Compare commits

...
Author SHA1 Message Date
swung0x48 5331106e90 [CI] Use uname for Arch runner hostname 2026-10-05 08:57:17 -04:00
swung0x48 ca981a8fc1 [CI] Add Arch WSL GPU runner smoke workflow and configuration 2026-10-05 08:56:27 -04:00
swung0x48 c90383db16 [Fix] (MG_Remote): bound socket buffers in log flood tests 2026-10-05 07:13:39 -04:00
swung0x48 698f04574b [Fix] (MG_Remote): wait for records emitted while draining return events 2026-10-05 06:21:42 -04:00
swung0x48 f497ff9741 [Fix] (DirectVulkan): detach ended render passes before attachment layout reconciliation 2026-10-05 05:12:42 -04:00
swung0x48 946e63d810 [Fix] (MG_IntegrationTest): check pre-finish work completion in present credit tests 2026-10-05 04:22:18 -04:00
swung0x48 918d40dd4f [Fix] (Init): allow MobileGL initialization with LIBGL_ALWAYS_SOFTWARE 2026-10-05 02:55:35 -04:00
swung0x48 f5b8fe48aa [Fix] (MGPipe, MG_Remote): verify server pack state and reject records after guilty GPU resets 2026-10-05 02:51:19 -04:00
swung0x48 c5d25868cf [Fix] (ShaderTranspiler): preserve block identity when deduplicating image types 2026-10-05 01:27:50 -04:00
swung0x48 c464947bdd [Fix] (TraceReplay): share Android window surfaces and release current bindings before resize 2026-10-05 00:47:56 -04:00
swung0x48 4fb75f4e00 [Fix] (DirectVulkan): prepare image copy recording after uploads and share texture usage checks 2026-10-05 00:19:21 -04:00
swung0x48 4d6904184d [Fix] (ShaderTranspiler): deduplicate image types after Vulkan image rewrites 2026-10-05 00:08:32 -04:00
swung0x48 2f1a673513 [Fix] (BackendLoader, DirectGLES): check EGL extensions before resolving damage swap entry points 2026-10-05 00:08:31 -04:00
swung0x48 88bc35b711 [Refactor] (BackendLoader): share extension name matching between GLES and EGL probes 2026-10-04 20:54:28 -04:00
swung0x48 880be57bb6 [CI] (Watchdog): disable GPU hang detection in all CI runs 2026-10-04 10:39:39 -04:00
swung0x48 70e4691b3a [Docs] (Anland): refresh the handoff to the 2026-10-04 state, with the perf, CI, YUV and VA-API streams 2026-10-04 07:25:23 -04:00
Swung0x48 2d42414667 [CI] (Test): annotate failed ctest entries with their failure lines, since the run logs need a signed-in reader 2026-10-04 07:20:04 -04:00
Swung0x48 32d27b1491 [Fix] (DirectGLES): a session's backend registration no longer trips the strict apply-thread pin while another session is applying 2026-10-04 07:20:04 -04:00
Swung0x48 78c5f8dae7 [CI] (Census): record Refuse{Recover} as a local refusal; the client declines to redial and there is no peer to tell 2026-10-04 07:20:04 -04:00
swung0x48 96bd540ba1 [Docs] (Performance): record the device state at the glmark2 handoff 2026-10-04 06:46:51 -04:00
swung0x48 710beefd02 [Docs] (Performance): glmark2 optimization handoff with the landed fixes, current numbers, remaining work and the bench scripts 2026-10-04 06:44:49 -04:00
Swung0x48 c1e957cb98 [CI] (Gates): retire the P3a/P4a untouched-region gates; those phases closed and later phases legitimately changed the regions 2026-10-04 06:30:24 -04:00
Swung0x48 eb71b669b3 [Fix] (Build): make the pull and verify builds compile again 2026-10-04 06:30:24 -04:00
swung0x48 bb31f374a3 [Perf] (DirectVulkan): keep the wire draw's render pass open across draws to the same attachments, closing it only for work that cannot sit inside one 2026-10-04 06:18:35 -04:00
swung0x48 e5ee31b022 [Build] (Anland): build the anland server for android-30, the APK's minSdk, so thread_local uses native TLS 2026-10-04 06:02:45 -04:00
swung0x48 915882b293 [Perf] (Server): time records only when the per-frame statistics are on 2026-10-04 06:02:44 -04:00
swung0x48 963f4d8177 [Perf] (DirectVulkan): reuse completed command buffers instead of freeing and allocating one per submission 2026-10-04 06:02:43 -04:00
swung0x48 d627fca387 [Perf] (Transport): the server rings a parked client only once the sequence it waits for is applied 2026-10-04 05:40:14 -04:00
swung0x48 2de0d6dd25 [Perf] (Pipe): reuse client-array buffers instead of a blocking create per draw, and copy contiguous ranges in one go 2026-10-04 04:54:54 -04:00
swung0x48 66b47a101b [Fix] (Remote): glFinish waits for the GPU, Magma bounds unsubmitted clear objects, ES contexts advertise OES_mapbuffer/rgb8_rgba8/depth24 2026-10-04 04:54:53 -04:00
swung0x48 4722fa38bb [Docs] (Performance): glmark2 baseline of Espryt and Magma against the stock kgsl stack on anland, with profiles and ranked bottlenecks 2026-10-04 04:33:24 -04:00
swung0x48 3a289222af [Docs] (MobileVA): plan a vendor-agnostic VA-API driver on Android MediaCodec, H.264 first 2026-10-03 23:46:13 -04:00
swung0x48 18a5dba0e4 [Docs] (Anland): Chrome decodes video in hardware through the YUV shared images, with probes to check it 2026-10-03 22:46:01 -04:00
swung0x48 7218bacb49 [Feature] (SharedImage): sample NV12/P010 dma-bufs through GL_OES_EGL_image_external on both backends, copying foreign buffers into server YUV images 2026-10-03 22:46:00 -04:00
145 changed files with 6690 additions and 630 deletions
@@ -126,9 +126,9 @@ timeout 25 mgrun offscreen glmark2-es2-wayland -s 1280x720 --run-forever > /tmp/
sleep 8; grep -a "presented through" /tmp/mgl-run.client.log
```
Chrome: `bash scripts/chrome-launch.sh <tag> [flags]` (log `/tmp/mgl-chrome-<tag>.client.log`, screenshot `ch-<tag>.png`; `CHROME_GPU=in-process` puts the GPU back in the browser, which cannot recover from a device loss).
Video: the container has a VA-API driver on the render node the GPU process gets (`/usr/lib/dri/msm_drm_drv_video.so`, V4L2 decode into its own NV12 dma-bufs); MobileGL cannot import those (only its own RGBA shared images; no dma-buf import in the phone's Vulkan, no `GL_OES_EGL_image_external`), and Chrome's GPU process aborts (SIGTRAP, "GPU process exited unexpectedly: exit_code=5") on the first decoded frame - window flicker + stalled video in a loop. `mobilegl-startup.sh chrome` therefore passes `--disable-accelerated-video-decode`. Diagnose a video with a CDP `Media.enable` dump (`kVideoDecoderName`); remote debugging needs a non-default `--user-data-dir`, and a fresh profile also needs `--password-store=basic` (else page loads wait forever on KWallet).
Video: Chrome decodes in hardware. The GPU process's render node has a VA-API driver in the container (`/usr/lib/dri/msm_drm_drv_video.so`, DroidSpaces' V4L2 driver): Chrome allocates its frames with `vaCreateSurfaces` and exports them (`vaExportSurfaceHandle`, one fd, R8 + GR88 layers) - the driver's own buffers, never GBM ones - and imports each as an NV12 (P010 for 10-bit) EGLImage bound to `GL_TEXTURE_EXTERNAL_OES`. MobileGL takes such a FOREIGN YUV dma-buf through its labelled CPU-copy fallback: the server maps it and copies it into a YUV AHardwareBuffer of its own (`AHardwareBuffer_lockPlanes`) at the reader's first use in each frame (any present, publishing flush or native fence), then converts that into the texture's RGBA level on the GPU (Espryt: `GL_EXT_YUV_target` with the import's colour hints, `MOBILEGL_YUV_DRIVER_CONVERSION=1` for the driver's external sampler; Magma: `VkSamplerYcbcrConversion` on the AHB's external format). A server-allocated NV12/P010 buffer (GBM `gbm_bo_create(NV12)`) needs no copy. `MOBILEGL_CHROME_VIDEO_DECODE=software` in `mobilegl-startup.sh chrome` brings back `--disable-accelerated-video-decode`. Check a video with `scripts/probes/cdpmedia.py <port> <s>` in the container (CDP `Media.enable`: `kVideoDecoderName = VaapiVideoDecoder`, `kIsPlatformVideoDecoder = true`; `CDP_NAVIGATE=<url>` / `CDP_EXPR=<js>`), which needs Chrome started with `--remote-debugging-port` and a non-default `--user-data-dir` (a fresh profile also needs `--password-store=basic`, else page loads wait forever on KWallet), and grep Chrome's log for `exit_code` (a GPU-process restart). `scripts/probes/yuvprobe.c` pixel-checks foreign NV12/P010 imports, their per-frame refresh and a GBM NV12 buffer (`mgrun offscreen mgl-yuvprobe`, exit 0 = pass). KWin's own NV12 path (it splits a YUV buffer into R8 + GR88 imports) is not served: MobileGL imports no single-channel dma-bufs, so KWin advertises no YUV format over linux-dmabuf. A vendor-agnostic replacement driver (MobileVA: VA-API on Android MediaCodec, decoding into server YUV images with no CPU copy, the current driver kept as an A/B fallback) is planned, not built: `docs/MobileVA/plan.md`.
Device loss: `setprop debug.mobilegl.inject_device_lost_pid <pid>` (as root) loses the server session of the client whose pid - as it sees itself in the container - is `<pid>`, once per value, on any backend and whether or not it presents; `debug.mobilegl.inject_device_lost_at N` instead loses whichever session makes the N-th device check (Espryt presents and read-backs only). `bash scripts/chrome-recover.sh <in-process|process> [rounds]` checks that Chrome repaints after such a loss without a restart (battery-guarded, stops below 40%). `bash scripts/kwin-recover.sh [rounds]` loses KWin's own session (default twice) and checks that KWin restarts compositing on a fresh session - same KWin, plasmashell and client processes, the screen still animating (glmark2 on Wayland, glxgears on X11, `CHROME=1` adds Chrome) - which needs the anland backend + kwin.patch reset handling deployed (`kwin-stage.sh build`).
Probes: `scripts/probes/texfmt.c` (texture formats keep alpha?), `dmahash.c` (hash a process's dma-bufs); build in the container, e.g. `gcc -O1 /root/texfmt.c -o /usr/local/bin/mgl-texfmt -lEGL`, run with `mgrun offscreen`.
Probes: `scripts/probes/texfmt.c` (texture formats keep alpha?), `dmahash.c` (hash a process's dma-bufs), `yuvprobe.c` (YUV dma-buf sampling, `gcc -O1 /root/yuvprobe.c -o /usr/local/bin/mgl-yuvprobe -lEGL -lGLESv2 -lgbm -lm`), `cdpmedia.py` (which decoder a Chrome page uses); build in the container, e.g. `gcc -O1 /root/texfmt.c -o /usr/local/bin/mgl-texfmt -lEGL`, run with `mgrun offscreen`.
GPU hangs: a shader that never ends stops the WHOLE GPU (every process, MobileGL or not) until the kernel resets it - ~2 s, or ~10 s ("Preemption Fault" in dmesg) when a higher-priority context waits; nothing in user space shortens one stall. What the server does is end the guilty session so it cannot hang the GPU again (Magma: event-bracketed submissions + `mgl-gpu-watch`, budget `MOBILEGL_GPU_HANG_BUDGET_MS` / `debug.mobilegl.gpu_hang_budget_ms`, default 2000; Espryt: the driver's GUILTY reset status, which it reports after the ~2 s reset only). Repro: `scripts/probes/hangclient.c` (Wayland + GL 3.3 core; `-H 60 -N 1 -E 60` hangs every 60 frames, `-H -1` is a frame-gap observer; build with wayland-scanner's xdg-shell files, `-lwayland-client -lwayland-egl -lEGL -lOpenGL -lm`), and `gpuprobe.c` (NDK, plain adb shell; `gpuprobe 30 high` is a high-priority GLES context outside MobileGL - it both measures the device-wide stall and forces the 10 s preemption path).
A result counts only if ALL hold:
@@ -72,7 +72,7 @@ bash scripts/build-android.sh # -> <worktree>/build-android/libMobileGL
```
The script runs, with the NDK toolchain:
```
-DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-26 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-30 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-DMOBILEGL_PIPE_PUSH=ON -DMOBILEGL_BUILD_DISAGGREGATED=ON -DMOBILEGL_BUILD_DISAGGREGATED_INPROC=ON
-DMOBILEGL_BUILD_TEST=OFF -DMOBILEGL_BUILD_BENCHMARK=OFF
```
@@ -4,7 +4,8 @@
# llvm-addr2line, it is what cyc.sh pushes). Stamp = worktree HEAD, so the container client
# (xbuild.sh, same stamp) and the server agree on the build ("compatible wire with different build"
# WARN otherwise). Env: MGL_WORKTREE, ANDROID_NDK, BUILD_DIR (default build-android), JOBS (12).
# Log level is compiled to INFO.
# Log level is compiled to INFO. android-30 = the APK's minSdk: thread_local is then the platform's
# own TLS rather than the emulated one (which cost 7-12% of an apply thread on the API 26 build).
set -euo pipefail
. "$(dirname "$0")/env.sh"
W=$MGL_WORKTREE
@@ -12,7 +13,7 @@ B=$W/${BUILD_DIR:-build-android}
STAMP=$(git -C "$W" rev-parse HEAD)
if [ ! -f "$B/build.ninja" ]; then
cmake -S "$W" -B "$B" -G Ninja -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK/build/cmake/android.toolchain.cmake" \
-DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-26 -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-30 -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-DMOBILEGL_PIPE_PUSH=ON -DMOBILEGL_BUILD_DISAGGREGATED=ON -DMOBILEGL_BUILD_DISAGGREGATED_INPROC=ON \
-DMOBILEGL_BUILD_TEST=OFF -DMOBILEGL_BUILD_BENCHMARK=OFF \
-DCMAKE_CXX_FLAGS=-DMOBILEGL_LOG_ACTIVE_LEVEL=MOBILEGL_LOG_LEVEL_INFO \
@@ -0,0 +1,94 @@
#!/usr/bin/env python3
# cdpmedia.py [port] [seconds] - ask a Chrome started with --remote-debugging-port=<port> (and a
# non-default --user-data-dir) which video decoder its first page's players use, and evaluate
# `window.__mglStats` / the element `#t` text of the page. Standard library only (no websocket
# module in the container): a minimal RFC 6455 client.
# python3 cdpmedia.py 9333 6
# Prints kVideoDecoderName / kIsPlatformVideoDecoder / kVideoTracks properties and media errors as
# they arrive, then the page's stats text.
import base64, json, os, socket, struct, sys, time, urllib.request
port = int(sys.argv[1]) if len(sys.argv) > 1 else 9333
seconds = float(sys.argv[2]) if len(sys.argv) > 2 else 6.0
targets = json.load(urllib.request.urlopen(f"http://127.0.0.1:{port}/json"))
page = next(t for t in targets if t.get("type") == "page")
url = page["webSocketDebuggerUrl"]
path = url.split(f"127.0.0.1:{port}", 1)[1]
sock = socket.create_connection(("127.0.0.1", port))
key = base64.b64encode(os.urandom(16)).decode()
sock.sendall((f"GET {path} HTTP/1.1\r\nHost: 127.0.0.1:{port}\r\nUpgrade: websocket\r\nConnection: Upgrade\r\n"
f"Sec-WebSocket-Key: {key}\r\nSec-WebSocket-Version: 13\r\n\r\n").encode())
response = b""
while b"\r\n\r\n" not in response:
response += sock.recv(4096)
buffer = response.split(b"\r\n\r\n", 1)[1]
def send(message):
data = json.dumps(message).encode()
header = bytes([0x81])
mask = os.urandom(4)
if len(data) < 126:
header += bytes([0x80 | len(data)])
elif len(data) < 65536:
header += bytes([0x80 | 126]) + struct.pack(">H", len(data))
else:
header += bytes([0x80 | 127]) + struct.pack(">Q", len(data))
sock.sendall(header + mask + bytes(b ^ mask[i % 4] for i, b in enumerate(data)))
def recv_exact(n):
global buffer
while len(buffer) < n:
chunk = sock.recv(65536)
if not chunk:
raise EOFError
buffer += chunk
out, buffer = buffer[:n], buffer[n:]
return out
def recv():
first, second = recv_exact(2)
length = second & 0x7F
if length == 126:
length = struct.unpack(">H", recv_exact(2))[0]
elif length == 127:
length = struct.unpack(">Q", recv_exact(8))[0]
payload = recv_exact(length)
return json.loads(payload) if first & 0x0F == 1 else None
send({"id": 1, "method": "Media.enable"})
# Players created before the domain was enabled report nothing: reload so the page makes new ones.
if os.environ.get("CDP_NAVIGATE"):
send({"id": 3, "method": "Page.navigate", "params": {"url": os.environ["CDP_NAVIGATE"]}})
elif os.environ.get("CDP_RELOAD", "1") == "1":
send({"id": 3, "method": "Page.reload"})
deadline = time.time() + seconds
sock.settimeout(0.5)
seen = set()
while time.time() < deadline:
try:
message = recv()
except socket.timeout:
continue
if not message or "method" not in message:
continue
if message["method"] == "Media.playerPropertiesChanged":
for prop in message["params"]["properties"]:
if prop["name"] in ("kVideoDecoderName", "kIsPlatformVideoDecoder", "kVideoTracks", "kResolution"):
line = f"{prop['name']} = {prop['value']}"
if line not in seen:
seen.add(line)
print(line)
elif message["method"] in ("Media.playerErrorsRaised",):
print("media error:", json.dumps(message["params"])[:300])
send({"id": 2, "method": "Runtime.evaluate",
"params": {"expression": os.environ.get("CDP_EXPR", "(document.getElementById('t')||{}).textContent"), "returnByValue": True}})
sock.settimeout(3)
while True:
message = recv()
if message and message.get("id") == 2:
print("page:", message.get("result", {}).get("result", {}).get("value"))
break
@@ -0,0 +1,410 @@
// yuvprobe.c - pixel check of YUV dma-buf sampling through MobileGL (GL_OES_EGL_image_external).
//
// gcc -O1 yuvprobe.c -o /usr/local/bin/mgl-yuvprobe -lEGL -lGLESv2 -lgbm
// mgrun offscreen mgl-yuvprobe # exit 0 = every check passed
//
// What it does, in an ES 3.0 context on a pbuffer:
// 1. FOREIGN NV12 (and P010): a dma-buf from /dev/dma_heap/system - a buffer the MobileGL server
// did not allocate, as a video decoder's would be - with padded pitches and a gap before the
// chroma plane, filled with four vertical bars (red, green, blue, white) in BT.709 narrow
// range. Imported with eglCreateImage(EGL_LINUX_DMA_BUF_EXT) and its colour hints, bound to
// GL_TEXTURE_EXTERNAL_OES, drawn through a samplerExternalOES shader into an RGBA8 FBO and read
// back: each bar must come out its colour. The buffer is then rewritten with the bars in
// reverse order and, after a native fence (what ends a frame for an offscreen producer), drawn
// again: the new order must show (the server refreshes a foreign buffer at each frame).
// 2. GBM NV12: gbm_bo_create(NV12) through the mobilegl GBM backend (a server-allocated YUV
// buffer, zero-copy), its planes printed, imported and sampled once (content undefined; the
// check is that every call succeeds).
// Also prints the dma-buf formats/modifiers EGL lists for NV12/P010.
#define _GNU_SOURCE
#include <EGL/egl.h>
#include <EGL/eglext.h>
#include <GLES3/gl3.h>
#include <GLES2/gl2ext.h>
#include <fcntl.h>
#include <gbm.h>
#include <linux/dma-buf.h>
#include <linux/dma-heap.h>
#include <math.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <unistd.h>
#ifndef DRM_FORMAT_NV12
#define DRM_FORMAT_NV12 0x3231564E
#endif
#ifndef DRM_FORMAT_P010
#define DRM_FORMAT_P010 0x30313050
#endif
static int g_failures = 0;
#define CHECK(cond, ...) \
do { \
if (!(cond)) { \
printf("FAIL: " __VA_ARGS__); \
printf("\n"); \
++g_failures; \
} \
} while (0)
static PFNEGLCREATEIMAGEKHRPROC pCreateImage;
static PFNEGLDESTROYIMAGEKHRPROC pDestroyImage;
static PFNGLEGLIMAGETARGETTEXTURE2DOESPROC pTargetTexture;
static PFNEGLQUERYDMABUFFORMATSEXTPROC pQueryFormats;
static PFNEGLQUERYDMABUFMODIFIERSEXTPROC pQueryModifiers;
static PFNEGLCREATESYNCKHRPROC pCreateSync;
static PFNEGLDESTROYSYNCKHRPROC pDestroySync;
static EGLDisplay g_display;
enum { W = 256, H = 128, BARS = 4 };
static const unsigned char kBars[BARS][3] = {{255, 0, 0}, {0, 255, 0}, {0, 0, 255}, {255, 255, 255}};
// RGB -> BT.709 narrow-range codes at `bits` (8 or 10).
static void Encode709(const unsigned char rgb[3], int bits, int out[3]) {
const double kr = 0.2126, kb = 0.0722, kg = 1.0 - kr - kb;
const double r = rgb[0] / 255.0, g = rgb[1] / 255.0, b = rgb[2] / 255.0;
const double y = kr * r + kg * g + kb * b;
const double cb = (b - y) / (2.0 * (1.0 - kb)), cr = (r - y) / (2.0 * (1.0 - kr));
const double scale = (double)(1 << (bits - 8));
out[0] = (int)lround((16.0 + 219.0 * y) * scale);
out[1] = (int)lround((128.0 + 224.0 * cb) * scale);
out[2] = (int)lround((128.0 + 224.0 * cr) * scale);
}
// Writes the bars (`reverse`d or not) into a semi-planar buffer: 1 byte samples, or P010's 16-bit
// samples with the 10 significant bits high.
static void FillBars(uint8_t* base, uint32_t pitchY, uint32_t offsetC, uint32_t pitchC, int bytes, int reverse) {
const int bits = bytes == 1 ? 8 : 10;
for (int y = 0; y < H; ++y) {
for (int x = 0; x < W; ++x) {
int bar = x * BARS / W;
if (reverse) bar = BARS - 1 - bar;
int code[3];
Encode709(kBars[bar], bits, code);
uint8_t* py = base + (size_t)y * pitchY + (size_t)x * bytes;
if (bytes == 1) py[0] = (uint8_t)code[0];
else *(uint16_t*)py = (uint16_t)(code[0] << 6);
if ((x & 1) == 0 && (y & 1) == 0) {
uint8_t* pc = base + offsetC + (size_t)(y / 2) * pitchC + (size_t)(x / 2) * 2 * bytes;
if (bytes == 1) {
pc[0] = (uint8_t)code[1];
pc[1] = (uint8_t)code[2];
} else {
((uint16_t*)pc)[0] = (uint16_t)(code[1] << 6);
((uint16_t*)pc)[1] = (uint16_t)(code[2] << 6);
}
}
}
}
}
static GLuint Compile(GLenum type, const char* source) {
GLuint shader = glCreateShader(type);
glShaderSource(shader, 1, &source, NULL);
glCompileShader(shader);
GLint ok = 0;
glGetShaderiv(shader, GL_COMPILE_STATUS, &ok);
if (!ok) {
char log[2048];
glGetShaderInfoLog(shader, sizeof(log), NULL, log);
printf("shader: %s\n", log);
}
return shader;
}
static GLuint g_program, g_fbo, g_rgba;
static void SetUpDraw(void) {
const char* vs = "#version 300 es\n"
"out vec2 uv;\n"
"void main() {\n"
" vec2 p = vec2(float((gl_VertexID << 1) & 2), float(gl_VertexID & 2));\n"
" gl_Position = vec4(p * 2.0 - 1.0, 0.0, 1.0);\n"
" uv = p;\n"
"}\n";
const char* fs = "#version 300 es\n"
"#extension GL_OES_EGL_image_external_essl3 : require\n"
"precision mediump float;\n"
"uniform samplerExternalOES tex;\n"
"in vec2 uv;\n"
"out vec4 color;\n"
"void main() { color = texture(tex, uv); }\n";
g_program = glCreateProgram();
glAttachShader(g_program, Compile(GL_VERTEX_SHADER, vs));
glAttachShader(g_program, Compile(GL_FRAGMENT_SHADER, fs));
glLinkProgram(g_program);
GLint linked = 0;
glGetProgramiv(g_program, GL_LINK_STATUS, &linked);
CHECK(linked, "the samplerExternalOES program did not link");
GLint type = 0, size = 0;
char name[64];
glGetActiveUniform(g_program, 0, sizeof(name), NULL, &size, (GLenum*)&type, name);
printf("uniform 0: %s type 0x%04x\n", name, type);
CHECK(type == GL_SAMPLER_EXTERNAL_OES, "the sampler reports type 0x%04x, not GL_SAMPLER_EXTERNAL_OES", type);
glGenTextures(1, &g_rgba);
glBindTexture(GL_TEXTURE_2D, g_rgba);
glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, W, H);
glGenFramebuffers(1, &g_fbo);
glBindFramebuffer(GL_FRAMEBUFFER, g_fbo);
glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, g_rgba, 0);
CHECK(glCheckFramebufferStatus(GL_FRAMEBUFFER) == GL_FRAMEBUFFER_COMPLETE, "the readback FBO is incomplete");
GLuint vao;
glGenVertexArrays(1, &vao);
glBindVertexArray(vao);
}
// Draws the external texture `texture` and checks each bar's center (row `y` of the image).
static void DrawAndCheck(GLuint texture, const char* what, int reverse, int tolerance) {
glBindFramebuffer(GL_FRAMEBUFFER, g_fbo);
glViewport(0, 0, W, H);
glUseProgram(g_program);
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_EXTERNAL_OES, texture);
glUniform1i(glGetUniformLocation(g_program, "tex"), 0);
glDrawArrays(GL_TRIANGLES, 0, 3);
static unsigned char pixels[W * H * 4];
glReadPixels(0, 0, W, H, GL_RGBA, GL_UNSIGNED_BYTE, pixels);
GLenum error = glGetError();
CHECK(error == GL_NO_ERROR, "%s: GL error 0x%04x", what, error);
if (tolerance < 0) {
printf("%s: center pixel %d %d %d %d (content undefined)\n", what, pixels[(H / 2 * W + W / 2) * 4],
pixels[(H / 2 * W + W / 2) * 4 + 1], pixels[(H / 2 * W + W / 2) * 4 + 2],
pixels[(H / 2 * W + W / 2) * 4 + 3]);
return;
}
// Rows: GL row 0 is the image's first row (the uv mapping above), and the bars are vertical,
// so any row does; take two to catch a broken chroma row pitch.
const int rows[2] = {H / 4, (3 * H) / 4};
for (int r = 0; r < 2; ++r) {
for (int bar = 0; bar < BARS; ++bar) {
const int x = bar * (W / BARS) + W / (2 * BARS);
const unsigned char* p = &pixels[(rows[r] * W + x) * 4];
const unsigned char* want = kBars[reverse ? BARS - 1 - bar : bar];
int worst = 0;
for (int c = 0; c < 3; ++c) {
int d = abs((int)p[c] - (int)want[c]);
if (d > worst) worst = d;
}
printf("%s: row %d bar %d -> %3d %3d %3d %3d (want %3d %3d %3d)%s\n", what, rows[r], bar, p[0], p[1], p[2],
p[3], want[0], want[1], want[2], worst > tolerance ? " <-- off" : "");
CHECK(worst <= tolerance, "%s: bar %d is off by %d", what, bar, worst);
}
}
}
static int AllocateDmaHeap(size_t size) {
int heap = open("/dev/dma_heap/system", O_RDWR | O_CLOEXEC);
if (heap < 0) {
perror("open /dev/dma_heap/system");
return -1;
}
struct dma_heap_allocation_data data = {.len = size, .fd_flags = O_RDWR | O_CLOEXEC};
if (ioctl(heap, DMA_HEAP_IOCTL_ALLOC, &data) != 0) {
perror("DMA_HEAP_IOCTL_ALLOC");
close(heap);
return -1;
}
close(heap);
return (int)data.fd;
}
static void Sync(int fd, int start) {
struct dma_buf_sync sync = {.flags = DMA_BUF_SYNC_WRITE | (start ? DMA_BUF_SYNC_START : DMA_BUF_SYNC_END)};
ioctl(fd, DMA_BUF_IOCTL_SYNC, &sync);
}
static EGLImageKHR Import(int fd, uint32_t fourcc, uint32_t pitchY, uint32_t offsetC, uint32_t pitchC, int modifier) {
EGLint attribs[64];
int n = 0;
#define A(k, v) (attribs[n++] = (k), attribs[n++] = (EGLint)(v))
A(EGL_WIDTH, W);
A(EGL_HEIGHT, H);
A(EGL_LINUX_DRM_FOURCC_EXT, fourcc);
A(EGL_DMA_BUF_PLANE0_FD_EXT, fd);
A(EGL_DMA_BUF_PLANE0_OFFSET_EXT, 0);
A(EGL_DMA_BUF_PLANE0_PITCH_EXT, pitchY);
A(EGL_DMA_BUF_PLANE1_FD_EXT, fd);
A(EGL_DMA_BUF_PLANE1_OFFSET_EXT, offsetC);
A(EGL_DMA_BUF_PLANE1_PITCH_EXT, pitchC);
if (modifier) {
A(EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT, 0);
A(EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT, 0);
A(EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT, 0);
A(EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT, 0);
}
A(EGL_YUV_COLOR_SPACE_HINT_EXT, EGL_ITU_REC709_EXT);
A(EGL_SAMPLE_RANGE_HINT_EXT, EGL_YUV_NARROW_RANGE_EXT);
attribs[n] = EGL_NONE;
#undef A
EGLImageKHR image = pCreateImage(g_display, EGL_NO_CONTEXT, EGL_LINUX_DMA_BUF_EXT, NULL, attribs);
if (image == EGL_NO_IMAGE_KHR) printf("eglCreateImage failed: 0x%04x\n", eglGetError());
return image;
}
static GLuint ExternalTexture(EGLImageKHR image) {
GLuint texture;
glGenTextures(1, &texture);
glBindTexture(GL_TEXTURE_EXTERNAL_OES, texture);
pTargetTexture(GL_TEXTURE_EXTERNAL_OES, image);
GLenum error = glGetError();
CHECK(error == GL_NO_ERROR, "glEGLImageTargetTexture2DOES(GL_TEXTURE_EXTERNAL_OES) -> 0x%04x", error);
GLint bound = 0;
glGetIntegerv(GL_TEXTURE_BINDING_EXTERNAL_OES, &bound);
CHECK((GLuint)bound == texture, "GL_TEXTURE_BINDING_EXTERNAL_OES reads %d", bound);
return texture;
}
// A frame boundary as an offscreen producer makes one: a native fence.
static void EndFrame(void) {
EGLint attribs[] = {EGL_NONE};
EGLSyncKHR sync = pCreateSync(g_display, EGL_SYNC_NATIVE_FENCE_ANDROID, attribs);
CHECK(sync != EGL_NO_SYNC_KHR, "eglCreateSync(native fence) failed");
if (sync != EGL_NO_SYNC_KHR) pDestroySync(g_display, sync);
}
static void ForeignTest(uint32_t fourcc, int bytes, int tolerance) {
const char* what = fourcc == DRM_FORMAT_NV12 ? "foreign NV12" : "foreign P010";
// Padded pitches and a gap before the chroma plane, as decoders lay buffers out.
const uint32_t pitchY = W * bytes + 64, pitchC = W * bytes + 128;
const uint32_t offsetC = pitchY * (H + 16);
const size_t size = offsetC + (size_t)pitchC * (H / 2) + 4096;
int fd = AllocateDmaHeap(size);
if (fd < 0) {
CHECK(0, "%s: no dma-buf to test with", what);
return;
}
uint8_t* map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
CHECK(map != MAP_FAILED, "%s: mmap failed", what);
if (map == MAP_FAILED) return;
Sync(fd, 1);
FillBars(map, pitchY, offsetC, pitchC, bytes, 0);
Sync(fd, 0);
EGLImageKHR image = Import(fd, fourcc, pitchY, offsetC, pitchC, fourcc == DRM_FORMAT_NV12);
CHECK(image != EGL_NO_IMAGE_KHR, "%s: import refused", what);
if (image == EGL_NO_IMAGE_KHR) return;
GLuint texture = ExternalTexture(image);
DrawAndCheck(texture, what, 0, tolerance);
EndFrame();
// New content, same buffer: shows after the frame boundary.
Sync(fd, 1);
FillBars(map, pitchY, offsetC, pitchC, bytes, 1);
Sync(fd, 0);
EndFrame();
DrawAndCheck(texture, what, 1, tolerance);
glDeleteTextures(1, &texture);
pDestroyImage(g_display, image);
munmap(map, size);
close(fd);
}
static void GbmTest(void) {
int node = open("/dev/dri/renderD128", O_RDWR | O_CLOEXEC);
if (node < 0) node = open("/dev/null", O_RDWR | O_CLOEXEC);
struct gbm_device* gbm = gbm_create_device(node);
if (gbm == NULL) {
CHECK(0, "gbm_create_device failed");
return;
}
printf("gbm backend: %s\n", gbm_device_get_backend_name(gbm));
const int supported = gbm_device_is_format_supported(gbm, DRM_FORMAT_NV12, 0);
printf("gbm NV12 supported: %d, P010: %d\n", supported,
gbm_device_is_format_supported(gbm, DRM_FORMAT_P010, 0));
CHECK(supported == 1, "gbm does not support NV12");
struct gbm_bo* bo = gbm_bo_create(gbm, W, H, DRM_FORMAT_NV12, 0);
CHECK(bo != NULL, "gbm_bo_create(NV12) failed");
if (bo != NULL) {
const int planes = gbm_bo_get_plane_count(bo);
printf("gbm NV12 bo: planes %d, stride %u/%u, offset %u/%u, modifier 0x%llx\n", planes,
gbm_bo_get_stride_for_plane(bo, 0), gbm_bo_get_stride_for_plane(bo, 1), gbm_bo_get_offset(bo, 0),
gbm_bo_get_offset(bo, 1), (unsigned long long)gbm_bo_get_modifier(bo));
CHECK(planes == 2, "the NV12 bo has %d planes", planes);
int fd = gbm_bo_get_fd(bo);
EGLImageKHR image =
Import(fd, DRM_FORMAT_NV12, gbm_bo_get_stride_for_plane(bo, 0), gbm_bo_get_offset(bo, 1),
gbm_bo_get_stride_for_plane(bo, 1), 0);
CHECK(image != EGL_NO_IMAGE_KHR, "the server's own NV12 buffer was not imported");
if (image != EGL_NO_IMAGE_KHR) {
GLuint texture = ExternalTexture(image);
DrawAndCheck(texture, "gbm NV12", 0, -1);
glDeleteTextures(1, &texture);
pDestroyImage(g_display, image);
}
close(fd);
gbm_bo_destroy(bo);
}
gbm_device_destroy(gbm);
close(node);
}
int main(void) {
setvbuf(stdout, NULL, _IONBF, 0);
g_display = eglGetDisplay(EGL_DEFAULT_DISPLAY);
EGLint major, minor;
if (!eglInitialize(g_display, &major, &minor)) {
printf("FAIL: eglInitialize\n");
return 2;
}
const char* eglExtensions = eglQueryString(g_display, EGL_EXTENSIONS);
printf("EGL %d.%d dma_buf_import: %s\n", major, minor,
strstr(eglExtensions, "EGL_EXT_image_dma_buf_import") ? "yes" : "no");
eglBindAPI(EGL_OPENGL_ES_API);
const EGLint configAttribs[] = {EGL_SURFACE_TYPE, EGL_PBUFFER_BIT, EGL_RENDERABLE_TYPE, EGL_OPENGL_ES3_BIT,
EGL_RED_SIZE, 8, EGL_GREEN_SIZE, 8, EGL_BLUE_SIZE, 8, EGL_NONE};
EGLConfig config;
EGLint count = 0;
eglChooseConfig(g_display, configAttribs, &config, 1, &count);
if (count < 1) {
printf("FAIL: no config\n");
return 2;
}
const EGLint pbufferAttribs[] = {EGL_WIDTH, 16, EGL_HEIGHT, 16, EGL_NONE};
EGLSurface surface = eglCreatePbufferSurface(g_display, config, pbufferAttribs);
const EGLint contextAttribs[] = {EGL_CONTEXT_MAJOR_VERSION, 3, EGL_NONE};
EGLContext context = eglCreateContext(g_display, config, EGL_NO_CONTEXT, contextAttribs);
if (context == EGL_NO_CONTEXT || !eglMakeCurrent(g_display, surface, surface, context)) {
printf("FAIL: no ES 3 context\n");
return 2;
}
pCreateImage = (PFNEGLCREATEIMAGEKHRPROC)eglGetProcAddress("eglCreateImageKHR");
pDestroyImage = (PFNEGLDESTROYIMAGEKHRPROC)eglGetProcAddress("eglDestroyImageKHR");
pTargetTexture = (PFNGLEGLIMAGETARGETTEXTURE2DOESPROC)eglGetProcAddress("glEGLImageTargetTexture2DOES");
pQueryFormats = (PFNEGLQUERYDMABUFFORMATSEXTPROC)eglGetProcAddress("eglQueryDmaBufFormatsEXT");
pQueryModifiers = (PFNEGLQUERYDMABUFMODIFIERSEXTPROC)eglGetProcAddress("eglQueryDmaBufModifiersEXT");
pCreateSync = (PFNEGLCREATESYNCKHRPROC)eglGetProcAddress("eglCreateSyncKHR");
pDestroySync = (PFNEGLDESTROYSYNCKHRPROC)eglGetProcAddress("eglDestroySyncKHR");
printf("GL_RENDERER: %s\n", glGetString(GL_RENDERER));
const char* glExtensions = (const char*)glGetString(GL_EXTENSIONS);
const int external = strstr(glExtensions, "GL_OES_EGL_image_external_essl3") != NULL;
printf("GL_OES_EGL_image_external_essl3: %s\n", external ? "yes" : "no");
CHECK(external, "GL_OES_EGL_image_external_essl3 is not advertised");
EGLint formats[32], formatCount = 0;
pQueryFormats(g_display, 32, formats, &formatCount);
int nv12 = 0;
for (int i = 0; i < formatCount; ++i) {
printf("dma-buf format %.4s", (const char*)&formats[i]);
EGLuint64KHR modifiers[4];
EGLBoolean externalOnly[4];
EGLint modifierCount = 0;
pQueryModifiers(g_display, formats[i], 4, modifiers, externalOnly, &modifierCount);
for (int m = 0; m < modifierCount; ++m)
printf(" [modifier 0x%llx%s]", (unsigned long long)modifiers[m], externalOnly[m] ? " external-only" : "");
printf("\n");
nv12 |= formats[i] == (EGLint)DRM_FORMAT_NV12;
}
CHECK(nv12, "NV12 is not a listed dma-buf format");
SetUpDraw();
ForeignTest(DRM_FORMAT_NV12, 1, 12);
ForeignTest(DRM_FORMAT_P010, 2, 16);
GbmTest();
printf("%s (%d failure(s))\n", g_failures ? "FAILED" : "PASSED", g_failures);
eglMakeCurrent(g_display, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT);
eglTerminate(g_display);
return g_failures ? 1 : 0;
}
+7 -3
View File
@@ -18,6 +18,10 @@ on:
- '**/*.md'
workflow_dispatch:
env:
# Software-rendered CI submissions may exceed the interactive GPU hang budget.
MOBILEGL_GPU_HANG_BUDGET_MS: "0"
# ONE LIVE RUN PER BRANCH, as in test.yml: a superseded push is cancelled rather than left queued
# behind its successor; a manual dispatch gets a group of its own; dev queues rather than cancels.
concurrency:
@@ -477,7 +481,7 @@ jobs:
extra_retrace_args=()
if [ "${MOBILEGL_CI_TRANSPORT}" = "inproc" ]; then
extra_retrace_args+=(
--env "MOBILEGL_TRANSPORT=inproc;MOBILEGL_IPC_ROLE_SPLIT_STATE=1;MOBILEGL_IPC_STRICT_ERRORS=1;MOBILEGL_IPC_RUN_AHEAD=1"
--env "MOBILEGL_TRANSPORT=inproc;MOBILEGL_IPC_ROLE_SPLIT_STATE=1;MOBILEGL_IPC_STRICT_ERRORS=1;MOBILEGL_IPC_RUN_AHEAD=1;MOBILEGL_GPU_HANG_BUDGET_MS=${MOBILEGL_GPU_HANG_BUDGET_MS}"
--require-inproc
)
elif [ "${MOBILEGL_CI_TRANSPORT}" = "spawn" ]; then
@@ -493,12 +497,12 @@ jobs:
# into the CLIENT's process and SetWindowHandle is refused on the way across with
# Fatal{UnmigratedSurface, "AndroidNativeWindow@P12"} (Rule H).
extra_retrace_args+=(
--env "MOBILEGL_TRANSPORT=spawn;MOBILEGL_IPC_ROLE_SPLIT_STATE=1;MOBILEGL_IPC_STRICT_ERRORS=1;MOBILEGL_IPC_RUN_AHEAD=1"
--env "MOBILEGL_TRANSPORT=spawn;MOBILEGL_IPC_ROLE_SPLIT_STATE=1;MOBILEGL_IPC_STRICT_ERRORS=1;MOBILEGL_IPC_RUN_AHEAD=1;MOBILEGL_GPU_HANG_BUDGET_MS=${MOBILEGL_GPU_HANG_BUDGET_MS}"
--use-pbuffer
--require-spawn
)
else
extra_retrace_args+=(--env "MOBILEGL_TRANSPORT=monolith")
extra_retrace_args+=(--env "MOBILEGL_TRANSPORT=monolith;MOBILEGL_GPU_HANG_BUDGET_MS=${MOBILEGL_GPU_HANG_BUDGET_MS}")
fi
# Bliss needs the newer signed ANGLE variant plus sampler mipmap
# min-filter downgrading on ANGLE llvmpipe.
+35 -90
View File
@@ -21,6 +21,10 @@ on:
required: false
default: "087685d1"
env:
# Software-rendered CI submissions may exceed the interactive GPU hang budget.
MOBILEGL_GPU_HANG_BUDGET_MS: "0"
# ONE LIVE RUN PER BRANCH. The disaggregation branch pushes (often docs-only) commits minutes apart
# and each push fans out to ~330 retrace legs here and ~275 jobs in apk.yml; with no group, runs
# sat queued for 10+ hours behind their own successors. A manual dispatch gets a group of its own,
@@ -413,6 +417,12 @@ jobs:
--discovery "${RUNNER_TEMP}/primary-inproc-lane.json" \
--junit "${RUNNER_TEMP}/primary-inproc.xml" --output "${RUNNER_TEMP}/primary-inproc-proof.json"
# The run log needs a signed-in reader; annotations do not. Name every failed ctest entry,
# with the lines of its output that say why, where anyone reading the run can see them.
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py "${RUNNER_TEMP}"
- name: Upload integration lane evidence
if: always()
uses: actions/upload-artifact@v7
@@ -1169,6 +1179,12 @@ jobs:
# arm proof read; verify-split-logs/ is whatever the control steps left behind, which for
# twelve entries is a deliberately red re-run under the same name. Uploading only the
# second - which is what this step did before the copy above existed - loses the first.
# The run log needs a signed-in reader; annotations do not. Name every failed ctest entry,
# with the lines of its output that say why, where anyone reading the run can see them.
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py "${RUNNER_TEMP}"
- name: Upload verify-split lane logs
if: always()
uses: actions/upload-artifact@v7
@@ -1481,7 +1497,7 @@ jobs:
working-directory: build-split
env:
MOBILEGL_ITEST_REQUIRE_GPU: "1"
run: ctest --output-on-failure -L unit --no-tests=error -j "$(nproc)"
run: ctest --output-on-failure -L unit --no-tests=error -j "$(nproc)" --output-junit "${RUNNER_TEMP}/unit-split.xml"
# P7 wave 2-F, PH-7 (1). THE TCP SUPERVISOR LANE NEEDS flatc, AND CI HAD NONE.
#
@@ -1648,6 +1664,12 @@ jobs:
CONTROL_TMPDIR: ${{ runner.temp }}
run: bash "${GITHUB_WORKSPACE}/scripts/ci/split_negative_controls.sh"
# The run log needs a signed-in reader; annotations do not. Name every failed ctest entry,
# with the lines of its output that say why, where anyone reading the run can see them.
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py "${RUNNER_TEMP}"
- name: Upload split lane logs
if: always()
uses: actions/upload-artifact@v7
@@ -1737,7 +1759,7 @@ jobs:
env:
MOBILEGL_ITEST_REQUIRE_GPU: "1"
MOBILEGL_IPC_STRICT_ERRORS: "1"
run: ctest --output-on-failure -L unit --no-tests=error -j "$(nproc)"
run: ctest --output-on-failure -L unit --no-tests=error -j "$(nproc)" --output-junit "${RUNNER_TEMP}/unit-strict.xml"
# ---- P5e (ra), MG_Remote/CONTRACT-P5E.md §7: ONE STEP, TWO ARMS, AND THE TREE PICKS ----
#
@@ -2246,6 +2268,12 @@ jobs:
slp.require_green(lane, sys.argv[3], names)
PY
# The run log needs a signed-in reader; annotations do not. Name every failed ctest entry,
# with the lines of its output that say why, where anyone reading the run can see them.
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py "${RUNNER_TEMP}"
- name: Upload strict lane logs
if: always()
uses: actions/upload-artifact@v7
@@ -2751,7 +2779,11 @@ jobs:
&& [ '${{ matrix.case }}' = 'improved-transparency-minecraft-26.3' ]; then
export MOBILEGL_MAGMA_DISABLE_BLENDED_DEPTH_WRITE=1
fi
ctest -V --no-tests=error -R '^MobileGLTraceReplay\.${{ matrix.case }}\.${{ matrix.backend }}$'
ctest -V --no-tests=error -R '^MobileGLTraceReplay\.${{ matrix.case }}\.${{ matrix.backend }}$' --output-junit "${RUNNER_TEMP}/retrace.xml"
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py --excerpts 1 "${RUNNER_TEMP}"
- name: Upload core dumps
if: failure()
@@ -3506,27 +3538,10 @@ jobs:
runs-on: ubuntu-latest
# Deliberately independent of build-linux: these are source-level gates, they take
# seconds, and a broken build must not hide a drifted interface.
env:
# THE CURRENT PHASE's base ref, for the two G5 region gates below. It is 37da3c3a - P4a's
# base ref, INTEGRATOR-DECISIONS ID-1 - and NOT the workflow's baseline_sha input: that input
# is the SYMBOL baseline (087685d1, P1's G1 reading) and it is empty on a push, whereas these
# gates ask "did the do-not-touch list move since the phase started".
#
# IT MOVED FROM P3a's 44c2b5cf TO P4a's 37da3c3a WITH THE PHASE, and that is a deliberate
# narrowing rather than a loss: P3a's eleven functions were compared against 44c2b5cf at P3a's
# own exit and were byte-identical there, so 37da3c3a carries the same bodies (measured: the
# eleven shas at 37da3c3a are the eleven shas at 44c2b5cf, and FlushPendingRangesFrom's is
# still the sha pinned in the script at 3e298c9a). What the two gates now both answer is "did
# anything on the list move during P4a", which is the question this phase can act on.
BASELINE: "37da3c3a"
steps:
- name: Checkout repo
uses: actions/checkout@v6
with:
# The G5 gate reads Managers.cpp at BASELINE with `git show`, which a depth-1 checkout
# does not have. Nothing else in this job needs history.
fetch-depth: 0
# The seven generators all read MG_Pipe/*.def, so regenerating and diffing is what
# keeps the two interface tables, the wire records, the verify comparators, the
@@ -3655,76 +3670,6 @@ jobs:
python3 tools/trace_replay/test_run_tcp_matrix.py
python3 tools/trace_replay/test_compare_actuals.py
# A GATE as of P3a (G5). "pool 与延迟释放原样搬" (ROADMAP.md:19) is meant literally: the
# buffer pool, the deferred-release drain and the three persistently mapped rings move
# VERBATIM, and ARCHITECTURE.md:515 says why - their retire happens only inside Present, so
# a batching or ordering change there starves them, and nothing else in this workflow can
# see it. P3a rewrites the rest of Managers.cpp by design, so a file diff says nothing; the
# script extracts the ELEVEN named bodies and compares their hashes on their own.
#
# Eleven and not ten (ID-15): Managers.cpp carries the three-tier flush drain TWICE, once
# per preprocessor arm, and a push build compiles only FlushPendingRangesFrom while the
# untouched FlushPendingRangesNow lives in the `#else`. Hashing the pull name alone would
# protect text no shipping build compiles, so both are hashed - the pull ladder against
# BASELINE, the push ladder against a sha pinned in the script at 3dadd4c1 (ID-41), because
# that one was born in P3a and is compared against the reviewed body rather than against the
# base ref.
#
# Scoped to the disaggregation branch and to a manual dispatch, deliberately: the question
# is "did these eleven move since P3a started", and BASELINE is P3a's base ref. On dev,
# where unrelated buffer fixes land on their own schedule, the same comparison would be
# asking a question nobody posed - it belongs with the TEMPORARY trigger lines at the top
# of this file and retires with them.
#
# --self-test is the half that keeps it honest, and it is not optional: a comparison that
# silently stopped comparing produces exactly the same green as eleven untouched bodies. It
# runs eight canned controls - eleven bodies extracted, an untouched copy compared equal, an
# edit OUTSIDE them ignored, each of the three perturbation targets (ClearBufferPool and BOTH
# flush ladders) reported BY NAME, plus the pin-precedence control and a one-token edit to the
# pinned ladder compared against the pin - and fails if any of them does not answer.
# Same shape as gen_pipe.py --self-test above.
- name: P3a regions unchanged (G5)
if: ${{ github.ref == 'refs/heads/feat/disaggregated' || github.event_name == 'workflow_dispatch' }}
run: bash scripts/p3a_untouched_regions.sh "${BASELINE}" HEAD
- name: P3a untouched-region self-test (G5)
if: ${{ github.ref == 'refs/heads/feat/disaggregated' || github.event_name == 'workflow_dispatch' }}
run: bash scripts/p3a_untouched_regions.sh --self-test
# A GATE AS OF P4a (G5), and a SECOND script rather than an edit to the one above. P4a extends
# the same claim to the rest of ARCHITECTURE.md:318's do-not-touch list - the unpack PBO ring's
# staging repack and its two helpers, the attachment permutation, the D24S8 sampling-emulation
# core and the format-caveat handler - which is SEVENTEEN regions across THREE files
# (BRIEF-P4A.md D-N: P3a's eleven, which P4a must not touch either, plus P4a's six). The
# parent's SOURCE_PATH is a single file, so the extension needed a per-region source path and
# a region KIND (DepthStencilSamplingReadImpl is a namespace, not a function); everything else
# about the extraction is its parent's, verbatim.
#
# Both scripts run. The parent keeps answering its own question against its own eleven, so a
# regression in either half names itself, and neither gate can be silenced by editing the
# other's list.
#
# Same feat/disaggregated-or-dispatch guard as the P3a step, for the same reason: the question
# is "did these move since the phase started", and on dev - where unrelated buffer and texture
# fixes land on their own schedule - it would be a question nobody posed. It belongs with the
# TEMPORARY trigger lines at the top of this file and retires with them.
#
# --self-test is the half that keeps it honest and is not optional: a comparison that silently
# stopped comparing produces exactly the same green as seventeen untouched regions. It runs
# three positive controls (seventeen regions extracted, an untouched copy compared equal, an
# edit OUTSIDE them invisible in all three files) and NINE negative ones - ClearBufferPool,
# FlushPendingRangesNow, RecomputeBackendColorSlots and StageBlocksIntoUnpackRing, each
# perturbed at its HEAD and at its TAIL and each required to be named BY NAME, plus a
# one-token edit to the pinned FlushPendingRangesFrom body compared against the pin - and
# fails if any of them does not answer.
- name: P4a regions unchanged (G5)
if: ${{ github.ref == 'refs/heads/feat/disaggregated' || github.event_name == 'workflow_dispatch' }}
run: bash scripts/p4a_untouched_regions.sh "${BASELINE}" HEAD
- name: P4a untouched-region self-test (G5)
if: ${{ github.ref == 'refs/heads/feat/disaggregated' || github.event_name == 'workflow_dispatch' }}
run: bash scripts/p4a_untouched_regions.sh --self-test
# Warning only for now: the disaggregation documents are still being written, and a
# lint that fails a rewrite in progress teaches people to ignore it. It becomes
# --strict when the documents settle.
+31
View File
@@ -0,0 +1,31 @@
name: Arch WSL GPU smoke
on:
workflow_dispatch:
push:
branches:
- ci/arch-wsl-gpu-runner
paths:
- .github/workflows/wsl-gpu-smoke.yml
- docs/ci/arch-wsl-gpu-runner.md
permissions: {}
concurrency:
group: arch-wsl-gpu-smoke
cancel-in-progress: false
jobs:
gpu-smoke:
runs-on: [self-hosted, Linux, X64, arch, wsl, gpu, gles, vulkan]
timeout-minutes: 5
steps:
- name: Report runner identity
shell: bash
run: |
printf 'Runner: %s\n' "$RUNNER_NAME"
uname -n
uname -a
- name: Verify hardware GLES and Vulkan
shell: bash
run: /usr/local/libexec/mobilegl-runner/gpu-smoke
+1
View File
@@ -372,6 +372,7 @@ set(SOURCE_FILES
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/FixIterationRPSubgroupScratchPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/EmulateSubgroupsPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/NormalizeRectCoordinatesPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/DeduplicateImageTypesPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/Lower1DArrayImagesPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/Lower1DSampledImagesPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/BakeImageFormatsPass.cpp
-12
View File
@@ -238,18 +238,6 @@ namespace MobileGL {
#endif
return true;
}
#if defined(__linux__) && !defined(__ANDROID__)
// LIBGL_ALWAYS_SOFTWARE is how a process (or whoever launched it) asks the system's GL for a
// software renderer. MobileGL is a GPU forwarder, so it steps aside and lets the system's
// other implementation give the process what it asked for.
{
String software;
if (MG_ConfigLoader::LookupSetting("LIBGL_ALWAYS_SOFTWARE", software) && !software.empty() &&
software != "0" && software != "false") {
return false;
}
}
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// Not yet: the bring-up would dial the configured server and wait for it. Ask first.
return MG_Remote::Client::ConfiguredServerAvailable();
+2
View File
@@ -97,6 +97,8 @@ namespace MobileGL::MG_Backend {
static_cast<SizeT>(target) >= kFormatCapabilityTextureTargetCount) {
return kFormatCapabilityTargetCount;
}
// An external texture is a 2D one wherever a format can go.
if (target == TextureTarget::External) return static_cast<SizeT>(TextureTarget::Texture2D);
return static_cast<SizeT>(target);
}
+26 -1
View File
@@ -9,9 +9,9 @@
#pragma once
#include <Includes.h>
#include "MG_State/GLState/TextureState/TextureEnum.h"
#include <MG_Util/Damage/Damage.h>
#if MOBILEGL_BUILD_DISAGGREGATED
#include <MG_Pipe/MGPipeHandles.h>
#include <MG_Util/Damage/Damage.h>
#endif
namespace MobileGL {
@@ -610,6 +610,24 @@ namespace MobileGL {
Uint32 Stride = 0;
Uint32 Offset = 0;
Uint64 Modifier = 0;
// A YUV image's Cb Cr plane: in the same descriptor, at this offset and pitch.
Uint32 Plane1Stride = 0;
Uint32 Plane1Offset = 0;
};
// An import of a two-plane (YUV) dma-buf: where its planes are in the one descriptor, and
// EGL_EXT_image_dma_buf_import's colour hints (0 = not given).
struct SharedImageImportLayout {
Uint32 PlaneCount = 1;
Uint32 Offset[2] = {0, 0};
Uint32 Pitch[2] = {0, 0};
Uint32 ColorSpace = 0;
Uint32 Range = 0;
Uint32 SitingX = 0;
Uint32 SitingY = 0;
// The DRM format modifier the import named; HasModifier false = none (the implicit one).
Bool HasModifier = false;
Uint64 Modifier = 0;
};
// THE DAMAGE OF THE PRESENT BEING APPLIED on this thread: the server's apply thread sets it
@@ -697,6 +715,13 @@ namespace MobileGL {
(void)fd, (void)width, (void)height, (void)fourcc, (void)outId;
return false;
}
// The same for a YUV fourcc, whose planes and hints `layout` gives. A YUV buffer the
// server did not export is taken too (its labelled CPU-copy fallback).
virtual Bool ImportSharedImagePlanes(int fd, Uint32 width, Uint32 height, Uint32 fourcc,
const SharedImageImportLayout& layout, Uint64* outId) {
(void)fd, (void)width, (void)height, (void)fourcc, (void)layout, (void)outId;
return false;
}
// Drops this client's reference (an allocation's or an import's).
virtual Bool ReleaseSharedImage(Uint64 id) {
(void)id;
+233 -10
View File
@@ -427,6 +427,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
// targets with no mip chain map to Unknown so the lookup falls back to no bias.
TextureTarget SamplerUniformTextureTarget(GLenum uniformType) {
switch (uniformType) {
case GL_SAMPLER_EXTERNAL_OES:
return TextureTarget::External;
case GL_SAMPLER_1D:
case GL_INT_SAMPLER_1D:
case GL_UNSIGNED_INT_SAMPLER_1D:
@@ -17765,6 +17767,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
MGLOG_E_ONCE("DirectGLES::MakeCurrent failed: native eglMakeCurrent returned error 0x%04x", error);
return false;
}
g_EGLFuncs.eglSwapBuffersWithDamageEXT =
MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(g_EGLFuncs);
InvalidateEglVerifiedStamp();
t_boundNativeContext = tuple.Context;
t_boundNativeDraw = session.Draw;
@@ -18604,18 +18608,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
WaitSyncKhrFn WaitSync = nullptr;
ClientWaitSyncKhrFn ClientWaitSync = nullptr;
DupNativeFenceFdFn DupNativeFenceFd = nullptr;
// YUV images: sampled as external textures (GL_OES_EGL_image_external_essl3), and
// through GL_EXT_YUV_target's raw Y'CbCr sampler where the driver has it, so the
// conversion is the import's own (its colour hints) rather than the driver's guess.
Bool ExternalEssl3 = false;
Bool YuvTarget = false;
};
Bool HasExtensionWord(const char* list, const char* word) {
if (list == nullptr) return false;
const SizeT n = std::strlen(word);
for (const char* at = std::strstr(list, word); at != nullptr; at = std::strstr(at + 1, word)) {
const Bool startOk = at == list || at[-1] == ' ';
const Bool endOk = at[n] == '\0' || at[n] == ' ';
if (startOk && endOk) return true;
}
return false;
}
using MG_Util::BackendLoader::HasExtensionWord;
// None of these is in the loader's tables, and eglGetProcAddress hands back live-looking
// stubs for entry points a driver lacks, so the extension strings decide first. Resolved
@@ -18667,6 +18667,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
entry->Why = "GL_OES_EGL_image is not advertised";
return *entry;
}
entry->ExternalEssl3 = HasExtensionWord(gl, "GL_OES_EGL_image_external_essl3");
entry->YuvTarget = HasExtensionWord(gl, "GL_EXT_YUV_target");
entry->GetNativeClientBuffer = reinterpret_cast<GetNativeClientBufferFn>(
g_EGLFuncs.eglGetProcAddress("eglGetNativeClientBufferANDROID"));
entry->CreateImage =
@@ -18706,6 +18708,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
Uint64 WaitedGeneration = 0;
Uint64 WaitedContext = 0;
Uint64 NotedFrame = 0;
// A YUV image (ConvertYuvForSampling): an external texture over the EGLImage, the source
// of the conversion into the texture's own RGBA storage, and the reader's frame the
// conversion last ran in.
GLuint ExternalTexture = 0;
Uint64 ConvertedFrame = 0;
~EglImage() {
// eglDestroyImageKHR needs the display, not a context, so a session's twin teardown
@@ -18713,6 +18720,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
// display - which destroyed the image itself - skip it.
if (Handle == nullptr || Destroy == nullptr || InProcessExit()) return;
if (g_displayTerminations.load(std::memory_order_acquire) != DisplayTerminations) return;
// Texture names are the share group's: deleting one needs any context of it, and
// without one current the name goes with the display.
if (ExternalTexture != 0 && t_boundNativeContext != EGL_NO_CONTEXT)
g_GLESFuncs.glDeleteTextures(1, &ExternalTexture);
Destroy(Display, Handle);
}
#endif
@@ -18938,14 +18949,23 @@ namespace MobileGL::MG_Backend::DirectGLES {
// The session ends frames at publishing flushes (PublishPendingAccesses): it writes
// images, so its first use of one in a frame waits for that image's pending reads.
Bool ImplicitSync = false;
// When a YUV image is converted again (ConvertYuvForSampling): renewed at EVERY
// boundary - present, publishing flush, and each native fence whatever the session
// is - since an offscreen producer (a browser's GPU process) may have no other.
Uint64 YuvEpoch = 0;
};
std::atomic<Uint64> g_readFrameIds{0};
ReadFrame& CurrentReadFrame() {
thread_local ReadFrame frame;
if (frame.Id == 0) frame.Id = g_readFrameIds.fetch_add(1, std::memory_order_relaxed) + 1;
if (frame.YuvEpoch == 0) frame.YuvEpoch = g_readFrameIds.fetch_add(1, std::memory_order_relaxed) + 1;
return frame;
}
void RenewYuvEpoch(ReadFrame& frame) {
frame.YuvEpoch = g_readFrameIds.fetch_add(1, std::memory_order_relaxed) + 1;
}
} // namespace
void AcquireForSampling(const EglImageRef& image) {
@@ -18975,9 +18995,201 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
}
Bool IsYuv(const EglImageRef& image) {
return image != nullptr && image->Image != nullptr && SI::FourccIsYuv(image->Image->Fourcc);
}
namespace {
// THE CONVERSION PASS: one triangle over the texture, sampling the image through an
// external texture. Programs are the share group's and rebuilt with a new context
// generation; the framebuffer and vertex array are per context (never shared).
struct YuvConversion {
Uint ContextGeneration = ~0u;
GLuint ExternalProgram = 0; // the driver's conversion (samplerExternalOES)
GLuint TargetProgram = 0; // the import's own (GL_EXT_YUV_target)
GLint TargetMatrix = -1;
GLint TargetOffset = -1;
Bool Failed = false;
NativeContextName Framebuffer;
NativeContextName VertexArray;
};
YuvConversion& Conversion() {
static auto* conversion = new YuvConversion();
return *conversion;
}
constexpr const char* kYuvVertexSource =
"#version 300 es\n"
"out vec2 vUv;\n"
"void main() {\n"
" vec2 p = vec2(float((gl_VertexID << 1) & 2), float(gl_VertexID & 2));\n"
" gl_Position = vec4(p * 2.0 - 1.0, 0.0, 1.0);\n"
" vUv = p;\n"
"}\n";
constexpr const char* kYuvExternalFragmentSource =
"#version 300 es\n"
"#extension GL_OES_EGL_image_external_essl3 : require\n"
"precision highp float;\n"
"uniform samplerExternalOES uSource;\n"
"in vec2 vUv;\n"
"out vec4 oColor;\n"
"void main() { oColor = vec4(texture(uSource, vUv).rgb, 1.0); }\n";
constexpr const char* kYuvTargetFragmentSource =
"#version 300 es\n"
"#extension GL_EXT_YUV_target : require\n"
"precision highp float;\n"
"uniform __samplerExternal2DY2YEXT uSource;\n"
"uniform mat3 uMatrix;\n"
"uniform vec3 uOffset;\n"
"in vec2 vUv;\n"
"out vec4 oColor;\n"
"void main() {\n"
" vec3 ycbcr = texture(uSource, vUv).xyz;\n"
" oColor = vec4(clamp(uMatrix * (ycbcr - uOffset), 0.0, 1.0), 1.0);\n"
"}\n";
GLuint BuildYuvProgram(const char* fragmentSource) {
const GLuint vertexShader = g_GLESFuncs.glCreateShader(GL_VERTEX_SHADER);
const GLuint fragmentShader = g_GLESFuncs.glCreateShader(GL_FRAGMENT_SHADER);
GLint linked = GL_FALSE;
GLuint program = 0;
if (vertexShader != 0 && fragmentShader != 0) {
g_GLESFuncs.glShaderSource(vertexShader, 1, &kYuvVertexSource, nullptr);
g_GLESFuncs.glCompileShader(vertexShader);
g_GLESFuncs.glShaderSource(fragmentShader, 1, &fragmentSource, nullptr);
g_GLESFuncs.glCompileShader(fragmentShader);
program = g_GLESFuncs.glCreateProgram();
if (program != 0) {
g_GLESFuncs.glAttachShader(program, vertexShader);
g_GLESFuncs.glAttachShader(program, fragmentShader);
g_GLESFuncs.glLinkProgram(program);
g_GLESFuncs.glGetProgramiv(program, GL_LINK_STATUS, &linked);
}
}
if (vertexShader != 0) g_GLESFuncs.glDeleteShader(vertexShader);
if (fragmentShader != 0) g_GLESFuncs.glDeleteShader(fragmentShader);
if (linked != GL_TRUE) {
if (program != 0) g_GLESFuncs.glDeleteProgram(program);
return 0;
}
return program;
}
// MOBILEGL_YUV_DRIVER_CONVERSION=1: always the driver's conversion (samplerExternalOES),
// even where GL_EXT_YUV_target would let the import's hints decide.
Bool DriverConversionForced() {
static const Bool forced = [] {
const char* value = std::getenv("MOBILEGL_YUV_DRIVER_CONVERSION");
return value != nullptr && value[0] == '1';
}();
return forced;
}
} // namespace
void ConvertYuvForSampling(const EglImageRef& image, GLuint target, Uint32 width, Uint32 height) {
if (!IsYuv(image) || target == 0 || width == 0 || height == 0) return;
EglImage& state = *image;
ReadFrame& frame = CurrentReadFrame();
if (state.ConvertedFrame == frame.YuvEpoch) return;
// Whatever happens below, this frame does not try again.
state.ConvertedFrame = frame.YuvEpoch;
const EntryPoints& entry = ResolvedEntryPoints();
if (!entry.Usable || !entry.ExternalEssl3) {
MGLOG_E_ONCE("Shared images: a YUV image cannot be sampled here (%s)",
entry.Usable ? "no GL_OES_EGL_image_external_essl3" : entry.Why.c_str());
return;
}
// THE CPU-COPY FALLBACK, for a buffer this server did not allocate (SharedImageRegistry.h).
String why;
if (state.Image->IsForeign() && !SI::RefreshForeign(*state.Image, why)) {
MGLOG_E_ONCE("Shared images: foreign YUV image %llu could not be refreshed: %s",
static_cast<unsigned long long>(state.Image->Id), why.c_str());
}
YuvConversion& conversion = Conversion();
if (conversion.ContextGeneration != g_backendContextGeneration) {
conversion.ExternalProgram = 0;
conversion.TargetProgram = 0;
conversion.TargetMatrix = conversion.TargetOffset = -1;
conversion.Failed = false;
conversion.ContextGeneration = g_backendContextGeneration;
}
if (conversion.Failed) return;
const Bool ownConversion = entry.YuvTarget && !DriverConversionForced();
GLuint& program = ownConversion ? conversion.TargetProgram : conversion.ExternalProgram;
if (program == 0) {
program = BuildYuvProgram(ownConversion ? kYuvTargetFragmentSource : kYuvExternalFragmentSource);
if (program == 0) {
conversion.Failed = true;
MGLOG_E_ONCE("Shared images: the YUV conversion program could not be built");
return;
}
if (ownConversion) {
conversion.TargetMatrix = g_GLESFuncs.glGetUniformLocation(program, "uMatrix");
conversion.TargetOffset = g_GLESFuncs.glGetUniformLocation(program, "uOffset");
}
MGLOG_I("Shared images: YUV images convert through %s",
ownConversion ? "GL_EXT_YUV_target with the import's colour hints"
: "the driver's external-sampler conversion");
}
if (conversion.Framebuffer == 0) g_GLESFuncs.glGenFramebuffers(1, &conversion.Framebuffer);
if (conversion.VertexArray == 0) g_GLESFuncs.glGenVertexArrays(1, &conversion.VertexArray);
if (conversion.Framebuffer == 0 || conversion.VertexArray == 0) return;
// Every binding goes through Espryt's shadows or a guard that puts it back.
ScopedEmulationDrawState drawState; // also activates TextureImpl::TempTextureUnit
ScopedFramebufferBinding binding(/*saveRead=*/false, /*saveDraw=*/true);
DrainErrors();
if (state.ExternalTexture == 0) {
g_GLESFuncs.glGenTextures(1, &state.ExternalTexture);
g_GLESFuncs.glBindTexture(GL_TEXTURE_EXTERNAL_OES, state.ExternalTexture);
entry.TargetTexture2D(GL_TEXTURE_EXTERNAL_OES, state.Handle);
g_GLESFuncs.glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
g_GLESFuncs.glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
g_GLESFuncs.glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
g_GLESFuncs.glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
const GLenum error = g_GLESFuncs.glGetError();
if (error != GL_NO_ERROR) {
MGLOG_E_ONCE("Shared images: YUV image %llu cannot be bound as an external texture (0x%04x)",
static_cast<unsigned long long>(state.Image->Id), static_cast<unsigned>(error));
}
} else {
g_GLESFuncs.glBindTexture(GL_TEXTURE_EXTERNAL_OES, state.ExternalTexture);
}
FramebufferImpl::BindFramebufferId(GL_DRAW_FRAMEBUFFER, conversion.Framebuffer);
g_GLESFuncs.glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, target, 0);
const GLenum status = g_GLESFuncs.glCheckFramebufferStatus(GL_DRAW_FRAMEBUFFER);
if (status == GL_FRAMEBUFFER_COMPLETE) {
VertexArrayImpl::BindBackendVAOId(conversion.VertexArray);
g_GLESFuncs.glUseProgram(program);
g_GLESFuncs.glUniform1i(g_GLESFuncs.glGetUniformLocation(program, "uSource"),
static_cast<GLint>(TextureImpl::TempTextureUnit));
if (ownConversion) {
// A Y2Y sample is the stored code normalized as the format stores it: 8 bits, or
// P010's 10 in the high end of 16.
const double sampleMax = state.Image->Fourcc == SI::kFourccP010 ? 65535.0 / 64.0 : 255.0;
const auto c = SI::Yuv::ConversionFor(SI::GetYuvHints(*state.Image), sampleMax);
g_GLESFuncs.glUniformMatrix3fv(conversion.TargetMatrix, 1, GL_FALSE, c.Matrix);
g_GLESFuncs.glUniform3fv(conversion.TargetOffset, 1, c.Offset);
}
g_GLESFuncs.glViewport(0, 0, static_cast<GLsizei>(width), static_cast<GLsizei>(height));
g_GLESFuncs.glDrawArrays(GL_TRIANGLES, 0, 3);
const GLenum error = g_GLESFuncs.glGetError();
if (error != GL_NO_ERROR) {
MGLOG_E_ONCE("Shared images: the YUV conversion draw failed (0x%04x)", static_cast<unsigned>(error));
}
} else {
MGLOG_E_ONCE("Shared images: a YUV conversion target is not renderable (0x%04x)",
static_cast<unsigned>(status));
}
g_GLESFuncs.glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
g_GLESFuncs.glBindTexture(GL_TEXTURE_EXTERNAL_OES, 0);
}
Bool PublishPendingAccesses() {
ReadFrame& frame = CurrentReadFrame();
frame.ImplicitSync = true;
RenewYuvEpoch(frame);
// Sub-data waiting for a sync that nothing of this session would otherwise trigger.
if (t_boundNativeContext != EGL_NO_CONTEXT) TextureImpl::WritePendingSharedImageUploads();
if (frame.Tracker.Pending() == 0) return true;
@@ -18994,6 +19206,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool ExportNativeFence(int* fence) {
*fence = -1;
ReadFrame& frame = CurrentReadFrame();
RenewYuvEpoch(frame);
// No context current: nothing of this session is in flight to fence.
if (t_boundNativeContext == EGL_NO_CONTEXT) return true;
TextureImpl::WritePendingSharedImageUploads();
@@ -19015,6 +19228,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
void PublishPendingReads() {
ReadFrame& frame = CurrentReadFrame();
RenewYuvEpoch(frame);
if (frame.Tracker.Pending() == 0) return;
int fence = -1;
// No context current: the reads' context is gone (and its work with it), and there is
@@ -19207,6 +19421,15 @@ namespace MobileGL::MG_Backend::DirectGLES {
return false;
}
void AcquireForSampling(const EglImageRef& image) { (void)image; }
Bool IsYuv(const EglImageRef& image) {
(void)image;
return false;
}
void ConvertYuvForSampling(const EglImageRef& image, GLuint target, Uint32 width, Uint32 height) {
(void)image, (void)target, (void)width, (void)height;
}
void PublishPendingReads() {}
Bool PublishPendingAccesses() { return true; }
Bool ExportNativeFence(int* fence) {
@@ -548,6 +548,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
// landed since this context last waited, the context's later commands wait for its fence
// on the GPU; and the image is noted as read by the calling session's current frame.
void AcquireForSampling(const EglImageRef& image);
// YUV IMAGES (NV12/P010, MG_Remote/Server/SharedImageYuv.h) are never a name's level 0: the
// name keeps its own RGBA8 storage `target` (width x height), into which the image is
// converted at its first use in each of the calling session's frames - the image bound as
// an external texture, one triangle drawn - after the registry's CPU copy for a foreign
// buffer. Every touched binding is put back. Called after AcquireForSampling.
Bool IsYuv(const EglImageRef& image);
void ConvertYuvForSampling(const EglImageRef& image, GLuint target, Uint32 width, Uint32 height);
// The calling session's frame boundary for what it has read: hands a fence for everything
// the current context has submitted to every image noted since the last boundary. Called
// at Present and wherever the current native context stops being current (its reads can
+26 -13
View File
@@ -2898,16 +2898,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
kT0MapReadBit | kT0MapWriteBit | kMapPersistentBit | kMapCoherentBit | kDynamicStorageBit;
constexpr GLbitfield kT0MapFlags = kT0MapReadBit | kT0MapWriteBit | kMapPersistentBit | kMapCoherentBit;
Bool HasExtensionWord(const char* list, const char* word) {
if (list == nullptr) return false;
const SizeT n = std::strlen(word);
for (const char* at = std::strstr(list, word); at != nullptr; at = std::strstr(at + 1, word)) {
const Bool startOk = at == list || at[-1] == ' ';
const Bool endOk = at[n] == '\0' || at[n] == ' ';
if (startOk && endOk) return true;
}
return false;
}
using MG_Util::BackendLoader::HasExtensionWord;
Bool ResolveT0EntryPoints(String& why) {
if (g_t0GetNativeClientBuffer != nullptr && g_t0BufferStorageExternal != nullptr) return true;
@@ -3903,8 +3894,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
MG_Pipe::MGPipeSetResourceOps(&g_glesResourceOps);
#endif
// Frontend writes issued while ops were unregistered advanced change
// serials with no per-op bump; re-open every draw-clean memo.
BumpBufferMutationEpoch();
// serials with no per-op bump; re-open every draw-clean memo. Advanced directly,
// past BumpBufferMutationEpoch's apply-thread pin: a registration mutates no buffer
// and only makes every memo probe again, and with several sessions in one process
// a new session's backend registers on its own thread while another session's
// apply thread is already running - which the pin would otherwise read as a
// client-thread producer and, under the strict lane, abort the server for.
g_bufferMutationEpoch.fetch_add(1, std::memory_order_release);
}
void UnregisterBufferBackendOps() {
@@ -8893,7 +8889,20 @@ namespace MobileGL::MG_Backend::DirectGLES {
return true;
}
Bind(GL_TEXTURE_2D);
if (!SharedImageImpl::TargetBoundTexture2D(image)) {
if (SharedImageImpl::IsYuv(image)) {
// A YUV image is CONVERTED into the name's own RGBA8 level at each frame's first use
// (SharedImageImpl::ConvertYuvForSampling), never made its storage.
while (g_GLESFuncs.glGetError() != GL_NO_ERROR) {
}
g_GLESFuncs.glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, static_cast<GLsizei>(record.Desc.Width),
static_cast<GLsizei>(record.Desc.Height));
if (g_GLESFuncs.glGetError() != GL_NO_ERROR) {
MGLOG_E_ONCE("Shared image texture %u: no %ux%u RGBA8 storage for YUV image %llu; left empty",
record.Desc.GlNameForDiag, record.Desc.Width, record.Desc.Height,
static_cast<unsigned long long>(imageId));
return true;
}
} else if (!SharedImageImpl::TargetBoundTexture2D(image)) {
MGLOG_E_ONCE("Shared image texture %u: glEGLImageTargetTexture2DOES refused image %llu; left empty",
record.Desc.GlNameForDiag, static_cast<unsigned long long>(imageId));
return true;
@@ -9042,6 +9051,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
// before the command this sync is for, and the session's frame notes the read.
if (m_sharedImage != nullptr) SharedImageImpl::AcquireForSampling(m_sharedImage);
ReplaySharedImageUploads(*pushedStorage);
// A YUV image is converted into the name at the frame's first use.
if (m_sharedImage != nullptr && SharedImageImpl::IsYuv(m_sharedImage))
SharedImageImpl::ConvertYuvForSampling(m_sharedImage, GetBackendTextureId(),
pushedStorage->Desc.Width, pushedStorage->Desc.Height);
return;
}
// P5e (tx2), CONTRACT-P5E §5.2 (scout G-S2-2): THE VIEW TEST MOVES BEHIND THE
@@ -1944,6 +1944,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
switch (target) {
case TextureTarget::Texture1D:
case TextureTarget::TextureRectangle:
case TextureTarget::External:
return TextureTarget::Texture2D;
case TextureTarget::Texture1DArray:
return TextureTarget::Texture2DArray;
@@ -66,7 +66,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
FreeRetiredCommandBuffers(frame);
}
vkFreeCommandBuffers(device, commandPool, frameCount * 2, commandBuffers.data());
if (!m_spareCommandBuffers.empty())
vkFreeCommandBuffers(device, commandPool, static_cast<Uint32>(m_spareCommandBuffers.size()),
m_spareCommandBuffers.data());
}
m_spareCommandBuffers.clear();
m_frames.clear();
currentFrameIndex = 0;
m_device = VK_NULL_HANDLE;
@@ -362,21 +366,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MOBILEGL_ASSERT(!frame.isPreCommandRecording,
"RetireCurrentCommandBuffer called while the pre-pass stream is still recording");
VkCommandBufferAllocateInfo allocInfo{};
allocInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
allocInfo.commandPool = m_commandPool;
allocInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
allocInfo.commandBufferCount = 1;
VkCommandBuffer replacement = VK_NULL_HANDLE;
VkResult result = vkAllocateCommandBuffers(m_device, &allocInfo, &replacement);
VkResult result = TakeCommandBuffer(replacement);
if (result != VK_SUCCESS) {
return result;
}
if (retirePreCommandBuffer) {
VkCommandBuffer preReplacement = VK_NULL_HANDLE;
result = vkAllocateCommandBuffers(m_device, &allocInfo, &preReplacement);
result = TakeCommandBuffer(preReplacement);
if (result != VK_SUCCESS) {
vkFreeCommandBuffers(m_device, m_commandPool, 1, &replacement);
RecycleCommandBuffer(replacement);
return result;
}
frame.retiredCommandBuffers.push_back({frame.preCommandBuffer, frame.lastSubmitIndex});
@@ -411,8 +410,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
SizeT completedCount = 0;
while (completedCount < frame.retiredCommandBuffers.size() &&
frame.retiredCommandBuffers[completedCount].submitIndex <= completedSubmitIndex) {
vkFreeCommandBuffers(m_device, m_commandPool, 1,
&frame.retiredCommandBuffers[completedCount].commandBuffer);
RecycleCommandBuffer(frame.retiredCommandBuffers[completedCount].commandBuffer);
++completedCount;
}
if (completedCount > 0) {
@@ -422,6 +420,31 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
VkResult FrameContext::TakeCommandBuffer(VkCommandBuffer& out) {
if (!m_spareCommandBuffers.empty()) {
out = m_spareCommandBuffers.back();
m_spareCommandBuffers.pop_back();
return VK_SUCCESS;
}
VkCommandBufferAllocateInfo allocInfo{};
allocInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
allocInfo.commandPool = m_commandPool;
allocInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
allocInfo.commandBufferCount = 1;
return vkAllocateCommandBuffers(m_device, &allocInfo, &out);
}
void FrameContext::RecycleCommandBuffer(VkCommandBuffer commandBuffer) {
// A handful covers the buffers in flight between two completions; beyond that, free.
constexpr SizeT kSpareCommandBuffers = 8;
if (commandBuffer == VK_NULL_HANDLE) return;
if (m_spareCommandBuffers.size() < kSpareCommandBuffers) {
m_spareCommandBuffers.push_back(commandBuffer);
return;
}
vkFreeCommandBuffers(m_device, m_commandPool, 1, &commandBuffer);
}
void FrameContext::FreeAllRetiredCommandBuffers() {
for (auto& frame : m_frames) {
FreeRetiredCommandBuffers(frame);
@@ -150,5 +150,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Stored at Initialize for retired-command-buffer management.
VkDevice m_device = VK_NULL_HANDLE;
VkCommandPool m_commandPool = VK_NULL_HANDLE;
// Completed command buffers kept for the next retirement instead of being freed and
// reallocated: the pool resets a buffer implicitly at its next vkBeginCommandBuffer.
Vector<VkCommandBuffer> m_spareCommandBuffers;
VkResult TakeCommandBuffer(VkCommandBuffer& out);
void RecycleCommandBuffer(VkCommandBuffer commandBuffer);
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -2529,6 +2529,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return SamplerNumericDomain::UnsignedInteger;
case GL_SAMPLER_1D:
case GL_SAMPLER_2D:
case GL_SAMPLER_EXTERNAL_OES:
case GL_SAMPLER_3D:
case GL_SAMPLER_CUBE:
case GL_SAMPLER_2D_RECT:
@@ -2719,6 +2720,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
case GL_UNSIGNED_INT_IMAGE_CUBE_MAP_ARRAY:
return TextureTarget::TextureCubeMapArray;
case GL_SAMPLER_2D:
case GL_SAMPLER_EXTERNAL_OES: // a 2D texture below the frontend
case GL_SAMPLER_2D_SHADOW:
case GL_INT_SAMPLER_2D:
case GL_UNSIGNED_INT_SAMPLER_2D:
@@ -0,0 +1,27 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/RenderPassGuard.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include "../VkIncludes.h"
#include <Includes.h>
namespace MobileGL::MG_Backend::DirectVulkan {
// Ends the render pass that is active on `commandBuffer`, if there is one. Called right before
// recording a command a render pass may not contain (a barrier, a copy, a blit, a clear of an
// image) into a command buffer that may hold an open pass: the wire draw path keeps its pass
// open across consecutive draws to the same attachments, so the resource work that falls
// between two draws must close it first. A pass on another command buffer is left alone.
void EndActiveRenderPassOn(VkCommandBuffer commandBuffer);
// A counter that moves whenever images may have been written since it was last read: a render
// pass ended (its attachments), resource work recorded outside a pass (copies, clears, blits,
// dispatches all close the pass first), or a draw with storage images. A sampled image needs
// the memory barrier that makes earlier writes visible only when this moved since the last one.
Uint64 WireImageWriteEpoch();
void BumpWireImageWriteEpoch();
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -6,6 +6,7 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "RenderPassGuard.h"
#include "UniformManager.h"
#include "MG_Backend/DirectVulkan/DirectVulkanResourceState.h"
@@ -251,11 +252,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
WireDescriptorFatal("image-descriptor-transition");
// A second descriptor/draw can keep GENERAL after a shader write. A layout
// equality fast return is not a memory dependency for that prior write.
VkMemoryBarrier memory{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
memory.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
memory.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &memory, 0, nullptr, 0, nullptr);
// Only when something may have written an image since the last such barrier
// (WireImageWriteEpoch): with nothing written in between, consecutive draws keep their
// render pass open instead of closing it for a barrier that orders nothing.
if (WireImageWriteEpoch() != m_wireImageBarrierEpoch) {
VkMemoryBarrier memory{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
memory.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
memory.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &memory, 0, nullptr, 0, nullptr);
m_wireImageBarrierEpoch = WireImageWriteEpoch();
}
VkImageViewCreateInfo info{VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO};
info.image = resource->image;
info.viewType = type;
@@ -3198,7 +3206,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const auto* resource = m_textureManager->SyncTextureResourceByHandle(handle, false, storage);
if (resource == nullptr) return false;
// Another session writes it: the renderer acquires it before this use.
if (resource->sharedImageId != 0) m_textureManager->NoteSharedImageUse(*resource);
if (resource->sharedImageId != 0 || resource->yuvImageId != 0)
m_textureManager->NoteSharedImageUse(*resource);
}
}
}
@@ -660,5 +660,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
m_samplerResolveMemoHighWater = binding + 1;
}
}
// WireImageWriteEpoch() when ResolveWireImageDescriptor last made image writes visible.
mutable Uint64 m_wireImageBarrierEpoch = 0;
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -6,6 +6,7 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "RenderPassGuard.h"
#include "VkBufferManager.h"
#include "../DirectVulkan.h"
#include "VulkanRenderer.h"
@@ -337,6 +338,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier barrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
barrier.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_HOST_READ_BIT | VK_ACCESS_HOST_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_HOST_BIT,
0, 1, &barrier, 0, nullptr, 0, nullptr);
// P8-D: counted here, past the idle early-out, so `whw` is waits TAKEN, not calls.
@@ -531,13 +533,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier before{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
before.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_HOST_WRITE_BIT;
before.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT | VK_PIPELINE_STAGE_HOST_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1, &before, 0, nullptr, 0, nullptr);
const VkBufferCopy copy{offset, dst.offset, size};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, resource->buffer.GetHandle(), dst.buffer, 1, &copy);
VkMemoryBarrier after{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
after.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
after.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &after, 0, nullptr, 0, nullptr);
resource->lastUseSerial = m_frameSerial;
@@ -575,22 +580,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier before{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
before.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_HOST_WRITE_BIT;
before.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT | VK_PIPELINE_STAGE_HOST_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1, &before, 0, nullptr, 0, nullptr);
const VkBufferCopy widen{alignedOffset, staging.offset, stagedSize};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, resource->buffer.GetHandle(), staging.buffer, 1, &widen);
// The shift reads what the widening copy just wrote, so it needs its own edge even
// though both halves are transfers on one command buffer.
VkMemoryBarrier staged{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
staged.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
staged.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
0, 1, &staged, 0, nullptr, 0, nullptr);
const VkBufferCopy shift{staging.offset + headPad, dst.offset + dstSkip, size};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, staging.buffer, dst.buffer, 1, &shift);
VkMemoryBarrier after{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
after.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
after.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &after, 0, nullptr, 0, nullptr);
resource->lastUseSerial = m_frameSerial;
@@ -908,14 +918,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier hostWrites{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
hostWrites.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT;
hostWrites.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&hostWrites, 0, nullptr, 0, nullptr);
VkBufferCopy region{0, 0, kWords * 4};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, imported.buffer, staging.buffer, 1, &region);
EndActiveRenderPassOn(commands);
vkCmdFillBuffer(commands, imported.buffer, kFillAt, kWords * 4, kFillWord);
VkMemoryBarrier toHost{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
toHost.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
toHost.dstAccessMask = VK_ACCESS_HOST_READ_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0, 1, &toHost,
0, nullptr, 0, nullptr);
vkEndCommandBuffer(commands);
@@ -965,11 +979,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
verdict = "steady: the second submission could not be recorded";
break;
}
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&hostWrites, 0, nullptr, 0, nullptr);
VkBufferCopy steadyRegion{kSteadyAt, 0, kWords * 4};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, imported.buffer, staging.buffer, 1, &steadyRegion);
EndActiveRenderPassOn(commands);
vkCmdFillBuffer(commands, imported.buffer, kFillAt, kWords * 4, kSteadyFillWord);
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0, 1, &toHost,
0, nullptr, 0, nullptr);
vkEndCommandBuffer(commands);
@@ -1488,6 +1506,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier beforeBarrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
beforeBarrier.srcAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
beforeBarrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&beforeBarrier, 0, nullptr, 0, nullptr);
@@ -1495,11 +1514,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
region.srcOffset = staging.offset;
region.dstOffset = static_cast<VkDeviceSize>(offset);
region.size = static_cast<VkDeviceSize>(size);
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBuffer(commandBuffer, staging.buffer, resource.buffer.GetHandle(), 1, &region);
VkMemoryBarrier afterBarrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
afterBarrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
afterBarrier.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1,
&afterBarrier, 0, nullptr, 0, nullptr);
@@ -1534,6 +1555,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier beforeBarrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
beforeBarrier.srcAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
beforeBarrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&beforeBarrier, 0, nullptr, 0, nullptr);
@@ -1541,11 +1563,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
region.srcOffset = staging.offset;
region.dstOffset = static_cast<VkDeviceSize>(offset);
region.size = static_cast<VkDeviceSize>(size);
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBuffer(commandBuffer, staging.buffer, resource.buffer.GetHandle(), 1, &region);
VkMemoryBarrier afterBarrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
afterBarrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
afterBarrier.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1,
&afterBarrier, 0, nullptr, 0, nullptr);
@@ -7,6 +7,8 @@
// End of Source File Header
#include "VkRenderPassManager.h"
#include "RenderPassGuard.h"
#include <atomic>
#include "MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.h"
#include "MG_State/GLState/TextureState/TextureObject2D.h"
@@ -1666,6 +1668,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
s_clearManager.Get()->PopPendingClear(pending.key);
}
}
s_activeRenderPass.Get().commandBuffer = commandBuffer;
s_activeRenderPass.Get().hash = renderPassEntry.hash;
s_activeRenderPass.Get().compatibilityHash = renderPassEntry.compatibilityHash;
s_activeRenderPass.Get().trackedAttachmentLayouts = renderPassEntry.trackedAttachmentLayouts;
@@ -1676,15 +1679,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
Bool VkRenderPassManager::EndRenderPass(VkCommandBuffer commandBuffer) {
BumpWireImageWriteEpoch();
auto* activeRenderPass = GetActiveRenderPass();
Vector<TrackedAttachmentLayoutInfo> trackedAttachmentLayouts;
if (activeRenderPass != nullptr) {
trackedAttachmentLayouts = Move(activeRenderPass->trackedAttachmentLayouts);
}
vkCmdEndRenderPass(commandBuffer);
// Layout reconciliation below may record barriers, whose guard closes an active
// pass. Detach the ended pass first so that guard cannot recursively end it again.
s_activeRenderPass.Get() = {};
s_hasActiveRenderPass.Get() = false;
// The fast-path memo reuses the ACTIVE render pass; once the pass ends it must not carry
// over (the next span may be a different FBO resolved before its render pass is begun).
if (s_renderPassManager.Get() != nullptr) {
s_renderPassManager.Get()->m_rpFastValid = false;
}
if (activeRenderPass != nullptr && !activeRenderPass->trackedAttachmentLayouts.empty()) {
for (const auto& trackedAttachment : activeRenderPass->trackedAttachmentLayouts) {
if (!trackedAttachmentLayouts.empty()) {
for (const auto& trackedAttachment : trackedAttachmentLayouts) {
switch (trackedAttachment.target) {
case TrackedAttachmentTarget::Texture:
MOBILEGL_ASSERT(s_textureManager.Get() != nullptr, "EndRenderPass: texture manager is null");
@@ -1726,12 +1738,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
}
s_activeRenderPass.Get() = {};
s_hasActiveRenderPass.Get() = false;
return true;
}
ActiveRenderPassInfo* VkRenderPassManager::GetActiveRenderPass() {
return s_hasActiveRenderPass.Get() ? &s_activeRenderPass.Get() : nullptr;
}
namespace {
std::atomic<Uint64> g_wireImageWriteEpoch{1};
}
Uint64 WireImageWriteEpoch() { return g_wireImageWriteEpoch.load(std::memory_order_relaxed); }
void BumpWireImageWriteEpoch() { g_wireImageWriteEpoch.fetch_add(1, std::memory_order_relaxed); }
void EndActiveRenderPassOn(VkCommandBuffer commandBuffer) {
BumpWireImageWriteEpoch();
const auto* active = VkRenderPassManager::GetActiveRenderPass();
if (active != nullptr && active->commandBuffer == commandBuffer && commandBuffer != VK_NULL_HANDLE)
VkRenderPassManager::EndRenderPass(commandBuffer);
}
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -185,6 +185,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
struct ActiveRenderPassInfo {
Uint64 hash = 0;
Uint64 compatibilityHash = 0;
VkCommandBuffer commandBuffer = VK_NULL_HANDLE;
Vector<TrackedAttachmentLayoutInfo> trackedAttachmentLayouts;
IntVec2 extent = {0, 0};
@@ -6,6 +6,7 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "RenderPassGuard.h"
#include "VkTextureManager.h"
#include "ProgramFactory.h"
@@ -55,6 +56,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
};
static SessionLocal<SampledReadStages> s_sampledReadStagesSlot;
static VkImageUsageFlags ResolveTextureUsage(VkImageAspectFlags aspect,
VkFormatFeatureFlags formatFeatures,
Bool supportsStorageImage) {
VkImageUsageFlags usage = VK_IMAGE_USAGE_SAMPLED_BIT;
if (supportsStorageImage) usage |= VK_IMAGE_USAGE_STORAGE_BIT;
// A color aspect does not imply renderability (for example, RGB9_E5).
if ((aspect & VK_IMAGE_ASPECT_COLOR_BIT) &&
(formatFeatures & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT)) {
usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
}
if (aspect & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
}
return usage;
}
static Uint32 ComputeFullMipLevelCount(const IntVec3& baseTexelSize) {
Int maxDimension = std::max<Int>(baseTexelSize.x(),
std::max<Int>(baseTexelSize.y(), std::max<Int>(baseTexelSize.z(), 1)));
@@ -376,6 +393,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
copyRegions.push_back(copy);
}
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyImage(commandBuffer,
oldResource.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
newResource.image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
@@ -1658,6 +1676,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
std::max(source->depth >> level, 1u)};
copyRegions.push_back(copy);
}
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyImage(commandBuffer, source->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, snapshot.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, static_cast<Uint32>(copyRegions.size()), copyRegions.data());
@@ -1772,6 +1791,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// slices" but will read as "depth slice 0" once VK_KHR_maintenance9 is enabled. The
// validation layer warns about that literal 1 by name.
barrier.subresourceRange.layerCount = VK_REMAINING_ARRAY_LAYERS;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, srcStageMask, dstStageMask, 0, 0, nullptr, 0, nullptr, 1, &barrier);
trackedLayout = newLayout;
@@ -2115,9 +2135,41 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool VkTextureManager::SyncWireTextureShape(const MG_Pipe::MGPipeResourceRecord& record,
TextureResource& resource, Bool requireStorage) {
const MG_Pipe::MGPResourceDesc& desc = record.Desc;
// A YUV shared image is converted INTO the texture's own RGBA8 storage, which the rest of
// this function defines as for any texture; the image is only named here, and the
// renderer converts it at its first use in each frame (WireYuvImage.inc).
if (record.SharedImageId != resource.yuvImageId && !m_syncingYuvTextureStorage) {
// Respecified away from (or to another) YUV image: its source goes with it, once the
// GPU is done with it (a released resource's Reset is deferred).
if (resource.yuvSource != nullptr) {
TextureResource retired;
retired.yuvSource = std::move(resource.yuvSource);
DeferResourceRelease(Move(retired));
}
resource.yuvImageId = 0;
resource.yuvOwner.reset();
resource.yuvSource.reset();
resource.yuvConvertedFrame = 0;
}
if (record.SharedImageId != 0 && !m_syncingYuvTextureStorage) {
const auto yuv = MG_Remote::Server::SharedImages::Find(record.SharedImageId);
if (yuv != nullptr && MG_Remote::Server::SharedImages::FourccIsYuv(yuv->Fourcc)) {
m_syncingYuvTextureStorage = true;
const Bool defined = SyncWireTextureShape(record, resource, requireStorage);
m_syncingYuvTextureStorage = false;
if (!defined) return false;
if (resource.yuvImageId != record.SharedImageId) {
resource.yuvImageId = record.SharedImageId;
resource.yuvOwner = yuv;
resource.yuvSource.reset();
resource.yuvConvertedFrame = 0;
}
return true;
}
}
// Level 0 is a shared image: bind its buffer. An image that cannot be bound here leaves
// the texture the storage its NULL-data definition asked for, below.
if (record.SharedImageId != 0) {
if (record.SharedImageId != 0 && !m_syncingYuvTextureStorage) {
const SharedImageBind bound = SyncWireSharedImage(record, resource, requireStorage);
if (bound != SharedImageBind::Unresolved) return bound == SharedImageBind::Bound;
}
@@ -2220,13 +2272,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
VkImageUsageFlags desiredUsage =
VK_IMAGE_USAGE_SAMPLED_BIT | (supportsStorageImage ? VK_IMAGE_USAGE_STORAGE_BIT : 0) |
((aspect & VK_IMAGE_ASPECT_COLOR_BIT) &&
(formatProperties.optimalTilingFeatures & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT)
? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT : 0) |
(((aspect & VK_IMAGE_ASPECT_DEPTH_BIT) || (aspect & VK_IMAGE_ASPECT_STENCIL_BIT))
? VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT
: 0);
ResolveTextureUsage(aspect, formatProperties.optimalTilingFeatures, supportsStorageImage);
// Multisample attachments are also transfer sources for vkCmdResolveImage.
desiredUsage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
@@ -2702,9 +2748,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
stencilCopy.imageSubresource.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
stencilCopy.bufferOffset = stagingBase + item.offset + static_cast<VkDeviceSize>(texelCount) * 4;
const VkBufferImageCopy copies[2] = {depthCopy, stencilCopy};
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, resource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 2, copies);
} else {
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, resource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &copy);
}
@@ -3057,7 +3105,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
handle.Slot, handle.Gen);
return nullptr;
}
resource.sharedImageKey = resource.sharedImageId != 0 ? key : 0;
resource.sharedImageKey = resource.sharedImageId != 0 || resource.yuvImageId != 0 ? key : 0;
resource.syncedWireSerial = record.Serial;
return &resource;
}
@@ -3179,6 +3227,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if ((aspect & copyAspect) == 0) continue;
copy.srcSubresource.aspectMask = copyAspect;
copy.dstSubresource.aspectMask = copyAspect;
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyImage(commandBuffer, old.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, grown.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &copy);
}
@@ -3474,12 +3523,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
VkImageUsageFlags desiredUsage =
VK_IMAGE_USAGE_SAMPLED_BIT |
(supportsStorageImage ? VK_IMAGE_USAGE_STORAGE_BIT : 0) |
((aspect & VK_IMAGE_ASPECT_COLOR_BIT) ? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT : 0) |
(((aspect & VK_IMAGE_ASPECT_DEPTH_BIT) || (aspect & VK_IMAGE_ASPECT_STENCIL_BIT)) ?
VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT :
0);
ResolveTextureUsage(aspect, formatProperties.optimalTilingFeatures, supportsStorageImage);
if (!isMultisampleTexture) {
desiredUsage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
}
@@ -3774,7 +3818,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
resource.sampledView == VK_NULL_HANDLE &&
resource.perMipViews.empty() && resource.perMipSampledViews.empty() &&
resource.attachmentViews.empty() && resource.alternateSampledViews.empty() &&
resource.storageImageViews.empty()) {
resource.storageImageViews.empty()
#if MOBILEGL_BUILD_DISAGGREGATED
&& resource.yuvSource == nullptr
#endif
) {
return;
}
@@ -4734,6 +4782,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
rectCopies[rectCopyCount++] = rectCopy;
runningOffset += static_cast<VkDeviceSize>(rect.TexelCount() * item.texelBytes);
}
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, outResource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, rectCopyCount, rectCopies.data());
continue;
@@ -4775,10 +4824,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
stencilCopy.imageSubresource.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
stencilCopy.bufferOffset = stagingBase + item.offset + static_cast<VkDeviceSize>(texelCount) * 4;
const VkBufferImageCopy copies[2] = {depthCopy, stencilCopy};
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, outResource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 2, copies);
continue;
}
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, outResource.image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
1, &copy);
}
@@ -372,6 +372,15 @@ public:
Bool sharedImageHeld = false;
Uint64 sharedImageGeneration = 0;
Uint64 sharedImageKey = 0;
// A YUV shared image (NV12/P010) is never the texture's image: the texture keeps its own
// RGBA8 level 0, into which the renderer converts the image at its first use in each frame
// (WireYuvImage.inc) - `yuvSource` being that conversion's import of the buffer, made on
// first use, and `yuvConvertedFrame` the frame it last ran in. `yuvOwner` holds the
// registry's image for as long as the texture names it.
Uint64 yuvImageId = 0;
SharedPtr<const void> yuvOwner;
SharedPtr<void> yuvSource;
Uint64 yuvConvertedFrame = 0;
#endif
TextureResource() = default;
@@ -415,6 +424,10 @@ public:
std::swap(this->sharedImageHeld, that.sharedImageHeld);
std::swap(this->sharedImageGeneration, that.sharedImageGeneration);
std::swap(this->sharedImageKey, that.sharedImageKey);
std::swap(this->yuvImageId, that.yuvImageId);
std::swap(this->yuvOwner, that.yuvOwner);
std::swap(this->yuvSource, that.yuvSource);
std::swap(this->yuvConvertedFrame, that.yuvConvertedFrame);
#endif
}
@@ -465,6 +478,10 @@ public:
sharedImageHeld = false;
sharedImageGeneration = 0;
sharedImageKey = 0;
yuvImageId = 0;
yuvOwner.reset();
yuvSource.reset();
yuvConvertedFrame = 0;
#endif
fullView = VK_NULL_HANDLE;
sampledView = VK_NULL_HANDLE;
@@ -882,6 +899,8 @@ private:
// Active only between BeginDrawSyncScope/EndDrawSyncScope; identities of
// textures already fully synced in the current draw (small N -> flat scan).
Bool m_drawSyncScopeActive = false;
// SyncWireTextureShape defining the RGBA8 storage a YUV shared image is converted into.
Bool m_syncingYuvTextureStorage = false;
// Per-draw sync memo: the identity plus the resolved resource pointer. The pointer is stable
// across rehash in the node-based m_textureResources and stays valid for the draw (a texture
// synced this draw is alive and is not erased mid-draw), so a repeat sync of the same texture
@@ -6,6 +6,7 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "RenderPassGuard.h"
#include "VulkanRenderer.h"
#include "SubmitFencePrefix.h"
#include "MG_Util/X11/DisplayGuard.h"
@@ -3525,6 +3526,7 @@ void main() {
CollectWireObjects(m_submitCounter, true);
ClearAllWireDrawPassCaches();
DestroyWireColorBlitResources();
DestroyWireYuvResources();
DestroyWireDepthMipmapResources();
DestroyWireMultisampleResolveResources();
DestroySharedImagePresentTargets(false);
@@ -5199,6 +5201,7 @@ void main() {
static_cast<Uint32>(dstTexelSize.x()), static_cast<Uint32>(dstTexelSize.y())
};
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdBeginRenderPass(frame.commandBuffer, &renderPassBeginInfo, VK_SUBPASS_CONTENTS_INLINE);
VkViewport viewport{};
@@ -7828,6 +7831,7 @@ void main() {
}
MGLOG_D("DirectVulkan: glDispatchCompute(%u, %u, %u)", numGroupsX, numGroupsY, numGroupsZ);
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdDispatch(frame.commandBuffer, numGroupsX, numGroupsY, numGroupsZ);
}
@@ -7905,6 +7909,7 @@ void main() {
}
MGLOG_D("DirectVulkan: glDispatchComputeIndirect(offset=%zu)", static_cast<SizeT>(indirect));
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdDispatchIndirect(frame.commandBuffer, slice.buffer, slice.offset + static_cast<VkDeviceSize>(indirect));
}
@@ -8107,6 +8112,8 @@ void main() {
#include "WireDraw.inc"
// AFTER WireFramebuffer.inc: the shared-image present is that file's readback, into an image.
#include "WireSharedImage.inc"
// AFTER WireSharedImage.inc: a YUV image is that file's reading side, converted on the way.
#include "WireYuvImage.inc"
void VulkanRenderer::Clear(GLbitfield mask) {
#if MOBILEGL_BUILD_DISAGGREGATED
@@ -8747,6 +8754,7 @@ void main() {
beginInfo.clearValueCount = 1;
beginInfo.pClearValues = &clearValue;
// The load op is the whole operation: begin and end with nothing in between.
EndActiveRenderPassOn(commandBuffer);
vkCmdBeginRenderPass(commandBuffer, &beginInfo, VK_SUBPASS_CONTENTS_INLINE);
vkCmdEndRenderPass(commandBuffer);
@@ -10670,15 +10678,6 @@ void main() {
return;
}
auto& frame = m_frameContext.GetCurrent();
if (!frame.isCommandRecording) {
m_frameContext.BeginCommandRecording();
}
if (VkRenderPassManager::GetActiveRenderPass() != nullptr) {
VkRenderPassManager::EndRenderPass(frame.commandBuffer);
}
// One resolver for both object kinds. The texture arm is the same
// SyncTextureAndGetDescriptor the copy always used; the renderbuffer arm goes through the
// render-pass manager, which is where a renderbuffer's VkImage lives.
@@ -10930,6 +10929,16 @@ void main() {
}
#endif
// Syncing either endpoint can flush and retire the current command buffer for
// a texture upload. Prepare recording only after both images and uploads are ready.
auto& frame = m_frameContext.GetCurrent();
if (!frame.isCommandRecording) {
m_frameContext.BeginCommandRecording();
}
if (VkRenderPassManager::GetActiveRenderPass() != nullptr) {
VkRenderPassManager::EndRenderPass(frame.commandBuffer);
}
const auto materializeClear = [this, &frame](const CopyImageEndpoint& endpoint) {
#if MOBILEGL_BUILD_DISAGGREGATED
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) return true;
@@ -16032,6 +16041,25 @@ void main() {
}
}
#if MOBILEGL_BUILD_DISAGGREGATED
// YUV shared images (WireYuvImage.inc) are sampled through a VkSamplerYcbcrConversion, a
// feature of its own; asked only where AHardwareBuffers are imported at all.
m_samplerYcbcrConversion = false;
VkPhysicalDeviceSamplerYcbcrConversionFeatures ycbcrFeatures{};
ycbcrFeatures.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES;
if (m_wireAhbImport && getPhysicalDeviceFeatures2 != nullptr) {
VkPhysicalDeviceFeatures2 featureQuery{};
featureQuery.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
featureQuery.pNext = &ycbcrFeatures;
getPhysicalDeviceFeatures2(m_physicalDevice.handle, &featureQuery);
if (ycbcrFeatures.samplerYcbcrConversion == VK_TRUE) {
ycbcrFeatures.pNext = const_cast<void*>(deviceCreateInfo.pNext);
deviceCreateInfo.pNext = &ycbcrFeatures;
m_samplerYcbcrConversion = true;
}
}
#endif
// VK_EXT_multi_draw: tier 1 of the multi-draw dispatch - one vkCmdDrawMulti(Indexed)EXT
// for a whole glMultiDraw* batch (VkMultiDrawIndexedInfoEXT carries per-draw
// firstIndex/indexCount/vertexOffset, so glMultiDrawElementsBaseVertex fits natively).
@@ -146,6 +146,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// IBufferCopyCommandProvider: recording command buffer, outside any
// render pass, for immediate staged buffer copies.
VkCommandBuffer AcquireBufferCopyCommandBuffer() override;
// EGL_BUFFER_AGE_EXT of the active target's default framebuffer for the next frame: the age
// of the swapchain image acquired for it (SwapchainObject::BufferAgeOf). Asking makes the
// target keep its presented images' content from then on.
Int32 CurrentDrawBufferAge();
#if MOBILEGL_BUILD_DISAGGREGATED
VkBufferManager& GetWireBufferManager() { return m_bufferManager; }
Bool FlushWirePendingCommandsForTextureUpdate() {
@@ -201,10 +205,6 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// submission's sync_file in `*fence` (-1: none, and the work is waited out here), and - for
// a session that writes images - the boundary PublishSharedImageAccesses makes.
Bool ExportNativeFence(int* fence);
// EGL_BUFFER_AGE_EXT of the active target's default framebuffer for the next frame: the age
// of the swapchain image acquired for it (SwapchainObject::BufferAgeOf). Asking makes the
// target keep its presented images' content from then on.
Int32 CurrentDrawBufferAge();
// The reading side's per-use hook (WireSharedImage.inc): every shared-image texture the
// wire draw or dispatch being set up samples (VkTextureManager::NoteSharedImageUse) is
// acquired from the foreign family before it, when it moved or is not held.
@@ -717,6 +717,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// one entry point the import needs (PFN_vkGetAndroidHardwareBufferPropertiesANDROID).
Bool m_wireAhbImport = false;
void* m_wireGetAhbProperties = nullptr;
// YUV shared images (WireYuvImage.inc): the device enabled samplerYcbcrConversion, and the
// conversion pipelines made so far (one per buffer format and colour conversion).
Bool m_samplerYcbcrConversion = false;
struct WireYuvResources;
WireYuvResources* m_wireYuvResources = nullptr;
void DestroyWireYuvResources();
// A YUV texture's first use in a frame converts its image (yuvConvertedFrame against this,
// which every frame boundary - present, publishing flush, native fence - moves).
Uint64 m_yuvFrameSerial = 1;
Bool ConvertYuvSharedImage(VkTextureManager::TextureResource& resource);
// Shared images this session presents into, imported once each and keyed by the image's
// id (monotonic, never reused). `owner` watches the registry's image: an entry whose image
// died is dropped at a later present once `lastSubmit`, its last copy, has completed.
@@ -378,6 +378,7 @@
begin.renderPass = target->renderPass;
begin.framebuffer = framebuffer;
begin.renderArea = {{0, 0}, destination.extent};
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdBeginRenderPass(frame.commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, target->pipeline);
const VkViewport viewport{0.0f, 0.0f, static_cast<float>(destination.extent.width),
@@ -299,6 +299,7 @@
begin.renderPass = target->renderPass;
begin.framebuffer = framebuffer;
begin.renderArea = {{0, 0}, destination.extent};
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdBeginRenderPass(frame.commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, target->pipeline);
const VkViewport viewport{0.0f, 0.0f, static_cast<float>(destination.extent.width),
@@ -11,6 +11,7 @@
// PrimitivesGeneratedNoXfbProbe.cpp): one throwaway command buffer, a bounded fence wait that
// leaks rather than idle-waits a hung GPU, teardown on every other path.
#include "RenderPassGuard.h"
#include "WireDepthResolveProbe.h"
#include <MG_Util/Debug/Log.h>
@@ -237,6 +238,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// The RECTANGLE is the render area, exactly as the render-pass arm's renderArea is - the
// scratch outside it is never read.
begin.renderArea = rect;
EndActiveRenderPassOn(commandBuffer);
vkCmdBeginRenderPass(commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
// THE VIEWPORT IS THE WHOLE ATTACHMENT AND THE SCISSOR IS THE RECT, and the two are not
@@ -746,6 +748,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
begin.renderPass = clear.pass;
begin.framebuffer = clear.framebuffer;
begin.renderArea = fullRect;
EndActiveRenderPassOn(cmd);
vkCmdBeginRenderPass(cmd, &begin, VK_SUBPASS_CONTENTS_INLINE);
VkClearAttachment attachment{};
attachment.aspectMask = clear.aspects;
@@ -772,6 +775,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
begin.renderPass = item.resolvePass;
begin.framebuffer = item.resolveFramebuffer;
begin.renderArea = fullRect;
EndActiveRenderPassOn(cmd);
vkCmdBeginRenderPass(cmd, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdEndRenderPass(cmd);
}
@@ -215,7 +215,10 @@
MakeXfbWritesVisible();
const auto* fbo = state.DrawFramebuffer();
if (!fbo) MagmaWireFatal("draw-framebuffer-record");
RetireWireDrawPass();
// THE PREVIOUS DRAW'S RENDER PASS STAYS OPEN until this draw's attachments are known: a
// draw to the same attachments continues it (below), and on a tiler a pass per draw also
// loaded and stored every attachment once per draw. Resource work in between that may not
// sit inside a pass (a barrier, a copy, a clear) closes it itself (RenderPassGuard.h).
// CPU preparation may synchronize GPU-produced vertex/index bytes. Do
// not rewind an arena while this draw owns as-yet-unrecorded slices.
WirePreparationScope preparing(m_wirePreparationDepth);
@@ -490,6 +493,7 @@
// attachments below resolve the same index, so this key has to be taken after it.
passKey.swapchainImageIndex = DefaultFramebufferWriteIndex(true);
}
Vector<WireImage> sameLayoutAttachments;
const auto add = [&](const MG_Pipe::MGPSurface& surface, VkImageAspectFlags aspect) -> Uint32 {
WireImage image = ResolveWireImage(*fbo, surface, aspect, /*isWriteTarget=*/true);
if (!image.image) return VK_ATTACHMENT_UNUSED;
@@ -499,7 +503,10 @@
MGB_CTX->IsCapabilityEnabled(CapabilityInput::FramebufferSrgb));
const VkImageLayout layout = aspect == VK_IMAGE_ASPECT_COLOR_BIT ?
VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_GENERAL;
TransitionWireImage(image, layout);
// Already in its attachment layout: its barrier (which closes any open pass) waits until
// it is known whether the open pass continues with this very attachment.
if (image.layout == layout) sameLayoutAttachments.push_back(image);
else TransitionWireImage(image, layout);
VkImageViewCreateInfo info{VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO};
info.image = image.image;
info.viewType = image.viewType;
@@ -565,7 +572,17 @@
passInfo.pAttachments = attachments.data();
passInfo.subpassCount = 1;
passInfo.pSubpasses = &subpass;
if (!AcquireCachedWireDrawPass(passKey)) {
const Bool continuing = m_wireDrawPass && m_wireDrawPassKey == passKey;
{
const auto* active = VkRenderPassManager::GetActiveRenderPass();
const Bool stillOpen = continuing && active != nullptr && active->commandBuffer == frame.commandBuffer;
// A new pass orders itself after earlier writes to its attachments; a continued one
// is the same pass, whose attachment writes are ordered by rasterization.
if (!stillOpen)
for (auto& image : sameLayoutAttachments) TransitionWireImage(image, VK_IMAGE_LAYOUT_GENERAL);
}
if (!continuing) RetireWireDrawPass();
if (!continuing && !AcquireCachedWireDrawPass(passKey)) {
m_wireDrawPassKey = Move(passKey);
Vector<VkImageView> views;
views.reserve(viewInfos.size());
@@ -629,13 +646,20 @@
vkCmdBindVertexBuffers(frame.commandBuffer, binding++, 1, &slice.buffer, &slice.offset);
}
if (indexSlice.IsValid()) vkCmdBindIndexBuffer(frame.commandBuffer, indexSlice.buffer, indexSlice.offset, indexType);
if (!VkRenderPassManager::BeginRenderPass(frame.commandBuffer, pass)) {
// Still open from the previous draw (nothing closed it, same command buffer): continue.
const auto* active = VkRenderPassManager::GetActiveRenderPass();
const Bool open = continuing && active != nullptr && active->commandBuffer == frame.commandBuffer &&
active->hash == pass.hash;
if (!open && active != nullptr) VkRenderPassManager::EndRenderPass(frame.commandBuffer);
if (!open && !VkRenderPassManager::BeginRenderPass(frame.commandBuffer, pass)) {
MGL_WIRE_DECLINE_AT(BeginRenderPass, "the render pass for this draw could not be begun");
return false;
}
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
ApplyDynamicDrawStateTail(frame, pass.extent, fbo->IsDefault,
ResolveDrawViewportCount(programObj.writesViewportIndexBuiltin));
// Storage images this draw writes are what the next sampled-image barrier must cover.
if (programObj.hasStorageImages) BumpWireImageWriteEpoch();
if (MG_Util::PipeStats::Enabled()) MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::Draws, 1);
return true;
}
@@ -800,6 +824,7 @@
return;
}
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdDispatch(frame.commandBuffer, x, y, z);
}
@@ -878,6 +903,7 @@
return;
}
vkCmdBindPipeline(dispatchFrame.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
EndActiveRenderPassOn(dispatchFrame.commandBuffer);
vkCmdDispatchIndirect(dispatchFrame.commandBuffer, parameters.buffer,
parameters.offset + static_cast<VkDeviceSize>(offset));
if (MG_Util::PipeStats::Enabled())
@@ -233,6 +233,13 @@
void VulkanRenderer::ClearWireFramebuffer(const MG_Pipe::MGPFramebufferState& fbo,
const ClearAttachmentPayload& payload, GLint drawbuffer) {
RetireWireDrawPass();
// A stream that neither presents nor finishes (an off-screen loop) never submits, so its
// recording and the per-clear objects retired below would grow until host memory runs
// out. Past a bound, submit what is recorded: the objects then retire with that work.
constexpr SizeT kRetiredObjectsBeforeSubmit = 1024;
if (m_wireRetiredObjects.size() >= kRetiredObjectsBeforeSubmit && m_wirePreparationDepth == 0 &&
HasPendingRecordedWork())
(void)FlushPendingCommands();
if (MGB_CTX->IsCapabilityEnabled(CapabilityInput::RasterizerDiscard)) return;
const auto clearSurface = [&](const MG_Pipe::MGPSurface& surface, VkImageAspectFlags aspect,
Uint32 colorIndex) {
@@ -319,6 +326,7 @@
if (!color) clear.clearValue.depthStencil = {payload.depth, payload.stencil};
VkClearRect rect{area, 0, image.layers};
auto& frame = m_frameContext.GetCurrent();
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdBeginRenderPass(frame.commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdClearAttachments(frame.commandBuffer, 1, &clear, 1, &rect);
vkCmdEndRenderPass(frame.commandBuffer);
@@ -830,6 +838,7 @@
begin.renderPass = pass;
begin.framebuffer = framebuffer;
begin.renderArea = {{sxMin, syMin}, {static_cast<Uint32>(width), static_cast<Uint32>(height)}};
EndActiveRenderPassOn(m_frameContext.GetCurrent().commandBuffer);
vkCmdBeginRenderPass(m_frameContext.GetCurrent().commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdEndRenderPass(m_frameContext.GetCurrent().commandBuffer);
} else if (!resolvedByShader) {
@@ -215,6 +215,7 @@
VkRenderPassBeginInfo begin{VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO};
begin.renderPass = image.clearPass; begin.framebuffer = image.clearFramebuffer;
begin.renderArea.extent = {1, 1}; begin.clearValueCount = 1; begin.pClearValues = &clear;
EndActiveRenderPassOn(commands);
vkCmdBeginRenderPass(commands, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdEndRenderPass(commands);
image.layout = VK_IMAGE_LAYOUT_GENERAL;
@@ -226,6 +227,7 @@
const VkImageSubresourceRange range{image.aspect, 0, 1, 0, image.layers};
if (shadow) {
const VkClearDepthStencilValue depth{0.0f, 0};
EndActiveRenderPassOn(commands);
vkCmdClearDepthStencilImage(commands, image.image, image.layout, &depth, 1, &range);
} else vkCmdClearColorImage(commands, image.image, image.layout, &clear.color, 1, &range);
if (!VkTextureManager::TransitionImageLayout(commands, image.image, image.layout,
@@ -236,6 +238,7 @@
VkMemoryBarrier visible{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
visible.srcAccessMask = multisample ? VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT : VK_ACCESS_TRANSFER_WRITE_BIT;
visible.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, multisample ? VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT : VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1, &visible, 0, nullptr, 0, nullptr);
}
@@ -499,6 +499,11 @@
void VulkanRenderer::AcquireSharedImage(VkTextureManager::TextureResource& resource) {
namespace SI = MG_Remote::Server::SharedImages;
// A YUV image is converted into the texture's own image instead (WireYuvImage.inc).
if (resource.yuvImageId != 0) {
(void)ConvertYuvSharedImage(resource);
return;
}
if (resource.sharedImageId == 0 || resource.image == VK_NULL_HANDLE || resource.sharedImageOwner == nullptr)
return;
const auto image = std::static_pointer_cast<const SI::Image>(resource.sharedImageOwner);
@@ -567,6 +572,8 @@
// family in the frame's last recording, and that recording's submission signals the semaphore
// whose sync_file is published as the frame's read fence for all of them at once.
void VulkanRenderer::ReleaseHeldSharedImages() {
// A frame boundary: a YUV texture's next use converts its image again.
++m_yuvFrameSerial;
if (m_sharedImagesHeld.empty()) return;
const Uint32 family = static_cast<Uint32>(m_physicalDevice.queueFamilies.graphicsFamily);
Vector<VkImageMemoryBarrier> releases;
@@ -633,6 +640,9 @@
// submission, and it is tracked like any other (the semaphore's reuse waits for it).
Bool VulkanRenderer::ExportNativeFence(int* fence) {
*fence = -1;
// Every fence the application asks for ends one of its frames as far as a YUV texture is
// concerned (an offscreen producer - a browser's GPU process - has no present to do it).
++m_yuvFrameSerial;
if (m_device == VK_NULL_HANDLE || m_graphicsQueue == VK_NULL_HANDLE || m_frameContext.GetFrameCount() == 0)
return true; // nothing was ever submitted: nothing to fence
// A session that writes images ends its frame here, as at a flush; a compositor that only
@@ -0,0 +1,10 @@
#version 450
// The image is bound through an immutable sampler that carries its VkSamplerYcbcrConversion, so
// a plain sample already returns RGB.
layout(set = 0, binding = 0) uniform sampler2D yuvImage;
layout(location = 0) in vec2 texCoord;
layout(location = 0) out vec4 outColor;
void main() {
outColor = vec4(textureLod(yuvImage, texCoord, 0.0).rgb, 1.0);
}
@@ -0,0 +1,11 @@
#version 450
// A YUV shared image converted into an RGBA texture (WireYuvImage.inc): one triangle over the
// whole target, its first row at the image's first row.
layout(location = 0) out vec2 texCoord;
void main() {
const vec2 corners[3] = vec2[](vec2(-1.0, -1.0), vec2(3.0, -1.0), vec2(-1.0, 3.0));
vec2 p = corners[gl_VertexIndex];
gl_Position = vec4(p, 0.0, 1.0);
texCoord = (p + 1.0) * 0.5;
}
@@ -0,0 +1,78 @@
// Generated from WireYuvConvert.vert / WireYuvConvert.frag (Vulkan 1.0 / SPIR-V 1.0) by
// scripts/bake_internal_shaders.py, which runs MG_Test's BakedInternalShadersTest with the
// IN-TREE glslang and splices the words it emits into the arrays below. Do not edit the words
// by hand: MOBILEGL_BAKED_INTERNAL_SHADERS (the same test, run as a gate) recompiles every
// source in its table and reds on the first word that disagrees.
#pragma once
static constexpr Uint32 kWireYuvConvertVertexSpirv[] = {
0x07230203u, 0x00010000u, 0x0008000bu, 0x00000031u, 0x00000000u, 0x00020011u, 0x00000001u, 0x0006000bu,
0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u, 0x0003000eu, 0x00000000u, 0x00000001u,
0x0008000fu, 0x00000000u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00000015u, 0x00000020u, 0x0000002bu,
0x00030003u, 0x00000002u, 0x000001c2u, 0x00040005u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00030005u,
0x00000009u, 0x00000070u, 0x00060005u, 0x00000015u, 0x565f6c67u, 0x65747265u, 0x646e4978u, 0x00007865u,
0x00050005u, 0x00000018u, 0x65646e69u, 0x6c626178u, 0x00000065u, 0x00060005u, 0x0000001eu, 0x505f6c67u,
0x65567265u, 0x78657472u, 0x00000000u, 0x00060006u, 0x0000001eu, 0x00000000u, 0x505f6c67u, 0x7469736fu,
0x006e6f69u, 0x00070006u, 0x0000001eu, 0x00000001u, 0x505f6c67u, 0x746e696fu, 0x657a6953u, 0x00000000u,
0x00070006u, 0x0000001eu, 0x00000002u, 0x435f6c67u, 0x4470696cu, 0x61747369u, 0x0065636eu, 0x00070006u,
0x0000001eu, 0x00000003u, 0x435f6c67u, 0x446c6c75u, 0x61747369u, 0x0065636eu, 0x00030005u, 0x00000020u,
0x00000000u, 0x00050005u, 0x0000002bu, 0x43786574u, 0x64726f6fu, 0x00000000u, 0x00040047u, 0x00000015u,
0x0000000bu, 0x0000002au, 0x00030047u, 0x0000001eu, 0x00000002u, 0x00050048u, 0x0000001eu, 0x00000000u,
0x0000000bu, 0x00000000u, 0x00050048u, 0x0000001eu, 0x00000001u, 0x0000000bu, 0x00000001u, 0x00050048u,
0x0000001eu, 0x00000002u, 0x0000000bu, 0x00000003u, 0x00050048u, 0x0000001eu, 0x00000003u, 0x0000000bu,
0x00000004u, 0x00040047u, 0x0000002bu, 0x0000001eu, 0x00000000u, 0x00020013u, 0x00000002u, 0x00030021u,
0x00000003u, 0x00000002u, 0x00030016u, 0x00000006u, 0x00000020u, 0x00040017u, 0x00000007u, 0x00000006u,
0x00000002u, 0x00040020u, 0x00000008u, 0x00000007u, 0x00000007u, 0x00040015u, 0x0000000au, 0x00000020u,
0x00000000u, 0x0004002bu, 0x0000000au, 0x0000000bu, 0x00000003u, 0x0004001cu, 0x0000000cu, 0x00000007u,
0x0000000bu, 0x0004002bu, 0x00000006u, 0x0000000du, 0xbf800000u, 0x0005002cu, 0x00000007u, 0x0000000eu,
0x0000000du, 0x0000000du, 0x0004002bu, 0x00000006u, 0x0000000fu, 0x40400000u, 0x0005002cu, 0x00000007u,
0x00000010u, 0x0000000fu, 0x0000000du, 0x0005002cu, 0x00000007u, 0x00000011u, 0x0000000du, 0x0000000fu,
0x0006002cu, 0x0000000cu, 0x00000012u, 0x0000000eu, 0x00000010u, 0x00000011u, 0x00040015u, 0x00000013u,
0x00000020u, 0x00000001u, 0x00040020u, 0x00000014u, 0x00000001u, 0x00000013u, 0x0004003bu, 0x00000014u,
0x00000015u, 0x00000001u, 0x00040020u, 0x00000017u, 0x00000007u, 0x0000000cu, 0x00040017u, 0x0000001bu,
0x00000006u, 0x00000004u, 0x0004002bu, 0x0000000au, 0x0000001cu, 0x00000001u, 0x0004001cu, 0x0000001du,
0x00000006u, 0x0000001cu, 0x0006001eu, 0x0000001eu, 0x0000001bu, 0x00000006u, 0x0000001du, 0x0000001du,
0x00040020u, 0x0000001fu, 0x00000003u, 0x0000001eu, 0x0004003bu, 0x0000001fu, 0x00000020u, 0x00000003u,
0x0004002bu, 0x00000013u, 0x00000021u, 0x00000000u, 0x0004002bu, 0x00000006u, 0x00000023u, 0x00000000u,
0x0004002bu, 0x00000006u, 0x00000024u, 0x3f800000u, 0x00040020u, 0x00000028u, 0x00000003u, 0x0000001bu,
0x00040020u, 0x0000002au, 0x00000003u, 0x00000007u, 0x0004003bu, 0x0000002au, 0x0000002bu, 0x00000003u,
0x0004002bu, 0x00000006u, 0x0000002fu, 0x3f000000u, 0x00050036u, 0x00000002u, 0x00000004u, 0x00000000u,
0x00000003u, 0x000200f8u, 0x00000005u, 0x0004003bu, 0x00000008u, 0x00000009u, 0x00000007u, 0x0004003bu,
0x00000017u, 0x00000018u, 0x00000007u, 0x0004003du, 0x00000013u, 0x00000016u, 0x00000015u, 0x0003003eu,
0x00000018u, 0x00000012u, 0x00050041u, 0x00000008u, 0x00000019u, 0x00000018u, 0x00000016u, 0x0004003du,
0x00000007u, 0x0000001au, 0x00000019u, 0x0003003eu, 0x00000009u, 0x0000001au, 0x0004003du, 0x00000007u,
0x00000022u, 0x00000009u, 0x00050051u, 0x00000006u, 0x00000025u, 0x00000022u, 0x00000000u, 0x00050051u,
0x00000006u, 0x00000026u, 0x00000022u, 0x00000001u, 0x00070050u, 0x0000001bu, 0x00000027u, 0x00000025u,
0x00000026u, 0x00000023u, 0x00000024u, 0x00050041u, 0x00000028u, 0x00000029u, 0x00000020u, 0x00000021u,
0x0003003eu, 0x00000029u, 0x00000027u, 0x0004003du, 0x00000007u, 0x0000002cu, 0x00000009u, 0x00050050u,
0x00000007u, 0x0000002du, 0x00000024u, 0x00000024u, 0x00050081u, 0x00000007u, 0x0000002eu, 0x0000002cu,
0x0000002du, 0x0005008eu, 0x00000007u, 0x00000030u, 0x0000002eu, 0x0000002fu, 0x0003003eu, 0x0000002bu,
0x00000030u, 0x000100fdu, 0x00010038u,
};
static constexpr Uint32 kWireYuvConvertFragmentSpirv[] = {
0x07230203u, 0x00010000u, 0x0008000bu, 0x0000001cu, 0x00000000u, 0x00020011u, 0x00000001u, 0x0006000bu,
0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u, 0x0003000eu, 0x00000000u, 0x00000001u,
0x0007000fu, 0x00000004u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00000009u, 0x00000011u, 0x00030010u,
0x00000004u, 0x00000007u, 0x00030003u, 0x00000002u, 0x000001c2u, 0x00040005u, 0x00000004u, 0x6e69616du,
0x00000000u, 0x00050005u, 0x00000009u, 0x4374756fu, 0x726f6c6fu, 0x00000000u, 0x00050005u, 0x0000000du,
0x49767579u, 0x6567616du, 0x00000000u, 0x00050005u, 0x00000011u, 0x43786574u, 0x64726f6fu, 0x00000000u,
0x00040047u, 0x00000009u, 0x0000001eu, 0x00000000u, 0x00040047u, 0x0000000du, 0x00000021u, 0x00000000u,
0x00040047u, 0x0000000du, 0x00000022u, 0x00000000u, 0x00040047u, 0x00000011u, 0x0000001eu, 0x00000000u,
0x00020013u, 0x00000002u, 0x00030021u, 0x00000003u, 0x00000002u, 0x00030016u, 0x00000006u, 0x00000020u,
0x00040017u, 0x00000007u, 0x00000006u, 0x00000004u, 0x00040020u, 0x00000008u, 0x00000003u, 0x00000007u,
0x0004003bu, 0x00000008u, 0x00000009u, 0x00000003u, 0x00090019u, 0x0000000au, 0x00000006u, 0x00000001u,
0x00000000u, 0x00000000u, 0x00000000u, 0x00000001u, 0x00000000u, 0x0003001bu, 0x0000000bu, 0x0000000au,
0x00040020u, 0x0000000cu, 0x00000000u, 0x0000000bu, 0x0004003bu, 0x0000000cu, 0x0000000du, 0x00000000u,
0x00040017u, 0x0000000fu, 0x00000006u, 0x00000002u, 0x00040020u, 0x00000010u, 0x00000001u, 0x0000000fu,
0x0004003bu, 0x00000010u, 0x00000011u, 0x00000001u, 0x0004002bu, 0x00000006u, 0x00000013u, 0x00000000u,
0x00040017u, 0x00000015u, 0x00000006u, 0x00000003u, 0x0004002bu, 0x00000006u, 0x00000017u, 0x3f800000u,
0x00050036u, 0x00000002u, 0x00000004u, 0x00000000u, 0x00000003u, 0x000200f8u, 0x00000005u, 0x0004003du,
0x0000000bu, 0x0000000eu, 0x0000000du, 0x0004003du, 0x0000000fu, 0x00000012u, 0x00000011u, 0x00070058u,
0x00000007u, 0x00000014u, 0x0000000eu, 0x00000012u, 0x00000002u, 0x00000013u, 0x0008004fu, 0x00000015u,
0x00000016u, 0x00000014u, 0x00000014u, 0x00000000u, 0x00000001u, 0x00000002u, 0x00050051u, 0x00000006u,
0x00000018u, 0x00000016u, 0x00000000u, 0x00050051u, 0x00000006u, 0x00000019u, 0x00000016u, 0x00000001u,
0x00050051u, 0x00000006u, 0x0000001au, 0x00000016u, 0x00000002u, 0x00070050u, 0x00000007u, 0x0000001bu,
0x00000018u, 0x00000019u, 0x0000001au, 0x00000017u, 0x0003003eu, 0x00000009u, 0x0000001bu, 0x000100fdu,
0x00010038u,
};
@@ -0,0 +1,515 @@
// Included by VulkanRenderer.cpp inside the backend namespace, after WireSharedImage.inc.
//
// YUV SHARED IMAGES (NV12, P010; MG_Remote/Server/SharedImageYuv.h). A YUV image is never a
// texture's image: the texture keeps its own RGBA8 level 0 (VkTextureManager::SyncWireTextureShape)
// and, at its first use in each frame, the image is CONVERTED into it here - the image's
// AHardwareBuffer imported with its external format, sampled through an immutable sampler that
// carries a VkSamplerYcbcrConversion built from the import's colour hints, one triangle over the
// target. Zero CPU copies for an image this server allocated; a foreign one (a video decoder's
// buffer) is first brought up to date by the registry's labelled CPU copy (RefreshForeign).
//
// Why converted rather than sampled in place: a Y'CbCr conversion is an immutable sampler, which
// lives in the descriptor set layout and therefore in every application pipeline that samples the
// texture. A conversion pass keeps those pipelines exactly what they are for an RGBA texture, at
// the cost of one small pass per new frame of the video.
#if MOBILEGL_BUILD_DISAGGREGATED
#include "WireYuvConvertSpirv.h"
struct VulkanRenderer::WireYuvResources {
PFN_vkCreateSamplerYcbcrConversion createConversion = nullptr;
PFN_vkDestroySamplerYcbcrConversion destroyConversion = nullptr;
VkShaderModule vertex = VK_NULL_HANDLE, fragment = VK_NULL_HANDLE;
// One per (buffer format, conversion, target format). Never dropped before shutdown: a
// device sees a handful of them.
struct Pipeline {
Uint64 externalFormat = 0;
VkFormat format = VK_FORMAT_UNDEFINED;
VkSamplerYcbcrModelConversion model = VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY;
VkSamplerYcbcrRange range = VK_SAMPLER_YCBCR_RANGE_ITU_FULL;
VkComponentMapping components{};
VkChromaLocation xOffset = VK_CHROMA_LOCATION_COSITED_EVEN;
VkChromaLocation yOffset = VK_CHROMA_LOCATION_COSITED_EVEN;
VkFilter chromaFilter = VK_FILTER_NEAREST;
VkFormat target = VK_FORMAT_UNDEFINED;
VkSamplerYcbcrConversion conversion = VK_NULL_HANDLE;
VkSampler sampler = VK_NULL_HANDLE;
VkDescriptorSetLayout descriptorLayout = VK_NULL_HANDLE;
VkPipelineLayout pipelineLayout = VK_NULL_HANDLE;
VkRenderPass renderPass = VK_NULL_HANDLE;
VkPipeline pipeline = VK_NULL_HANDLE;
};
Vector<Pipeline> pipelines;
};
namespace {
// One texture's import of its YUV buffer (TextureResource::yuvSource): the image on the
// buffer's memory and a view through the conversion of `pipeline`. Destroyed by the
// resource's (deferred) Reset, so never under a submission still reading it.
struct WireYuvSource {
VkDevice device = VK_NULL_HANDLE;
VkImage image = VK_NULL_HANDLE;
VkDeviceMemory memory = VK_NULL_HANDLE;
VkImageView view = VK_NULL_HANDLE;
SizeT pipeline = 0;
MG_Remote::Server::SharedImages::Yuv::Hints hints;
~WireYuvSource() {
if (device == VK_NULL_HANDLE) return;
if (view != VK_NULL_HANDLE) vkDestroyImageView(device, view, nullptr);
if (image != VK_NULL_HANDLE) vkDestroyImage(device, image, nullptr);
if (memory != VK_NULL_HANDLE) vkFreeMemory(device, memory, nullptr);
}
};
VkSamplerYcbcrModelConversion YcbcrModelFor(const MG_Remote::Server::SharedImages::Yuv::Hints& hints) {
namespace Yuv = MG_Remote::Server::SharedImages::Yuv;
switch (hints.EffectiveColorSpace()) {
case Yuv::kEglItuRec709:
return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709;
case Yuv::kEglItuRec2020:
return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020;
default:
return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601;
}
}
} // namespace
void VulkanRenderer::DestroyWireYuvResources() {
if (!m_wireYuvResources) return;
auto& cache = *m_wireYuvResources;
for (const auto& entry : cache.pipelines) {
vkDestroyPipeline(m_device, entry.pipeline, nullptr);
vkDestroyRenderPass(m_device, entry.renderPass, nullptr);
vkDestroyPipelineLayout(m_device, entry.pipelineLayout, nullptr);
vkDestroyDescriptorSetLayout(m_device, entry.descriptorLayout, nullptr);
vkDestroySampler(m_device, entry.sampler, nullptr);
if (cache.destroyConversion != nullptr) cache.destroyConversion(m_device, entry.conversion, nullptr);
}
vkDestroyShaderModule(m_device, cache.vertex, nullptr);
vkDestroyShaderModule(m_device, cache.fragment, nullptr);
delete m_wireYuvResources;
m_wireYuvResources = nullptr;
}
Bool VulkanRenderer::ConvertYuvSharedImage(VkTextureManager::TextureResource& resource) {
#if defined(__ANDROID__)
namespace SI = MG_Remote::Server::SharedImages;
if (resource.yuvConvertedFrame == m_yuvFrameSerial) return true;
const auto image = std::static_pointer_cast<const SI::Image>(resource.yuvOwner);
if (image == nullptr || image->Native == nullptr || resource.image == VK_NULL_HANDLE) return false;
if (!m_samplerYcbcrConversion) {
MGLOG_E_ONCE("Magma: YUV shared image %llu cannot be sampled: the device has no samplerYcbcrConversion",
static_cast<unsigned long long>(image->Id));
return false;
}
// Whatever happens below, this frame does not try again.
resource.yuvConvertedFrame = m_yuvFrameSerial;
// THE CPU-COPY FALLBACK, for a buffer this server did not allocate (SharedImageRegistry.h).
String why;
if (image->IsForeign() && !SI::RefreshForeign(*image, why)) {
MGLOG_E_ONCE("Magma: foreign YUV image %llu could not be refreshed: %s",
static_cast<unsigned long long>(image->Id), why.c_str());
}
const SI::Yuv::Hints hints = SI::GetYuvHints(*image);
if (!m_wireYuvResources) {
m_wireYuvResources = new WireYuvResources;
auto& cache = *m_wireYuvResources;
cache.createConversion = reinterpret_cast<PFN_vkCreateSamplerYcbcrConversion>(
vkGetDeviceProcAddr(m_device, "vkCreateSamplerYcbcrConversion"));
if (cache.createConversion == nullptr)
cache.createConversion = reinterpret_cast<PFN_vkCreateSamplerYcbcrConversion>(
vkGetDeviceProcAddr(m_device, "vkCreateSamplerYcbcrConversionKHR"));
cache.destroyConversion = reinterpret_cast<PFN_vkDestroySamplerYcbcrConversion>(
vkGetDeviceProcAddr(m_device, "vkDestroySamplerYcbcrConversion"));
if (cache.destroyConversion == nullptr)
cache.destroyConversion = reinterpret_cast<PFN_vkDestroySamplerYcbcrConversion>(
vkGetDeviceProcAddr(m_device, "vkDestroySamplerYcbcrConversionKHR"));
const auto shader = [&](const Uint32* code, SizeT bytes, VkShaderModule& module) {
VkShaderModuleCreateInfo info{VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO};
info.codeSize = bytes;
info.pCode = code;
VK_VERIFY(vkCreateShaderModule(m_device, &info, nullptr, &module));
};
shader(kWireYuvConvertVertexSpirv, sizeof(kWireYuvConvertVertexSpirv), cache.vertex);
shader(kWireYuvConvertFragmentSpirv, sizeof(kWireYuvConvertFragmentSpirv), cache.fragment);
}
auto& cache = *m_wireYuvResources;
if (cache.createConversion == nullptr || cache.destroyConversion == nullptr) {
MGLOG_E_ONCE("Magma: vkCreateSamplerYcbcrConversion is not available; YUV images cannot be sampled");
return false;
}
// ---- the texture's import of the buffer, and the pipeline its conversion needs ----
auto* source = static_cast<WireYuvSource*>(resource.yuvSource.get());
if (source != nullptr && !(source->hints == hints)) {
// Re-imported with other colour hints: the conversion (and so the view) changes.
auto retired = MakeUnique<VkTextureManager::TextureResource>();
retired->yuvSource = std::move(resource.yuvSource);
WireRetiredObjects objects{};
objects.textures.push_back(Move(retired));
RetireWireObjects(Move(objects));
source = nullptr;
}
if (source == nullptr) {
WireAhbImport ctx;
if (!GetWireAhbImport(ctx)) return false;
auto* ahb = static_cast<AHardwareBuffer*>(image->Native);
const auto getProperties =
reinterpret_cast<PFN_vkGetAndroidHardwareBufferPropertiesANDROID>(ctx.getAhbProperties);
VkAndroidHardwareBufferFormatPropertiesANDROID format{
VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_FORMAT_PROPERTIES_ANDROID};
VkAndroidHardwareBufferPropertiesANDROID properties{
VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_PROPERTIES_ANDROID};
properties.pNext = &format;
if (getProperties(m_device, ahb, &properties) != VK_SUCCESS) {
MGLOG_E_ONCE("Magma: YUV image %llu: vkGetAndroidHardwareBufferPropertiesANDROID failed",
static_cast<unsigned long long>(image->Id));
return false;
}
// The conversion: the import's hints where it gave them, else what the buffer suggests.
WireYuvResources::Pipeline key;
key.externalFormat = format.format == VK_FORMAT_UNDEFINED ? format.externalFormat : 0;
key.format = format.format;
key.model = hints.ColorSpace != 0 ? YcbcrModelFor(hints) : format.suggestedYcbcrModel;
key.range = hints.Range != 0 ? (hints.FullRange() ? VK_SAMPLER_YCBCR_RANGE_ITU_FULL
: VK_SAMPLER_YCBCR_RANGE_ITU_NARROW)
: format.suggestedYcbcrRange;
key.components = format.samplerYcbcrConversionComponents;
const auto chroma = [&format](Uint32 siting, VkChromaLocation suggested) {
const Bool midpoint = siting == SI::Yuv::kEglYuvChromaSiting0_5;
const Bool cosited = siting == SI::Yuv::kEglYuvChromaSiting0;
if (midpoint && (format.formatFeatures & VK_FORMAT_FEATURE_MIDPOINT_CHROMA_SAMPLES_BIT))
return VK_CHROMA_LOCATION_MIDPOINT;
if (cosited && (format.formatFeatures & VK_FORMAT_FEATURE_COSITED_CHROMA_SAMPLES_BIT))
return VK_CHROMA_LOCATION_COSITED_EVEN;
return suggested;
};
key.xOffset = chroma(hints.SitingX, format.suggestedXChromaOffset);
key.yOffset = chroma(hints.SitingY, format.suggestedYChromaOffset);
key.chromaFilter =
(format.formatFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_YCBCR_CONVERSION_LINEAR_FILTER_BIT) != 0
? VK_FILTER_LINEAR
: VK_FILTER_NEAREST;
key.target = resource.format;
SizeT index = cache.pipelines.size();
for (SizeT i = 0; i < cache.pipelines.size(); ++i) {
const auto& p = cache.pipelines[i];
if (p.externalFormat == key.externalFormat && p.format == key.format && p.model == key.model &&
p.range == key.range && p.components.r == key.components.r &&
p.components.g == key.components.g && p.components.b == key.components.b &&
p.components.a == key.components.a && p.xOffset == key.xOffset && p.yOffset == key.yOffset &&
p.chromaFilter == key.chromaFilter && p.target == key.target) {
index = i;
break;
}
}
if (index == cache.pipelines.size()) {
VkExternalFormatANDROID external{VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID};
external.externalFormat = key.externalFormat;
VkSamplerYcbcrConversionCreateInfo conversion{VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO};
conversion.pNext = key.externalFormat != 0 ? &external : nullptr;
conversion.format = key.format;
conversion.ycbcrModel = key.model;
conversion.ycbcrRange = key.range;
conversion.components = key.components;
conversion.xChromaOffset = key.xOffset;
conversion.yChromaOffset = key.yOffset;
conversion.chromaFilter = key.chromaFilter;
conversion.forceExplicitReconstruction = VK_FALSE;
if (cache.createConversion(m_device, &conversion, nullptr, &key.conversion) != VK_SUCCESS) {
MGLOG_E_ONCE("Magma: YUV image %llu: vkCreateSamplerYcbcrConversion failed",
static_cast<unsigned long long>(image->Id));
return false;
}
VkSamplerYcbcrConversionInfo conversionInfo{VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO};
conversionInfo.conversion = key.conversion;
VkSamplerCreateInfo sampler{VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO};
sampler.pNext = &conversionInfo;
// Without separate reconstruction filters the sampler filters as the conversion does.
sampler.magFilter = sampler.minFilter = key.chromaFilter;
sampler.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST;
sampler.addressModeU = sampler.addressModeV = sampler.addressModeW =
VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sampler.maxLod = 0.0f;
VK_VERIFY(vkCreateSampler(m_device, &sampler, nullptr, &key.sampler));
const VkDescriptorSetLayoutBinding binding{0, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1,
VK_SHADER_STAGE_FRAGMENT_BIT, &key.sampler};
VkDescriptorSetLayoutCreateInfo layout{VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO};
layout.bindingCount = 1;
layout.pBindings = &binding;
VK_VERIFY(vkCreateDescriptorSetLayout(m_device, &layout, nullptr, &key.descriptorLayout));
VkPipelineLayoutCreateInfo pipelineLayout{VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO};
pipelineLayout.setLayoutCount = 1;
pipelineLayout.pSetLayouts = &key.descriptorLayout;
VK_VERIFY(vkCreatePipelineLayout(m_device, &pipelineLayout, nullptr, &key.pipelineLayout));
// Every texel of the target is written: nothing to load.
VkAttachmentDescription attachment{};
attachment.format = key.target;
attachment.samples = VK_SAMPLE_COUNT_1_BIT;
attachment.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
attachment.initialLayout = attachment.finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
const VkAttachmentReference color{0, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL};
VkSubpassDescription subpass{};
subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
subpass.colorAttachmentCount = 1;
subpass.pColorAttachments = &color;
VkRenderPassCreateInfo pass{VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO};
pass.attachmentCount = 1;
pass.pAttachments = &attachment;
pass.subpassCount = 1;
pass.pSubpasses = &subpass;
VK_VERIFY(vkCreateRenderPass(m_device, &pass, nullptr, &key.renderPass));
VkPipelineShaderStageCreateInfo stages[2]{};
for (auto& stage : stages) {
stage.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stage.pName = "main";
}
stages[0].stage = VK_SHADER_STAGE_VERTEX_BIT;
stages[0].module = cache.vertex;
stages[1].stage = VK_SHADER_STAGE_FRAGMENT_BIT;
stages[1].module = cache.fragment;
const VkPipelineVertexInputStateCreateInfo vertex{
VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO};
VkPipelineInputAssemblyStateCreateInfo assembly{
VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO};
assembly.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
VkPipelineViewportStateCreateInfo viewport{VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO};
viewport.viewportCount = viewport.scissorCount = 1;
VkPipelineRasterizationStateCreateInfo raster{
VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO};
raster.polygonMode = VK_POLYGON_MODE_FILL;
raster.cullMode = VK_CULL_MODE_NONE;
raster.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE;
raster.lineWidth = 1.0f;
VkPipelineMultisampleStateCreateInfo samples{
VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO};
samples.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;
VkPipelineColorBlendAttachmentState colorMask{};
colorMask.colorWriteMask = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT |
VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT;
VkPipelineColorBlendStateCreateInfo blend{VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO};
blend.attachmentCount = 1;
blend.pAttachments = &colorMask;
const VkDynamicState dynamicStates[]{VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR};
VkPipelineDynamicStateCreateInfo dynamic{VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO};
dynamic.dynamicStateCount = 2;
dynamic.pDynamicStates = dynamicStates;
VkGraphicsPipelineCreateInfo pipeline{VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO};
pipeline.stageCount = 2;
pipeline.pStages = stages;
pipeline.pVertexInputState = &vertex;
pipeline.pInputAssemblyState = &assembly;
pipeline.pViewportState = &viewport;
pipeline.pRasterizationState = &raster;
pipeline.pMultisampleState = &samples;
pipeline.pColorBlendState = &blend;
pipeline.pDynamicState = &dynamic;
pipeline.layout = key.pipelineLayout;
pipeline.renderPass = key.renderPass;
VK_VERIFY(vkCreateGraphicsPipelines(m_device, VK_NULL_HANDLE, 1, &pipeline, nullptr, &key.pipeline));
cache.pipelines.push_back(key);
MGLOG_I("Magma: YUV conversion pipeline %zu: %s format %llu, model %d, range %d, chroma %d/%d, "
"filter %d",
index, key.externalFormat != 0 ? "external" : "Vulkan",
static_cast<unsigned long long>(key.externalFormat != 0 ? key.externalFormat
: static_cast<Uint64>(key.format)),
static_cast<Int>(key.model), static_cast<Int>(key.range), static_cast<Int>(key.xOffset),
static_cast<Int>(key.yOffset), static_cast<Int>(key.chromaFilter));
}
const auto& chosen = cache.pipelines[index];
// The buffer, imported with its external format, sampled only.
auto created = std::make_shared<WireYuvSource>();
created->device = m_device;
created->pipeline = index;
created->hints = hints;
VkExternalFormatANDROID external{VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID};
external.externalFormat = chosen.externalFormat;
VkExternalMemoryImageCreateInfo externalMemory{VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO};
externalMemory.pNext = chosen.externalFormat != 0 ? &external : nullptr;
externalMemory.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID;
VkImageCreateInfo imageInfo{VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO};
imageInfo.pNext = &externalMemory;
imageInfo.imageType = VK_IMAGE_TYPE_2D;
imageInfo.format = chosen.format;
imageInfo.extent = {image->Width, image->Height, 1};
imageInfo.mipLevels = 1;
imageInfo.arrayLayers = 1;
imageInfo.samples = VK_SAMPLE_COUNT_1_BIT;
imageInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
imageInfo.usage = VK_IMAGE_USAGE_SAMPLED_BIT;
imageInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
imageInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
if (vkCreateImage(m_device, &imageInfo, nullptr, &created->image) != VK_SUCCESS) {
created->image = VK_NULL_HANDLE;
MGLOG_E_ONCE("Magma: YUV image %llu: vkCreateImage(external format) failed",
static_cast<unsigned long long>(image->Id));
return false;
}
Int32 type = -1;
for (Uint32 i = 0; i < ctx.memory.memoryTypeCount && type < 0; ++i) {
if ((properties.memoryTypeBits & (1u << i)) != 0) type = static_cast<Int32>(i);
}
VkMemoryDedicatedAllocateInfo dedicated{VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO};
dedicated.image = created->image;
VkImportAndroidHardwareBufferInfoANDROID import{
VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID};
import.pNext = &dedicated;
import.buffer = ahb;
VkMemoryAllocateInfo allocate{VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO};
allocate.pNext = &import;
allocate.allocationSize = properties.allocationSize;
allocate.memoryTypeIndex = static_cast<Uint32>(type < 0 ? 0 : type);
if (type < 0 || vkAllocateMemory(m_device, &allocate, nullptr, &created->memory) != VK_SUCCESS ||
vkBindImageMemory(m_device, created->image, created->memory, 0) != VK_SUCCESS) {
MGLOG_E_ONCE("Magma: YUV image %llu: its buffer could not be imported",
static_cast<unsigned long long>(image->Id));
return false;
}
VkSamplerYcbcrConversionInfo conversionInfo{VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO};
conversionInfo.conversion = chosen.conversion;
VkImageViewCreateInfo view{VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO};
view.pNext = &conversionInfo;
view.image = created->image;
view.viewType = VK_IMAGE_VIEW_TYPE_2D;
view.format = chosen.format;
view.components = {VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY,
VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY};
view.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
if (vkCreateImageView(m_device, &view, nullptr, &created->view) != VK_SUCCESS) {
created->view = VK_NULL_HANDLE;
MGLOG_E_ONCE("Magma: YUV image %llu: its view could not be created",
static_cast<unsigned long long>(image->Id));
return false;
}
MGLOG_I("Magma: YUV image %llu (%ux%u%s) sampled through conversion pipeline %zu",
static_cast<unsigned long long>(image->Id), image->Width, image->Height,
image->IsForeign() ? ", foreign: CPU-copied" : ", zero-copy", index);
resource.yuvSource = created;
source = created.get();
}
const auto& chosen = cache.pipelines[source->pipeline];
// ---- the pass ----
auto& frame = m_frameContext.GetCurrent();
if (!frame.isCommandRecording) m_frameContext.BeginCommandRecording();
if (VkRenderPassManager::GetActiveRenderPass()) VkRenderPassManager::EndRenderPass(frame.commandBuffer);
m_textureManager->FlushPendingUploads();
const Uint32 family = static_cast<Uint32>(m_physicalDevice.queueFamilies.graphicsFamily);
// The buffer's content is its producer's (the CPU's copy, or the allocator's writer): it is
// acquired from the foreign family for the read and handed back after it.
VkImageMemoryBarrier acquire{VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER};
acquire.srcAccessMask = 0;
acquire.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
acquire.oldLayout = VK_IMAGE_LAYOUT_GENERAL;
acquire.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
acquire.srcQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT;
acquire.dstQueueFamilyIndex = family;
acquire.image = source->image;
acquire.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdPipelineBarrier(frame.commandBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &acquire);
WireImage target{};
target.image = resource.image;
target.format = resource.format;
target.viewType = VK_IMAGE_VIEW_TYPE_2D;
target.extent = resource.extent;
target.layout = resource.layout;
target.trackedLayout = &resource.layout;
target.aspect = VK_IMAGE_ASPECT_COLOR_BIT;
target.usageFlags = resource.usageFlags;
target.levels = resource.mipLevels;
TransitionWireImage(target, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
VkImageViewCreateInfo targetViewInfo{VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO};
targetViewInfo.image = resource.image;
targetViewInfo.viewType = VK_IMAGE_VIEW_TYPE_2D;
targetViewInfo.format = resource.format;
targetViewInfo.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
VkImageView targetView = VK_NULL_HANDLE;
VK_VERIFY(vkCreateImageView(m_device, &targetViewInfo, nullptr, &targetView));
VkFramebufferCreateInfo framebufferInfo{VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO};
framebufferInfo.renderPass = chosen.renderPass;
framebufferInfo.attachmentCount = 1;
framebufferInfo.pAttachments = &targetView;
framebufferInfo.width = resource.extent.width;
framebufferInfo.height = resource.extent.height;
framebufferInfo.layers = 1;
VkFramebuffer framebuffer = VK_NULL_HANDLE;
VK_VERIFY(vkCreateFramebuffer(m_device, &framebufferInfo, nullptr, &framebuffer));
// A multi-planar image may take more than one descriptor of its binding.
VkDescriptorPool descriptorPool = VK_NULL_HANDLE;
VkDescriptorSet descriptorSet = VK_NULL_HANDLE;
const VkDescriptorPoolSize poolSize{VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 3};
VkDescriptorPoolCreateInfo pool{VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO};
pool.maxSets = 1;
pool.poolSizeCount = 1;
pool.pPoolSizes = &poolSize;
VK_VERIFY(vkCreateDescriptorPool(m_device, &pool, nullptr, &descriptorPool));
VkDescriptorSetAllocateInfo alloc{VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO};
alloc.descriptorPool = descriptorPool;
alloc.descriptorSetCount = 1;
alloc.pSetLayouts = &chosen.descriptorLayout;
VK_VERIFY(vkAllocateDescriptorSets(m_device, &alloc, &descriptorSet));
const VkDescriptorImageInfo sampled{chosen.sampler, source->view, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL};
VkWriteDescriptorSet write{VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET};
write.dstSet = descriptorSet;
write.dstBinding = 0;
write.descriptorCount = 1;
write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
write.pImageInfo = &sampled;
vkUpdateDescriptorSets(m_device, 1, &write, 0, nullptr);
VkRenderPassBeginInfo begin{VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO};
begin.renderPass = chosen.renderPass;
begin.framebuffer = framebuffer;
begin.renderArea = {{0, 0}, resource.extent};
vkCmdBeginRenderPass(frame.commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, chosen.pipeline);
const VkViewport viewport{0.0f, 0.0f, static_cast<float>(resource.extent.width),
static_cast<float>(resource.extent.height), 0.0f, 1.0f};
const VkRect2D scissor{{0, 0}, resource.extent};
vkCmdSetViewport(frame.commandBuffer, 0, 1, &viewport);
vkCmdSetScissor(frame.commandBuffer, 0, 1, &scissor);
vkCmdBindDescriptorSets(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, chosen.pipelineLayout, 0, 1,
&descriptorSet, 0, nullptr);
vkCmdDraw(frame.commandBuffer, 3, 1, 0, 0);
vkCmdEndRenderPass(frame.commandBuffer);
VkImageMemoryBarrier release = acquire;
release.srcAccessMask = VK_ACCESS_SHADER_READ_BIT;
release.dstAccessMask = 0;
release.oldLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
release.newLayout = VK_IMAGE_LAYOUT_GENERAL;
release.srcQueueFamilyIndex = family;
release.dstQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT;
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdPipelineBarrier(frame.commandBuffer, VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, 0, 0, nullptr, 0, nullptr, 1, &release);
// GENERAL, the layout every wire descriptor samples in, with the pass's writes made visible.
target.layout = resource.layout;
TransitionWireImage(target, VK_IMAGE_LAYOUT_GENERAL);
ResetDynamicStateShadow();
InvalidatePipelineMemo();
if (m_uniformManager) m_uniformManager->OnCommandBufferBoundary();
m_textureManager->StampResourceRecordingUse(resource);
m_textureManager->StampResourceRecordingWrite(resource);
WireRetiredObjects retired{};
retired.framebuffers.push_back(framebuffer);
retired.imageViews = {targetView};
retired.descriptorPools.push_back(descriptorPool);
RetireWireObjects(Move(retired));
// The frame's fence orders the next refresh of a foreign buffer after this read.
m_sharedImageReads.NoteRead(image);
return true;
#else
(void)resource;
return false;
#endif
}
#endif // MOBILEGL_BUILD_DISAGGREGATED
+114 -17
View File
@@ -17,9 +17,14 @@
// process's MobileGL client session. Whatever node the caller holds is therefore equally good,
// and none is vendor-specific.
//
// THE LAYOUT IS THE ALLOCATOR'S. Buffers report DRM_FORMAT_MOD_INVALID and a single plane; a
// request that needs a known layout (LINEAR, or a modifier list without INVALID) is refused
// rather than answered with a buffer whose layout is not what was asked for.
// THE LAYOUT IS THE ALLOCATOR'S. Buffers report DRM_FORMAT_MOD_INVALID and a single plane (two for
// the YUV formats, both in the one descriptor); a request that needs a known layout (LINEAR, or a
// modifier list without INVALID) is refused rather than answered with a buffer whose layout is not
// what was asked for.
//
// YUV (NV12, P010) buffers exist when the server's platform allocates them (the shared-image ABI's
// format list says so). They are sampled, never rendered to: a GBM_BO_USE_RENDERING request for one
// is refused.
//
// Buffer operations reach the server through libMobileGL.so's shared-image C ABI
// (MG_Impl/SharedImageApi.h), resolved with dlsym on first use - preferably from the instance the
@@ -40,6 +45,7 @@
#include <dlfcn.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <unistd.h>
#define MOBILEGL_GBM_EXPORT extern "C" __attribute__((visibility("default")))
@@ -65,6 +71,10 @@ namespace {
int fd; // owned; every get_fd hands out a duplicate
uint32_t offset;
uint64_t modifier;
// A YUV buffer's plane 1, in the same descriptor.
uint32_t planes;
uint32_t plane1Stride;
uint32_t plane1Offset;
};
// ---- libMobileGL's shared-image ABI -----------------------------------------------------
@@ -74,6 +84,9 @@ namespace {
PFN_mobilegl_shared_image_import import = nullptr;
PFN_mobilegl_shared_image_release release = nullptr;
PFN_mobilegl_server_available serverAvailable = nullptr; // optional (older libraries lack it)
// ABI 2 (YUV); null on an older library.
PFN_mobilegl_shared_image_formats formats = nullptr;
PFN_mobilegl_shared_image_import_planes importPlanes = nullptr;
};
// The library the process already has (by its soname: glvnd loads it by path, which records
@@ -107,6 +120,12 @@ namespace {
std::fprintf(stderr, "mobilegl_gbm: this libMobileGL.so has no shared-image ABI\n");
return;
}
if (version() >= 2) {
api.formats = reinterpret_cast<PFN_mobilegl_shared_image_formats>(
dlsym(lib, "mobilegl_shared_image_formats"));
api.importPlanes = reinterpret_cast<PFN_mobilegl_shared_image_import_planes>(
dlsym(lib, "mobilegl_shared_image_import_planes"));
}
resolved = true;
});
return resolved ? &api : nullptr;
@@ -125,11 +144,41 @@ namespace {
// What a shared image can be. The ARGB/XRGB orders are stored like the ABGR/XBGR ones: the
// layout is the server's (modifier INVALID, no map is offered), so no reader can tell, and
// every GL that touches the image addresses its channels logically.
bool FormatIsYuv(uint32_t format) {
return format == MOBILEGL_SHARED_IMAGE_FOURCC_NV12 || format == MOBILEGL_SHARED_IMAGE_FOURCC_P010;
}
// Whether the server holds YUV images: the library's format list, asked once (it takes the
// process's session, which the first buffer would anyway).
bool YuvFormatAvailable(uint32_t format) {
static std::once_flag once;
static bool nv12 = false, p010 = false;
std::call_once(once, [] {
const SharedImageApi* api = Api();
if (api == nullptr || api->formats == nullptr || api->importPlanes == nullptr) return;
uint32_t formats[16] = {};
const int total = api->formats(formats, 16);
for (int i = 0; i < total && i < 16; ++i) {
nv12 |= formats[i] == MOBILEGL_SHARED_IMAGE_FOURCC_NV12;
p010 |= formats[i] == MOBILEGL_SHARED_IMAGE_FOURCC_P010;
}
});
return format == MOBILEGL_SHARED_IMAGE_FOURCC_NV12 ? nv12 : format == MOBILEGL_SHARED_IMAGE_FOURCC_P010 && p010;
}
bool FormatSupported(uint32_t format) {
if (FormatIsYuv(format)) return YuvFormatAvailable(format);
return format == MOBILEGL_SHARED_IMAGE_FOURCC_ABGR8888 || format == MOBILEGL_SHARED_IMAGE_FOURCC_XBGR8888 ||
format == MOBILEGL_SHARED_IMAGE_FOURCC_ARGB8888 || format == MOBILEGL_SHARED_IMAGE_FOURCC_XRGB8888;
}
uint32_t PlaneCount(uint32_t format) { return FormatIsYuv(format) ? 2 : 1; }
// A YUV buffer is sampled only.
uint32_t RefusedUsage(uint32_t format) {
return kRefusedUsage | (FormatIsYuv(format) ? static_cast<uint32_t>(GBM_BO_USE_RENDERING) : 0u);
}
bool ModifierAcceptable(uint64_t modifier) { return modifier == kModifierInvalid || modifier == kModifierLinear; }
// There is no GEM handle behind a shared image; callers that only use the handle as a key
@@ -142,7 +191,8 @@ namespace {
}
gbm_bo* NewBo(gbm_device* gbm, uint64_t imageId, int fd, uint32_t width, uint32_t height, uint32_t format,
uint32_t stride, uint32_t offset, uint64_t modifier) {
uint32_t stride, uint32_t offset, uint64_t modifier, uint32_t plane1Stride = 0,
uint32_t plane1Offset = 0) {
auto* bo = new (std::nothrow) Bo{};
if (bo == nullptr) {
errno = ENOMEM;
@@ -158,6 +208,9 @@ namespace {
bo->fd = fd;
bo->offset = offset;
bo->modifier = modifier;
bo->planes = PlaneCount(format);
bo->plane1Stride = plane1Stride;
bo->plane1Offset = plane1Offset;
return &bo->base;
}
@@ -168,18 +221,20 @@ namespace {
void DeviceDestroy(gbm_device* gbm) { delete reinterpret_cast<Device*>(gbm); }
int IsFormatSupported(gbm_device*, uint32_t format, uint32_t usage) {
return FormatSupported(Canonical(format)) && (usage & kRefusedUsage) == 0 ? 1 : 0;
format = Canonical(format);
return FormatSupported(format) && (usage & RefusedUsage(format)) == 0 ? 1 : 0;
}
int GetFormatModifierPlaneCount(gbm_device*, uint32_t format, uint64_t modifier) {
if (!FormatSupported(Canonical(format)) || !ModifierAcceptable(modifier)) return -1;
return 1;
format = Canonical(format);
if (!FormatSupported(format) || !ModifierAcceptable(modifier)) return -1;
return static_cast<int>(PlaneCount(format));
}
gbm_bo* BoCreate(gbm_device* gbm, uint32_t width, uint32_t height, uint32_t format, uint32_t usage,
const uint64_t* modifiers, const unsigned int count) {
format = Canonical(format);
if (!FormatSupported(format) || (usage & kRefusedUsage) != 0) {
if (!FormatSupported(format) || (usage & RefusedUsage(format)) != 0) {
errno = EINVAL;
return nullptr;
}
@@ -207,7 +262,7 @@ namespace {
return nullptr;
}
gbm_bo* bo = NewBo(gbm, image.id, image.fd, image.width, image.height, image.fourcc, image.stride,
image.offset, image.modifier);
image.offset, image.modifier, image.plane1_stride, image.plane1_offset);
if (bo == nullptr) {
api->release(image.id);
::close(image.fd);
@@ -215,6 +270,13 @@ namespace {
return bo;
}
// Whether two descriptors name one dma-buf: a YUV buffer's planes must be in one.
bool SameBuffer(int a, int b) {
if (a == b) return true;
struct stat sa {}, sb {};
return ::fstat(a, &sa) == 0 && ::fstat(b, &sb) == 0 && sa.st_dev == sb.st_dev && sa.st_ino == sb.st_ino;
}
gbm_bo* BoImport(gbm_device* gbm, uint32_t type, void* buffer, uint32_t usage) {
(void)usage;
if (buffer == nullptr) {
@@ -223,6 +285,7 @@ namespace {
}
int fd = -1;
uint32_t width = 0, height = 0, format = 0, stride = 0, offset = 0;
uint32_t plane1Stride = 0, plane1Offset = 0;
uint64_t modifier = kModifierInvalid;
switch (type) {
case GBM_BO_IMPORT_FD: {
@@ -232,11 +295,19 @@ namespace {
height = data->height;
format = data->format;
stride = data->stride;
// One fd and one stride cannot say where a second plane is.
if (FormatIsYuv(Canonical(format))) {
errno = EINVAL;
return nullptr;
}
break;
}
case GBM_BO_IMPORT_FD_MODIFIER: {
const auto* data = static_cast<const gbm_import_fd_modifier_data*>(buffer);
if (data->num_fds != 1 || !ModifierAcceptable(data->modifier)) {
const bool yuv = FormatIsYuv(Canonical(data->format));
const unsigned int planes = yuv ? 2u : 1u;
if (data->num_fds < 1 || data->num_fds > planes || !ModifierAcceptable(data->modifier) ||
(data->num_fds == 2 && !SameBuffer(data->fds[0], data->fds[1]))) {
errno = EINVAL;
return nullptr;
}
@@ -246,6 +317,10 @@ namespace {
format = data->format;
stride = static_cast<uint32_t>(data->strides[0]);
offset = static_cast<uint32_t>(data->offsets[0]);
if (yuv) {
plane1Stride = static_cast<uint32_t>(data->strides[1]);
plane1Offset = static_cast<uint32_t>(data->offsets[1]);
}
modifier = data->modifier;
break;
}
@@ -269,13 +344,26 @@ namespace {
const int owned = ::fcntl(fd, F_DUPFD_CLOEXEC, 0);
if (owned < 0) return nullptr;
uint64_t id = 0;
const int rc = api->import(owned, width, height, format, &id);
int rc = 0;
if (FormatIsYuv(format)) {
mobilegl_shared_image_planes planes{};
planes.struct_size = sizeof(planes);
planes.plane_count = 2;
planes.offset[0] = offset;
planes.offset[1] = plane1Offset;
planes.pitch[0] = stride;
planes.pitch[1] = plane1Stride;
planes.modifier = modifier;
rc = api->importPlanes(owned, width, height, format, &planes, &id);
} else {
rc = api->import(owned, width, height, format, &id);
}
if (rc != 0) {
::close(owned);
errno = -rc;
return nullptr;
}
gbm_bo* bo = NewBo(gbm, id, owned, width, height, format, stride, offset, modifier);
gbm_bo* bo = NewBo(gbm, id, owned, width, height, format, stride, offset, modifier, plane1Stride, plane1Offset);
if (bo == nullptr) {
api->release(id);
::close(owned);
@@ -299,11 +387,14 @@ namespace {
int BoGetFd(gbm_bo* bo) { return ::fcntl(AsBo(bo)->fd, F_DUPFD_CLOEXEC, 0); }
int BoGetPlanes(gbm_bo*) { return 1; }
int BoGetPlanes(gbm_bo* bo) { return static_cast<int>(AsBo(bo)->planes); }
bool ValidPlane(gbm_bo* bo, int plane) { return plane >= 0 && static_cast<uint32_t>(plane) < AsBo(bo)->planes; }
// Every plane is in the one buffer, so every plane has its handle.
gbm_bo_handle BoGetHandle(gbm_bo* bo, int plane) {
gbm_bo_handle handle{};
if (plane != 0) {
if (!ValidPlane(bo, plane)) {
errno = EINVAL;
handle.s32 = -1;
return handle;
@@ -312,16 +403,22 @@ namespace {
}
int BoGetPlaneFd(gbm_bo* bo, int plane) {
if (plane != 0) {
if (!ValidPlane(bo, plane)) {
errno = EINVAL;
return -1;
}
return BoGetFd(bo);
}
uint32_t BoGetStride(gbm_bo* bo, int plane) { return plane == 0 ? bo->v0.stride : 0; }
uint32_t BoGetStride(gbm_bo* bo, int plane) {
if (plane == 0) return bo->v0.stride;
return plane == 1 && AsBo(bo)->planes == 2 ? AsBo(bo)->plane1Stride : 0;
}
uint32_t BoGetOffset(gbm_bo* bo, int plane) { return plane == 0 ? AsBo(bo)->offset : 0; }
uint32_t BoGetOffset(gbm_bo* bo, int plane) {
if (plane == 0) return AsBo(bo)->offset;
return plane == 1 && AsBo(bo)->planes == 2 ? AsBo(bo)->plane1Offset : 0;
}
uint64_t BoGetModifier(gbm_bo* bo) { return AsBo(bo)->modifier; }
+73 -6
View File
@@ -38,6 +38,10 @@
#include <cstring>
#include <mutex>
#include <sstream>
#if !defined(_WIN32)
#include <sys/stat.h>
#include <unistd.h>
#endif
#if defined(__linux__) && !defined(__ANDROID__)
#include <dlfcn.h>
#endif
@@ -1608,6 +1612,38 @@ namespace MobileGL::MG_Impl::EGLImpl {
#endif
}
// Whether the server can hold YUV images (its platform allocates the formats): asked once, by
// allocating and dropping a small NV12 image, and only when a YUV question first comes up.
Bool SharedImageYuvAvailable() {
if (!SharedImagesAvailable()) return false;
static std::atomic<Int> answer{-1};
const Int known = answer.load(std::memory_order_acquire);
if (known >= 0) return known != 0;
auto* backendObject = MG_Backend::pActiveBackendObject.get();
if (backendObject == nullptr) return false;
MG_Backend::SharedImageExport probe{};
const Bool allocated = backendObject->AllocateSharedImage(
16, 16, MG_State::EGLState::EGLContext::kDrmFourccNv12, &probe);
#if !defined(_WIN32)
if (probe.Fd >= 0) ::close(probe.Fd);
#endif
if (allocated && probe.Id != 0) (void)backendObject->ReleaseSharedImage(probe.Id);
answer.store(allocated ? 1 : 0, std::memory_order_release);
MGLOG_I("Shared images: the server %s YUV (NV12) images", allocated ? "holds" : "has no");
return allocated;
}
// Whether two descriptors name one dma-buf (a YUV import's planes must be in one buffer).
Bool SameDmaBuf(int a, int b) {
if (a == b) return true;
#if defined(_WIN32)
return false;
#else
struct stat sa {}, sb {};
return ::fstat(a, &sa) == 0 && ::fstat(b, &sb) == 0 && sa.st_dev == sb.st_dev && sa.st_ino == sb.st_ino;
#endif
}
Bool LookupSharedImage(EGLImage image, Uint64* id, EGLint* width, EGLint* height) {
auto* state = GetState();
MG_State::EGLState::EGLContext::SharedImageInfo info;
@@ -1673,10 +1709,41 @@ namespace MobileGL::MG_Impl::EGLImpl {
}
auto* backendObject = GetBackendObject(state);
Uint64 sharedImageId = 0;
if (!backendObject ||
!backendObject->ImportSharedImage(request.Fd, static_cast<Uint32>(request.Width),
static_cast<Uint32>(request.Height), request.Fourcc, &sharedImageId) ||
sharedImageId == 0) {
Bool imported = false;
if (MG_State::EGLState::EGLContext::IsDmaBufFourccYuv(request.Fourcc)) {
// YUV: both planes in one buffer (what video decoders and this server's allocator hand
// out), sampled through GL_OES_EGL_image_external. Planes in two buffers are refused.
if (!SharedImageYuvAvailable() || !SameDmaBuf(request.Fd, request.Plane1Fd)) {
MGLOG_I_ONCE("eglCreateImage: a %dx%d YUV dma-buf cannot be imported (%s); refused (EGL_BAD_MATCH)",
request.Width, request.Height,
SharedImageYuvAvailable() ? "its planes are in two buffers"
: "the server has no YUV images");
state->SetError(EGL_BAD_MATCH);
return EGL_NO_IMAGE;
}
MG_Backend::SharedImageImportLayout layout;
layout.PlaneCount = 2;
layout.Offset[0] = static_cast<Uint32>(request.Offset);
layout.Offset[1] = static_cast<Uint32>(request.Plane1Offset);
layout.Pitch[0] = static_cast<Uint32>(request.Pitch);
layout.Pitch[1] = static_cast<Uint32>(request.Plane1Pitch);
layout.ColorSpace = request.ColorSpace;
layout.Range = request.Range;
layout.SitingX = request.SitingX;
layout.SitingY = request.SitingY;
layout.HasModifier = request.HasModifier;
layout.Modifier = request.Modifier;
imported = backendObject != nullptr &&
backendObject->ImportSharedImagePlanes(request.Fd, static_cast<Uint32>(request.Width),
static_cast<Uint32>(request.Height), request.Fourcc,
layout, &sharedImageId);
} else {
imported = backendObject != nullptr &&
backendObject->ImportSharedImage(request.Fd, static_cast<Uint32>(request.Width),
static_cast<Uint32>(request.Height), request.Fourcc,
&sharedImageId);
}
if (!imported || sharedImageId == 0) {
MGLOG_I_ONCE("eglCreateImage: a %dx%d dma-buf is not one of the server's shared images; refused "
"(EGL_BAD_MATCH)",
request.Width, request.Height);
@@ -1731,7 +1798,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
if (!state) {
return EGL_FALSE;
}
return state->QueryDmaBufFormats(dpy, SharedImagesAvailable(), max_formats, formats, num_formats)
return state->QueryDmaBufFormats(dpy, SharedImagesAvailable(), SharedImageYuvAvailable(), max_formats, formats, num_formats)
? EGL_TRUE
: EGL_FALSE;
}
@@ -1743,7 +1810,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
if (!state) {
return EGL_FALSE;
}
return state->QueryDmaBufModifiers(dpy, SharedImagesAvailable(), format, max_modifiers,
return state->QueryDmaBufModifiers(dpy, SharedImagesAvailable(), SharedImageYuvAvailable(), format, max_modifiers,
reinterpret_cast<Uint64*>(modifiers), external_only, num_modifiers)
? EGL_TRUE
: EGL_FALSE;
+3
View File
@@ -68,6 +68,9 @@ namespace MobileGL::MG_Impl::EGLImpl {
// a split client's does, a monolith's does not. Decides the dma-buf EGL extensions, the
// Wayland linux-dmabuf presentation and GL_OES_EGL_image.
Bool SharedImagesAvailable();
// Whether the server also holds YUV (NV12/P010) images: decides the YUV dma-buf formats and
// GL_OES_EGL_image_external. One round trip the first time it is asked, cached after.
Bool SharedImageYuvAvailable();
// Under a dispatching libEGL the dispatcher, not this library, owns eglBindAPI's state: it does
// not forward a bind of the API it already reports (EGL_OPENGL_ES_API, EGL's default), so a
// context would otherwise be created for whatever this library assumed. The glvnd vendor
+44 -16
View File
@@ -434,6 +434,7 @@ namespace MobileGL::MG_Impl::GLImpl {
case GL_TEXTURE_BINDING_CUBE_MAP: outTarget = TextureTarget::TextureCubeMap; return true;
case GL_TEXTURE_BINDING_CUBE_MAP_ARRAY: outTarget = TextureTarget::TextureCubeMapArray; return true;
case GL_TEXTURE_BINDING_RECTANGLE: outTarget = TextureTarget::TextureRectangle; return true;
case GL_TEXTURE_BINDING_EXTERNAL_OES: outTarget = TextureTarget::External; return true;
default: return false;
}
}
@@ -606,23 +607,50 @@ namespace MobileGL::MG_Impl::GLImpl {
Bool EsIdentityActive();
const Vector<const char*>& ClientSideExtensions() {
static const Vector<const char*> withImages = {"GL_OES_EGL_image", "GL_KHR_robustness",
"GL_ARB_robustness", "GL_EXT_robustness"};
static const Vector<const char*> withoutImages = {"GL_KHR_robustness", "GL_ARB_robustness",
"GL_EXT_robustness"};
// Under the ES identity (EsIdentityActive) the ES names of what desktop GL has in core:
// BGRA textures, render targets and read-backs. GL_BGRA_EXT / GL_BGRA8_EXT as internal
// formats are taken as RGBA storage (TextureEnumConverter), the byte order being the
// pixel transfer format's, as on desktop GL.
static const Vector<const char*> esWithImages = {
"GL_OES_EGL_image", "GL_KHR_robustness", "GL_ARB_robustness",
"GL_EXT_robustness", "GL_EXT_texture_format_BGRA8888", "GL_EXT_read_format_bgra"};
static const Vector<const char*> esWithoutImages = {"GL_KHR_robustness", "GL_ARB_robustness",
"GL_EXT_robustness", "GL_EXT_texture_format_BGRA8888",
"GL_EXT_read_format_bgra"};
// GL_OES_EGL_image_external(_essl3) only where the server samples YUV images: an
// external texture is a 2D texture holding the image converted to RGBA, which needs
// the server to hold the YUV image in the first place (EGLImpl::SharedImageYuvAvailable).
static const auto build = [](Bool images, Bool external, Bool es, Bool esApi) {
Vector<const char*> list;
if (images) list.push_back("GL_OES_EGL_image");
if (images && external) {
list.push_back("GL_OES_EGL_image_external");
list.push_back("GL_OES_EGL_image_external_essl3");
}
for (const char* name : {"GL_KHR_robustness", "GL_ARB_robustness", "GL_EXT_robustness"})
list.push_back(name);
// Under the ES identity (EsIdentityActive) the ES names of what desktop GL has in core:
// BGRA textures, render targets and read-backs. GL_BGRA_EXT / GL_BGRA8_EXT as internal
// formats are taken as RGBA storage (TextureEnumConverter), the byte order being the
// pixel transfer format's, as on desktop GL.
if (es) {
list.push_back("GL_EXT_texture_format_BGRA8888");
list.push_back("GL_EXT_read_format_bgra");
}
// Any context created for the OpenGL ES API, whatever it calls itself: the ES 2.0
// names of what desktop GL has in core - RGB8/RGBA8 and DEPTH_COMPONENT24
// renderbuffers, and glMapBuffer (exported under its OES names as well).
if (esApi) {
list.push_back("GL_OES_rgb8_rgba8");
list.push_back("GL_OES_depth24");
list.push_back("GL_OES_mapbuffer");
}
return list;
};
static const Vector<const char*> lists[16] = {
build(false, false, false, false), build(false, false, false, true),
build(false, false, true, false), build(false, false, true, true),
build(false, true, false, false), build(false, true, false, true),
build(false, true, true, false), build(false, true, true, true),
build(true, false, false, false), build(true, false, false, true),
build(true, false, true, false), build(true, false, true, true),
build(true, true, false, false), build(true, true, false, true),
build(true, true, true, false), build(true, true, true, true)};
const Bool images = MG_Impl::EGLImpl::SharedImagesAvailable();
if (EsIdentityActive()) return images ? esWithImages : esWithoutImages;
return images ? withImages : withoutImages;
const Bool external = images && MG_Impl::EGLImpl::SharedImageYuvAvailable();
const Bool esApi = MG_State::pEGLContext &&
MG_State::pEGLContext->GetCurrentContextClientAPI() == EGL_OPENGL_ES_API;
return lists[(images ? 8 : 0) + (external ? 4 : 0) + (EsIdentityActive() ? 2 : 0) + (esApi ? 1 : 0)];
}
// Robust buffer access as GL 4.5 / KHR_robust_buffer_access_behavior define it: the backend
+64 -11
View File
@@ -1194,6 +1194,25 @@ namespace MobileGL::MG_Impl::GLImpl {
return false;
}
// GL_OES_EGL_image_external: base level 0, CLAMP_TO_EDGE only, NEAREST or LINEAR only.
if (target == TextureTarget::External) {
if (pname == GL_TEXTURE_BASE_LEVEL && param != 0) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidOperation,
MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller, "External texture base level must be zero."));
return false;
}
if (((pname == GL_TEXTURE_WRAP_S || pname == GL_TEXTURE_WRAP_T || pname == GL_TEXTURE_WRAP_R) &&
param != GL_CLAMP_TO_EDGE) ||
(pname == GL_TEXTURE_MIN_FILTER && param != GL_NEAREST && param != GL_LINEAR)) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidEnum,
MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller,
"External textures clamp to edge and filter without mipmaps."));
return false;
}
}
if (target == TextureTarget::TextureRectangle && pname == GL_TEXTURE_BASE_LEVEL && param != 0) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidOperation,
@@ -1635,6 +1654,7 @@ namespace MobileGL::MG_Impl::GLImpl {
case GL_TEXTURE_CUBE_MAP_ARRAY:
case GL_TEXTURE_2D_MULTISAMPLE:
case GL_TEXTURE_2D_MULTISAMPLE_ARRAY:
case GL_TEXTURE_EXTERNAL_OES:
return true;
default:
return false;
@@ -3145,6 +3165,17 @@ namespace MobileGL::MG_Impl::GLImpl {
if (!TextureImpl::ValidateTextureObject(textureObject)) return false;
switch (pname) {
// GL_OES_EGL_image_external: one unit per external texture (the image arrives as RGBA).
case GL_REQUIRED_TEXTURE_IMAGE_UNITS_OES:
if (textureTarget != TextureTarget::External) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidEnum,
MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
"GL_REQUIRED_TEXTURE_IMAGE_UNITS_OES is an external texture's."));
return false;
}
if (params) *params = 1;
break;
case GL_TEXTURE_MAG_FILTER:
if (params) {
*params = (GLint)MG_Util::ConvertSamplerFilterModeToGLEnum(
@@ -6805,6 +6836,13 @@ namespace MobileGL::MG_Impl::GLImpl {
const char* caller) {
if (!textureObject) return false;
const auto target = textureObject->GetTarget();
// GL_OES_EGL_image_external: an external texture has no mip chain to generate.
if (target == TextureTarget::External) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidEnum,
MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller, "External textures have no mipmaps."));
return false;
}
if ((target == TextureTarget::TextureCubeMap || target == TextureTarget::TextureCubeMapArray) &&
!textureObject->IsComplete()) {
MG_State::pGLContext->RecordError(
@@ -7138,15 +7176,20 @@ namespace MobileGL::MG_Impl::GLImpl {
// GL_OES_EGL_image. The only EGLImages with storage behind them are the server's shared images
// (EGL_LINUX_DMA_BUF_EXT): level 0 of the bound texture is defined at the image's size through
// glTexImage2D's own path - so the respecification is recorded and announced as for any other
// - and the backend then swaps that storage for the image. The frontend has no external-image
// target, so GL_TEXTURE_2D is the only one taken.
// - and the backend then swaps that storage for the image.
//
// GL_OES_EGL_image_external's target takes it the same way: an external texture is a 2D one
// below the frontend. A YUV image (NV12/P010) is taken by either target; the backend converts it
// into the texture's RGBA level at the first use in each frame (SharedImageYuv.h).
void EGLImageTargetTexture2DOES(GLenum target, GLeglImageOES image) {
if (target != GL_TEXTURE_2D) {
if (target != GL_TEXTURE_2D && target != GL_TEXTURE_EXTERNAL_OES) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidEnum,
MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "target must be GL_TEXTURE_2D."));
MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
"target must be GL_TEXTURE_2D or GL_TEXTURE_EXTERNAL_OES."));
return;
}
const Bool external = target == GL_TEXTURE_EXTERNAL_OES;
Uint64 sharedImageId = 0;
EGLint width = 0;
EGLint height = 0;
@@ -7158,17 +7201,27 @@ namespace MobileGL::MG_Impl::GLImpl {
}
auto& activeUnit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
const SharedPtr<MG_State::GLState::ITextureObject> textureObject =
activeUnit.GetBindingSlot(TextureTarget::Texture2D).GetBoundObject();
activeUnit.GetBindingSlot(external ? TextureTarget::External : TextureTarget::Texture2D).GetBoundObject();
if (!TextureImpl::ValidateTextureObject(textureObject)) return;
if (!ValidateTextureMutable(textureObject, __func__)) return;
if (!TextureImpl::ValidateTextureSizeRange(width, height, 1)) return;
// NULL is an offset into a bound unpack buffer; the definition must read nothing.
GLint unpackBuffer = 0;
GetIntegerv(GL_PIXEL_UNPACK_BUFFER_BINDING, &unpackBuffer);
if (unpackBuffer != 0) BindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
TexImage2D_State(GL_TEXTURE_2D, 0, GL_RGBA8, width, height, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr);
if (unpackBuffer != 0) BindBuffer(GL_PIXEL_UNPACK_BUFFER, static_cast<GLuint>(unpackBuffer));
if (external) {
// glTexImage2D does not take the external target, so its level 0 is defined here the way
// that call defines a 2D one without data: an RGBA8 base, any older chain dropped.
textureObject->SetInternalFormat(TextureInternalFormat::RGBA8);
auto* mipmapObject = static_cast<MG_State::GLState::TextureObjectMipmap*>(textureObject.get());
DiscardMipmapChainOnBaseRespecification(mipmapObject, TextureUploadTarget::Texture2D, 0);
mipmapObject->AllocateStorage(TextureUploadTarget::Texture2D, 0,
{{width, height, 1}, static_cast<SizeT>(width) * height * 4});
} else {
// NULL is an offset into a bound unpack buffer; the definition must read nothing.
GLint unpackBuffer = 0;
GetIntegerv(GL_PIXEL_UNPACK_BUFFER_BINDING, &unpackBuffer);
if (unpackBuffer != 0) BindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
TexImage2D_State(GL_TEXTURE_2D, 0, GL_RGBA8, width, height, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr);
if (unpackBuffer != 0) BindBuffer(GL_PIXEL_UNPACK_BUFFER, static_cast<GLuint>(unpackBuffer));
}
auto* backendObject = MG_Backend::pActiveBackendObject.get();
if (backendObject == nullptr ||
+4
View File
@@ -246,6 +246,10 @@ namespace MobileGL::MG_Impl {
GETPROC(glBufferSubData, name);
GETPROC(glGetBufferSubData, name);
GETPROC(glMapBuffer, name);
// GL_OES_mapbuffer (ES contexts): the same three calls under their OES names.
if (strcmp(name, "glMapBufferOES") == 0) return (void*)glMapBuffer;
if (strcmp(name, "glUnmapBufferOES") == 0) return (void*)glUnmapBuffer;
if (strcmp(name, "glGetBufferPointervOES") == 0) return (void*)glGetBufferPointerv;
GETPROC(glUnmapBuffer, name);
GETPROC(glGetBufferParameteriv, name);
GETPROC(glGetBufferPointerv, name);
+3 -1
View File
@@ -234,7 +234,9 @@ namespace MobileGL::MG_Pipe {
m_vertices.push_back(static_cast<Uint64>(vertex));
}
}
std::sort(m_vertices.begin(), m_vertices.end());
// An array draw's single range arrives ascending already; sorting it again was most of
// the client's time on a client-array draw.
if (!std::is_sorted(m_vertices.begin(), m_vertices.end())) std::sort(m_vertices.begin(), m_vertices.end());
m_vertices.erase(std::unique(m_vertices.begin(), m_vertices.end()), m_vertices.end());
return true;
}
+1 -1
View File
@@ -132,7 +132,7 @@ namespace MobileGL::MG_Pipe {
// TextureUploadTarget -> TextureTarget inverse exists anywhere in the tree, so
// without this the inline InternalFormat cannot make them fall out at push time and
// the backend keeps reading the frontend attachment objects.
surface.TextureTarget = static_cast<Uint16>(texture->GetTarget());
surface.TextureTarget = static_cast<Uint16>(MGPipeWireTextureTarget(texture->GetTarget()));
return surface;
}
const auto& renderbuffer = attachment.GetRenderbuffer();
+20 -4
View File
@@ -34,6 +34,12 @@ namespace MobileGL::MG_Pipe {
MGPipeVertexInputEmitterInstance().EmitVertexBuffers(m_context, m_baseInstance);
if (m_indexBindingChanged)
MGPipeVertexInputEmitterInstance().EmitIndexBuffer(m_context);
// The bindings no longer name them: the buffers go back to the share group for the
// next client-array draw. Their next respecify orphans the store this draw reads.
auto& group = *m_context.GetShareGroup();
for (auto& buffer : m_vertexBuffers)
if (buffer) group.ReturnClientArrayBuffer(std::move(buffer));
if (m_indexBuffer) group.ReturnClientArrayBuffer(std::move(m_indexBuffer));
}
Bool Prepare(MGPDrawInfo& info, const MGPDrawRange* ranges, Uint32 rangeCount,
@@ -46,7 +52,8 @@ namespace MobileGL::MG_Pipe {
// Start remains an element offset, so base vertex and draw IDs are unchanged.
if (clientIndexBytes != 0) {
if (!clientIndices || clientIndexBytes > std::numeric_limits<Uint32>::max()) return false;
m_indexBuffer = MakeOwnedBuffer(clientIndices, static_cast<SizeT>(clientIndexBytes), BufferTarget::Index);
m_indexBuffer = MakeOwnedBuffer(m_context, clientIndices, static_cast<SizeT>(clientIndexBytes),
BufferTarget::Index);
info.IndexResource = MGPipeResourceTrackerInstance().Find(*m_indexBuffer);
info.Flags &= ~static_cast<Uint8>(kDrawHasUserIndices);
MGPIndexBuffer binding{};
@@ -115,8 +122,14 @@ namespace MobileGL::MG_Pipe {
if (attribute.Offset == 0 || byteCount > std::numeric_limits<SizeT>::max() - attribute.Offset)
return false;
const auto* source = reinterpret_cast<const Uint8*>(attribute.Offset);
const Uint64 first = referenced.front();
if (stride == 0) {
std::memcpy(bytes.data(), source, elementBytes);
} else if (last - first + 1 == referenced.size()) {
// Contiguous elements (every array draw, most indexed ones): the vertex
// records from the first fetched one to the end of the last, in one copy.
const SizeT begin = static_cast<SizeT>(first * stride);
std::memcpy(bytes.data() + begin, source + begin, byteCount - begin);
} else {
// Only dereference fetched elements: sparse indices and first>0
// do not grant permission to read intervening application memory.
@@ -126,7 +139,7 @@ namespace MobileGL::MG_Pipe {
}
}
}
m_vertexBuffers[location] = MakeOwnedBuffer(bytes.data(), bytes.size(), BufferTarget::Vertex);
m_vertexBuffers[location] = MakeOwnedBuffer(m_context, bytes.data(), bytes.size(), BufferTarget::Vertex);
handles[location] = MGPipeResourceTrackerInstance().Find(*m_vertexBuffers[location]);
}
MGPipeVertexInputEmitterInstance().EmitVertexBuffers(m_context, info.StartInstance, &handles);
@@ -173,8 +186,11 @@ namespace MobileGL::MG_Pipe {
return attribute.Size > 0 && attribute.Size <= 4 ? component * attribute.Size : 0;
}
static UniquePtr<Buffer> MakeOwnedBuffer(const void* bytes, SizeT size, BufferTarget target) {
auto buffer = MakeUnique<Buffer>(0);
static UniquePtr<Buffer> MakeOwnedBuffer(Context& context, const void* bytes, SizeT size, BufferTarget target) {
// A kept buffer is already a published resource: respecifying it is fire-and-forget,
// where a new one costs a blocking create (and a destroy after the draw).
auto buffer = context.GetShareGroup()->TakeClientArrayBuffer();
if (!buffer) buffer = MakeUnique<Buffer>(0);
const auto handle = MGPipeResourceTrackerInstance().Find(*buffer);
MGPipeResourceTrackerInstance().NoteBoundAs(handle, target);
buffer->SetUsage(BufferUsage::StreamDraw);
+28 -107
View File
@@ -70,6 +70,11 @@ namespace MobileGL::MG_Pipe {
// The one door into PipeInputs' storage on the client side. A struct rather than a
// list of friend functions so the header names exactly one friend.
struct MGPipeFillAccess {
static void CopyServerPackReference(PipeInputs& dst, const PipeInputs& actual,
const PixelStoreParameters& pack) {
dst.m_pixelStore[0] = pack;
dst.m_pixelStore[1] = actual.m_pixelStore[1];
}
// Copies ONE field's storage out of the live context by calling the GLContext
// accessor of the same name (P1 brief D4: no derivation logic is re-implemented
// here, which is what keeps the copy semantically identical by construction).
@@ -659,56 +664,6 @@ namespace MobileGL::MG_Pipe {
if (!MGPipeVerifyInputs(inputs, g_snapshot, mask, &differing)) ReportDivergence(differing, "entry");
}
#if MOBILEGL_BUILD_DISAGGREGATED
// ---- P7 wave 3 (V1): the compare-at-read oracle inside the server's read_pixels ----
//
// WHAT THE SPLIT ARM FOUND THE FIRST TIME VERIFY AND SPLIT SHARED A BUILD. Every
// glReadPixels on the inproc arm - 8530 reads over 220 of the lane's 1068 entries, and the
// VerifySplitArming. entries on both backends - aborted with
//
// MGPipe: verify read of GetPixelStoreParameters (index 0, 0) differs from the live context
// MGPipe: Fatal{PipeVerifyDiffer, "GetPixelStoreParameters@ReadPixels", verb=6, where=read}
//
// on the server's apply thread, and it was the ONLY field that diverged anywhere in the
// lane. The mechanism is ID-49 and is deliberate: the pack state never shapes the wire
// answer, so MG_Remote/Server/PipeApplier.cpp's read_pixels body writes a NEUTRAL pack
// into the applier's copy of the field (MGPipeApplySetPixelPackState), makes the backend
// read, and restores the pushed one; the client then scatters the tight reply through the
// application's own pack state. The frontend context - this comparator's oracle - still
// holds the application's pack (Alignment 4 by default), so the backend's read of the
// pack half answers {Alignment 1} where the live context says {Alignment 4}. On the
// monolith arm the same neutral window is opened on the frontend context itself
// (GL_Texture.cpp's ScopedNeutralPackState), which is why the monolith lane never saw it.
//
// SO THE ORACLE, NOT THE RULE, IS WHAT CHANGES, AND ONLY IN THAT WINDOW: under a server
// stamp, on the ReadPixels verb, the pack half's expected value is the neutral pack the
// applier reads with, and the UNPACK half and every other field keep the live context as
// their oracle. A server read of the pack half that is neither the application's value
// nor exactly the neutral one is still a divergence and still Fatal.
//
// WHAT THIS GIVES UP IS ONE VALUE, NOT THE FIELD (corrected in P7 wave 3's V1 fix round;
// the first statement of it here said the comparator could no longer see a client that
// pushed a WRONG pack state at all, which is too broad). PipeApplier.cpp's read_pixels
// reads `savedPack` THROUGH THE ACCESSOR - and therefore through this hook - BEFORE it
// installs the neutral value, so the pushed pack has already been compared against the
// live one by then: a correct push is equal and returns, and a wrong push that is not
// exactly the neutral pack fails both compares and is still Fatal. The blind spot is the
// single value {SwapBytes 0, LSBFirst 0, RowLength 0, ImageHeight 0, Skip* 0, Alignment 1},
// and only while the application's own pack differs from it. The client-side scatter that
// consumes the real pack is covered by the readback matrix cases, which compare bytes,
// not fields.
Bool ServerReadsInsideTheNeutralPackWindow(const PipeInputs& self, MGPipeInputField field) {
return field == MGPipeInputField::GetPixelStoreParameters && self.ServerStampedVerb() &&
self.CurrentVerb() == MGPipeVerb::ReadPixels;
}
// The constant itself is MG_Pipe's (MGPipeTypes.h's MGPipeNeutralReadPixelsPack), which is
// the same function MG_Remote/Server/PipeApplier.cpp's read_pixels installs. It used to be
// re-typed here, field for field, to avoid reaching into MG_Remote/Server from MG_Impl -
// but the owner of MGPPixelPackState is MG_Pipe, both sites already include it, and a
// drift between the two spellings would have cost a false divergence rather than a build
// break.
#endif // MOBILEGL_BUILD_DISAGGREGATED
#endif // MOBILEGL_PIPE_VERIFY
} // namespace
@@ -726,9 +681,28 @@ namespace MobileGL::MG_Pipe {
}
void MGPipeVerifyReadHook(const PipeInputs& self, MGPipeInputField field, Uint index0, Uint index1) {
if (&self != &gPipeInputs || !g_verify.Enabled || g_verify.InHook) return;
if (&self != &gPipeInputs || !g_verify.Enabled) return;
const auto index = static_cast<SizeT>(field);
if (kMGPipeInputFieldSticky[index]) return;
#if MOBILEGL_BUILD_DISAGGREGATED
if (self.ServerStampedVerb() && field == MGPipeInputField::GetPixelStoreParameters) {
// Compare the backend's pack field against the last payload applied to this
// context, including the neutral-pack override during readback. Never obtain
// the reference by reading the field being checked, or consult the client's
// GLContext from the server thread. Unpack has no server-side reader.
static thread_local PipeInputs reference;
MGPipeFillAccess::CopyServerPackReference(reference, self, MGPipeApplier().VerifyPixelPack);
if (g_verify.Corrupt && *g_verify.Corrupt == field) {
MGPipeApplyVerifyCorruption(reference, field);
}
if (!MGPipeInputsFieldEqual(field, self, reference)) {
MGLOG_E("MGPipe: verify read of pixel pack differs from the applied pack payload");
ReportDivergence(field, "read");
}
return;
}
#endif
if (g_verify.InHook) return;
auto* ctx = LiveContext();
if (ctx == nullptr) return;
// The whole field is re-read and compared - a superset of "the same indices", so a
@@ -743,68 +717,13 @@ namespace MobileGL::MG_Pipe {
// never recurses (and never reports the inner read against a half-copied scratch).
g_verify.InHook = true;
MGPipeFillAccess::CopyField(g_readScratch, *ctx, field);
// P7 wave 3 (V1): NEGATIVE CONTROL A REACHES THIS ARM TOO. Until this line the knob only
// ever perturbed EntryCompare's snapshot, so every red it could produce said `where=entry`
// and was produced on the CLIENT thread - which left the compare-at-read hook, the arm
// that runs on the server's apply thread and does the whole of the split lane's per-read
// work, with no falsifier at all: a hook that had stopped comparing would have looked
// exactly like a hook with nothing to report. The perturbation goes on the ORACLE, for
// EntryCompare's reason: the arm under test is the stored value, so corrupting THAT would
// be testing the corruption.
//
// WHICH READ THIS BLOCK TURNS RED (V1 fix round 2). The server reads the pack field TWICE
// inside the ReadPixels verb window, and both reads sit inside
// ServerReadsInsideTheNeutralPackWindow below (the verb stamp is up for the whole apply):
// PipeApplier::read_pixels saves the application's pack through the accessor BEFORE it
// installs the neutral one, and the backend's ReadPixels reads the field AFTER. This block
// is what turns the FIRST of them red: at that read `self` still holds the application's
// pack, so the first compare only reaches the window if this perturbation made it differ,
// and inside the window the application's pack against the neutral oracle differs on its
// own. It does nothing for the second read - `self` IS the neutral pack there - which is
// what the re-application inside the window is for.
// Corrupt the reference, not the backend field, to test the comparison itself.
if (g_verify.Corrupt && *g_verify.Corrupt == field) {
MGPipeApplyVerifyCorruption(g_readScratch, field);
}
const Bool equal = MGPipeInputsFieldEqual(field, self, g_readScratch);
g_verify.InHook = false;
if (equal) return;
#if MOBILEGL_BUILD_DISAGGREGATED
if (ServerReadsInsideTheNeutralPackWindow(self, field)) {
// The live context is the application's pack state and the server is, by ID-49,
// reading with the neutral one; the oracle for THAT HALF is the neutral pack, and
// every other byte of the field still has to match the live context.
g_verify.InHook = true;
PipeInputs::VisitStorage(field, g_readScratch, g_readScratch, [](auto& live, auto&) {
if constexpr (std::is_same_v<std::remove_reference_t<decltype(live)>, PixelStoreParameters[2]>) {
live[0] = MGPipeNeutralReadPixelsPack().Pack;
}
return true;
});
// AND THE CONTROL IS RE-APPLIED, because the overwrite above replaces the pack half
// WHOLESALE - the corruption included, since CorruptStorage perturbs an array's FIRST
// ELEMENT and the pack half is element 0. This is the block for the POST-INSTALL read
// (the backend's ReadPixels, after PipeApplier::read_pixels installed the neutral pack
// into gPipeInputs): there `self` is the neutral pack, the overwrite just made the
// oracle equal to it, and only a perturbation applied AFTER the overwrite can make
// that compare differ - the block above cannot reach it. Without this line the one
// window in which this arm is most load-bearing (the server's read_pixels, 8530 reads
// over 220 of the lane's entries) would be the one window MOBILEGL_PIPE_VERIFY_CORRUPT=
// GetPixelStoreParameters cannot turn red - and a control that accepted ONE report
// would never notice, because the saved-pack read still reports through the block
// above. The VerifySplitReadCorrupted. entries and the CI step therefore require at
// least TWO `where=read` reports in the server half (V1 fix round 2). Measured: 3 on
// DirectGLES (its backend reads the field twice after the install) / 2 on
// DirectVulkan; without this block both drop to 1, without the block above
// DirectVulkan drops to 1 while DirectGLES's two post-install reports keep it at 2 -
// so this block is falsified on both backends and the one above on DirectVulkan.
if (g_verify.Corrupt && *g_verify.Corrupt == field) {
MGPipeApplyVerifyCorruption(g_readScratch, field);
}
const Bool equalToTheNeutralPack = MGPipeInputsFieldEqual(field, self, g_readScratch);
g_verify.InHook = false;
if (equalToTheNeutralPack) return;
}
#endif
MGLOG_E("MGPipe: verify read of %s (index %u, %u) differs from the live context", kMGPipeInputFieldNames[index],
index0, index1);
ReportDivergence(field, "read");
@@ -3988,6 +3907,7 @@ namespace MobileGL::MG_Pipe {
}
void MGPipeClientDrainTextureUploads() {
#if MOBILEGL_BUILD_DISAGGREGATED
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
if (MG_Remote::Client::RunsAsTheServerRole()) return;
auto* ctx = LiveContext();
@@ -3999,6 +3919,7 @@ namespace MobileGL::MG_Pipe {
P4aFamilyDependenciesAreSet(kMGPipeSubsystemTextureResources, pushMask)) {
(void)MGPipeTextureEmitterInstance().DrainTextureSubData(*ctx);
}
#endif
}
#if MOBILEGL_BUILD_DISAGGREGATED
+4 -1
View File
@@ -629,6 +629,9 @@ namespace MobileGL::MG_Pipe {
// sample this unit".
inline MobileGL::TextureTarget MGPipeSamplerUniformTextureTarget(GLenum uniformType) {
switch (uniformType) {
// GL_OES_EGL_image_external: the unit's external binding, whose texture is a 2D one below.
case GL_SAMPLER_EXTERNAL_OES:
return TextureTarget::External;
case GL_SAMPLER_1D:
case GL_INT_SAMPLER_1D:
case GL_UNSIGNED_INT_SAMPLER_1D:
@@ -995,7 +998,7 @@ namespace MobileGL::MG_Pipe {
// and not the storage owner's, which is the whole point of the call; for an
// ordinary texture it is simply its format.
m_lastView.InternalFormat = static_cast<Uint32>(texture.GetFormat());
m_lastView.Target = static_cast<Uint8>(texture.GetTarget());
m_lastView.Target = static_cast<Uint8>(MGPipeWireTextureTarget(texture.GetTarget()));
// THE FOUR RESTRICTIONS COME FROM ONE PLACE. TextureObjectBase leaves all four at
// 0 for an ordinary texture and glTextureView writes them for a view, and a view
// always has NumLevels >= 1 - so a zero here unambiguously means "no restriction,
+50 -7
View File
@@ -15,6 +15,7 @@
#include <Defines.h>
#include <Init.h>
#include <MG_Backend/BackendObjects.h>
#include <MG_Impl/EGLImpl/EGLImpl.h>
#include <algorithm>
#include <cerrno>
@@ -26,14 +27,19 @@
namespace {
using MobileGL::MG_Backend::BackendObject;
constexpr uint32_t kFormats[] = {MOBILEGL_SHARED_IMAGE_FOURCC_ABGR8888, MOBILEGL_SHARED_IMAGE_FOURCC_XBGR8888,
MOBILEGL_SHARED_IMAGE_FOURCC_ARGB8888, MOBILEGL_SHARED_IMAGE_FOURCC_XRGB8888};
constexpr uint32_t kRgbFormats[] = {MOBILEGL_SHARED_IMAGE_FOURCC_ABGR8888, MOBILEGL_SHARED_IMAGE_FOURCC_XBGR8888,
MOBILEGL_SHARED_IMAGE_FOURCC_ARGB8888, MOBILEGL_SHARED_IMAGE_FOURCC_XRGB8888};
constexpr uint32_t kYuvFormats[] = {MOBILEGL_SHARED_IMAGE_FOURCC_NV12, MOBILEGL_SHARED_IMAGE_FOURCC_P010};
bool FourccIsYuv(uint32_t fourcc) {
return fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_NV12 || fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_P010;
}
bool FourccSupported(uint32_t fourcc) {
for (uint32_t format : kFormats) {
for (uint32_t format : kRgbFormats) {
if (format == fourcc) return true;
}
return false;
return FourccIsYuv(fourcc);
}
// Serializes the calls against each other (not against GL calls: see the header).
@@ -54,10 +60,20 @@ MOBILEGL_EXPORT uint32_t mobilegl_shared_image_abi_version(void) {
return MOBILEGL_SHARED_IMAGE_ABI_VERSION;
}
// The YUV formats only when the server holds them, which takes the session (and one round trip,
// once): a caller asking which formats exist is about to allocate one.
MOBILEGL_EXPORT int mobilegl_shared_image_formats(uint32_t* formats, uint32_t capacity) {
constexpr uint32_t total = sizeof(kFormats) / sizeof(kFormats[0]);
if (formats == nullptr && capacity != 0) return -EINVAL;
for (uint32_t i = 0; i < total && i < capacity; ++i) formats[i] = kFormats[i];
const std::lock_guard<std::mutex> lock(g_apiMutex);
uint32_t total = 0;
const auto put = [&](uint32_t format) {
if (total < capacity) formats[total] = format;
++total;
};
for (uint32_t format : kRgbFormats) put(format);
if (ActiveBackend(true) != nullptr && MobileGL::MG_Impl::EGLImpl::SharedImageYuvAvailable()) {
for (uint32_t format : kYuvFormats) put(format);
}
return static_cast<int>(total);
}
@@ -83,6 +99,8 @@ MOBILEGL_EXPORT int mobilegl_shared_image_allocate(uint32_t width, uint32_t heig
answer.stride = image.Stride;
answer.offset = image.Offset;
answer.modifier = image.Modifier;
answer.plane1_stride = image.Plane1Stride;
answer.plane1_offset = image.Plane1Offset;
// An older caller's struct is a prefix of this one; a newer caller's tail stays as it was.
std::memcpy(out, &answer, std::min<size_t>(out->struct_size, sizeof(answer)));
return 0;
@@ -90,7 +108,8 @@ MOBILEGL_EXPORT int mobilegl_shared_image_allocate(uint32_t width, uint32_t heig
MOBILEGL_EXPORT int mobilegl_shared_image_import(int fd, uint32_t width, uint32_t height, uint32_t fourcc,
uint64_t* out_id) {
if (fd < 0 || out_id == nullptr || width == 0 || height == 0 || !FourccSupported(fourcc)) return -EINVAL;
if (fd < 0 || out_id == nullptr || width == 0 || height == 0 || !FourccSupported(fourcc) || FourccIsYuv(fourcc))
return -EINVAL;
const std::lock_guard<std::mutex> lock(g_apiMutex);
BackendObject* backend = ActiveBackend(true);
if (backend == nullptr) return -EIO;
@@ -100,6 +119,30 @@ MOBILEGL_EXPORT int mobilegl_shared_image_import(int fd, uint32_t width, uint32_
return 0;
}
MOBILEGL_EXPORT int mobilegl_shared_image_import_planes(int fd, uint32_t width, uint32_t height, uint32_t fourcc,
const struct mobilegl_shared_image_planes* planes,
uint64_t* out_id) {
if (fd < 0 || out_id == nullptr || planes == nullptr || width == 0 || height == 0 || !FourccIsYuv(fourcc) ||
planes->struct_size < sizeof(mobilegl_shared_image_planes) || planes->plane_count != 2)
return -EINVAL;
const std::lock_guard<std::mutex> lock(g_apiMutex);
BackendObject* backend = ActiveBackend(true);
if (backend == nullptr) return -EIO;
MobileGL::MG_Backend::SharedImageImportLayout layout;
layout.PlaneCount = 2;
for (int i = 0; i < 2; ++i) {
layout.Offset[i] = planes->offset[i];
layout.Pitch[i] = planes->pitch[i];
}
constexpr uint64_t kModifierInvalid = 0x00ffffffffffffffull;
layout.HasModifier = planes->modifier != kModifierInvalid;
layout.Modifier = planes->modifier;
uint64_t id = 0;
if (!backend->ImportSharedImagePlanes(fd, width, height, fourcc, layout, &id)) return -EIO;
*out_id = id;
return 0;
}
MOBILEGL_EXPORT int mobilegl_server_available(void) {
return MobileGL::ImplementationAvailable() ? 1 : 0;
}
+28 -1
View File
@@ -36,7 +36,9 @@
extern "C" {
#endif
#define MOBILEGL_SHARED_IMAGE_ABI_VERSION 1u
/* 2: YUV images (NV12, P010), the plane-1 fields of mobilegl_shared_image, and
* mobilegl_shared_image_import_planes. */
#define MOBILEGL_SHARED_IMAGE_ABI_VERSION 2u
/* DRM fourccs (little-endian packed). ABGR8888 is R, G, B, A in memory - GL's RGBA8. */
#define MOBILEGL_SHARED_IMAGE_FOURCC_ABGR8888 0x34324241u /* 'AB24' */
@@ -46,6 +48,12 @@ extern "C" {
* writes them, so the name an image is allocated under never changes its colours. */
#define MOBILEGL_SHARED_IMAGE_FOURCC_ARGB8888 0x34325241u /* 'AR24' */
#define MOBILEGL_SHARED_IMAGE_FOURCC_XRGB8888 0x34325258u /* 'XR24' */
/* Two-plane YUV 4:2:0 (ABI 2), when the server's platform allocates it (listed by
* mobilegl_shared_image_formats only then): plane 0 is Y, plane 1 Cb Cr interleaved, both in the one
* descriptor at the offsets and pitches the allocation reports. They are SAMPLED only - an EGLImage
* of one binds through GL_OES_EGL_image_external - and their layout is still the allocator's. */
#define MOBILEGL_SHARED_IMAGE_FOURCC_NV12 0x3231564Eu /* 'NV12' */
#define MOBILEGL_SHARED_IMAGE_FOURCC_P010 0x30313050u /* 'P010' */
/* One allocated image. VERSIONED BY SIZE: the caller sets struct_size to the sizeof it was built
* against and the library writes no byte past it, so fields are only ever appended. */
@@ -60,6 +68,18 @@ struct mobilegl_shared_image {
uint32_t offset; /* plane 0, bytes */
uint32_t reserved; /* zero */
uint64_t modifier; /* DRM format modifier of the layout (DRM_FORMAT_MOD_INVALID today) */
/* ABI 2: a YUV image's plane 1 (same fd); zero for the one-plane formats. */
uint32_t plane1_stride;
uint32_t plane1_offset;
};
/* Where an imported two-plane buffer's planes are (ABI 2). */
struct mobilegl_shared_image_planes {
uint32_t struct_size;
uint32_t plane_count; /* 2 */
uint32_t offset[2];
uint32_t pitch[2];
uint64_t modifier; /* DRM_FORMAT_MOD_INVALID when none was given */
};
/* MOBILEGL_SHARED_IMAGE_ABI_VERSION of the library. */
@@ -78,6 +98,11 @@ int mobilegl_shared_image_allocate(uint32_t width, uint32_t height, uint32_t fou
* -EIO: not one of the server's images, or the size/format disagree. */
int mobilegl_shared_image_import(int fd, uint32_t width, uint32_t height, uint32_t fourcc, uint64_t* out_id);
/* The same for a YUV fourcc whose two planes are in `fd` as `planes` says (ABI 2). A YUV buffer the
* server did not allocate is taken too, read through the server's CPU-copy fallback. */
int mobilegl_shared_image_import_planes(int fd, uint32_t width, uint32_t height, uint32_t fourcc,
const struct mobilegl_shared_image_planes* planes, uint64_t* out_id);
/* Drops this process's reference (an allocation's or an import's). */
int mobilegl_shared_image_release(uint64_t id);
@@ -95,6 +120,8 @@ typedef int (*PFN_mobilegl_shared_image_formats)(uint32_t*, uint32_t);
typedef int (*PFN_mobilegl_shared_image_allocate)(uint32_t, uint32_t, uint32_t, struct mobilegl_shared_image*);
typedef int (*PFN_mobilegl_shared_image_import)(int, uint32_t, uint32_t, uint32_t, uint64_t*);
typedef int (*PFN_mobilegl_shared_image_release)(uint64_t);
typedef int (*PFN_mobilegl_shared_image_import_planes)(int, uint32_t, uint32_t, uint32_t,
const struct mobilegl_shared_image_planes*, uint64_t*);
#ifdef __cplusplus
}
@@ -374,5 +374,47 @@ void main() { o_color = vec4(0.1, 0.2, 0.8, 1.0); }
<< (std::fabs(blitted - 0.125f) < 1e-3f ? " - the destination kept its own clear" : "");
}
TEST_F(LayeredAttachmentBarrierScenario, EndingAMiddleMipDrawPreservesOtherLevels) {
if (!Ready()) return;
GLuint texture = 0;
glGenTextures(1, &texture);
m_textures.push_back(texture);
glBindTexture(GL_TEXTURE_2D, texture);
glTexStorage2D(GL_TEXTURE_2D, 3, GL_RGBA8, 16, 16);
for (int level = 0; level < 3; ++level) {
const int size = 16 >> level;
const std::vector<Rgba8> initial(size * size, LayerFill(level));
glTexSubImage2D(GL_TEXTURE_2D, level, 0, 0, size, size,
GL_RGBA, GL_UNSIGNED_BYTE, initial.data());
}
GLuint fbo = 0;
glGenFramebuffers(1, &fbo);
m_fbos.push_back(fbo);
glBindFramebuffer(GL_FRAMEBUFFER, fbo);
glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, 1);
ASSERT_EQ(glCheckFramebufferStatus(GL_FRAMEBUFFER), GL_FRAMEBUFFER_COMPLETE);
glViewport(0, 0, 8, 8);
glDisable(GL_SCISSOR_TEST);
glDisable(GL_DEPTH_TEST);
glDisable(GL_BLEND);
glDrawBuffer(GL_COLOR_ATTACHMENT0);
DrawFullViewportQuad(m_program);
// Ending this pass reconciles both lower and upper mip layouts. Their barriers
// must not see the pass as active and recursively end it a second time.
glFinish();
for (int level = 0; level < 3; ++level) {
glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, level);
glReadBuffer(GL_COLOR_ATTACHMENT0);
Rgba8 actual{};
glReadPixels(0, 0, 1, 1, GL_RGBA, GL_UNSIGNED_BYTE, &actual);
const Rgba8 expected = level == 1 ? kPaintedColor : LayerFill(level);
EXPECT_NEAR(actual.r, expected.r, 2) << "level " << level;
EXPECT_NEAR(actual.g, expected.g, 2);
EXPECT_NEAR(actual.b, expected.b, 2);
EXPECT_EQ(actual.a, expected.a);
}
EXPECT_EQ(FirstGLError(), 0u);
}
} // namespace
} // namespace MGITest
@@ -322,11 +322,14 @@ TEST_F(MagmaRunAheadScenario, PresentCreditActuallyParksTheClientAtItsConfigured
Gl().EndFrame(); // credit + 1 must park before it can publish its present.
stop.store(true); observer.join(); hold.Release();
EXPECT_TRUE(sawPark.load()) << "no real producer park while the present window was full";
const auto issuedBeforeFinish = PeekSplitRuntime().emitSeq;
glFinish();
const auto after=PeekSplitRuntime();
EXPECT_EQ(after.presentCreditWaits,before.presentCreditWaits+1);
EXPECT_EQ(after.presentAckSerial,before.presentCredit+1);
EXPECT_EQ(after.appliedSeq,after.emitSeq);
// Finish waits for prior work, then asynchronously destroys its internal fence.
// That cleanup may still be queued when Finish returns.
EXPECT_GE(after.appliedSeq,issuedBeforeFinish);
RecordProperty("observed_present_credit",std::to_string(before.presentCredit));
RecordProperty("observed_producer_park",sawPark.load() ? "yes" : "no");
}
+34 -1
View File
@@ -263,6 +263,11 @@ namespace MobileGL::MG_Pipe {
return static_cast<Uint32>(MGPipeResourceTarget::TexCubeArray);
case MobileGL::TextureTarget::Texture2DMultisampleArray:
return static_cast<Uint32>(MGPipeResourceTarget::Tex2DMSArray);
// THE ONE FOLD (rather than an enumerator of its own, D-A3): an external texture
// (GL_OES_EGL_image_external) is a 2D texture to every backend - a single RGBA level 0,
// sampled through a sampler2D - and only the frontend tells the two binding points apart.
case MobileGL::TextureTarget::External:
return static_cast<Uint32>(MGPipeResourceTarget::Tex2D);
// NOT TEXTURE TARGETS. Listed rather than defaulted so the completeness assert still
// sees them, and mapped to the sentinel because no descriptor may carry either: the
// count is the enum's bound and Unknown is what an unresolved GL enum becomes.
@@ -273,6 +278,12 @@ namespace MobileGL::MG_Pipe {
return kMGPipeResourceTargetUnmapped;
}
// The TextureTarget a record that carries one raw (MGPSamplerView, MGPSurface) names: the
// frontend's own, but for the external target, which is a 2D texture below the frontend.
constexpr MobileGL::TextureTarget MGPipeWireTextureTarget(MobileGL::TextureTarget target) {
return target == MobileGL::TextureTarget::External ? MobileGL::TextureTarget::Texture2D : target;
}
constexpr Bool MGPipeEveryTextureTargetIsMapped() {
for (SizeT i = 0; i < static_cast<SizeT>(MobileGL::TextureTarget::TextureTargetCount); ++i) {
if (MGPipeResourceTargetForTextureTarget(static_cast<MobileGL::TextureTarget>(i)) ==
@@ -2044,6 +2055,25 @@ namespace MobileGL::MG_Pipe {
inline constexpr Uint32 kMGPBufferAgeDamageRegionFollows = 1u;
inline constexpr Uint32 kMGPNativeFenceFdFollows = 1u;
// An Import of a YUV fourcc (NV12, P010) carries its planes and colour hints in Damage, with
// DamageCount = kMGPImportPlaneWords (the field is a Present's otherwise, and 0 on an RGBA
// Import): both planes live in the one descriptor the Import sends, at these byte offsets and
// row pitches. The hints are EGL_EXT_image_dma_buf_import's values, 0 where not given.
enum MGPImportPlaneWord : Uint32 {
kMGPImportPlaneCount = 0,
kMGPImportPlane0Offset = 1,
kMGPImportPlane1Offset = 2,
kMGPImportPlane0Pitch = 3,
kMGPImportPlane1Pitch = 4,
kMGPImportYuvColorSpace = 5,
kMGPImportYuvRange = 6,
kMGPImportYuvSiting = 7, // horizontal | vertical << 16
kMGPImportModifierLo = 8,
kMGPImportModifierHi = 9,
kMGPImportHasModifier = 10,
kMGPImportPlaneWords = 11,
};
struct MGPSharedImageOp {
Uint32 Op;
Uint32 Format; // DRM fourcc
@@ -2067,8 +2097,11 @@ namespace MobileGL::MG_Pipe {
Uint32 Stride;
Uint32 Offset;
Int32 BufferAge; // QueryBufferAge only
// A YUV image's Cb Cr plane (Allocate, Import): same descriptor, this offset and pitch.
Uint32 Plane1Stride;
Uint32 Plane1Offset;
};
MGP_ASSERT_POD(MGPSharedImageReply, 40);
MGP_ASSERT_POD(MGPSharedImageReply, 48);
// The aux-socket sideband an Import's descriptor travels with (and an Allocate's answer).
inline constexpr Uint32 kMGPSharedImageFdMagic = 0x49534D47u; // 'MGSI'
+3
View File
@@ -2097,6 +2097,9 @@ namespace MobileGL::MG_Pipe {
}
void MGPipeApplySetPixelPackState(const MGPPixelPackState& pack) {
#if MOBILEGL_PIPE_VERIFY
MGPipeApplier().VerifyPixelPack = pack.Pack;
#endif
MGPipeApplyAccess::PackState(gPipeInputs) = pack.Pack;
}
+5
View File
@@ -602,6 +602,11 @@ namespace MobileGL::MG_Pipe {
// The per-context half. Everything here is cleared by MGPipeApplierReset, EXCEPT the serials,
// which only ever advance (see there).
struct MGPipeWorkingState {
#if MOBILEGL_PIPE_VERIFY
// Reference from the pack-state payload, separate from the backend-visible storage.
// The server has no frontend GLContext to use as a read-time oracle.
PixelStoreParameters VerifyPixelPack{};
#endif
// Indexed by slot; slot 0 is the reserved null handle and is never live
// (MGPipeHandles.h kMGPipeFirstAllocatableSlot).
Vector<MGPipeRenderStateCsoRecord> RenderStateCsos;
@@ -604,6 +604,8 @@ namespace MobileGL::MG_Remote::Client {
out->Stride = reply.Stride;
out->Offset = reply.Offset;
out->Modifier = reply.Modifier;
out->Plane1Stride = reply.Plane1Stride;
out->Plane1Offset = reply.Plane1Offset;
{
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
++m_sessionSharedImages[reply.ImageId];
@@ -630,6 +632,40 @@ namespace MobileGL::MG_Remote::Client {
return true;
}
Bool BackendObject_Remote::ImportSharedImagePlanes(int fd, Uint32 width, Uint32 height, Uint32 fourcc,
const MG_Backend::SharedImageImportLayout& layout,
Uint64* outId) {
ClientSession* session = ClientSession::Active();
if (session == nullptr || fd < 0) return false;
const ClientSession::ScopedCurrentSessionWork tableOp;
MG_Pipe::MGPSharedImageOp op{};
op.Op = MG_Pipe::kMGPSharedImageImport;
op.Format = fourcc;
op.Width = width;
op.Height = height;
// The planes and hints ride in Damage (MGPImportPlaneWord).
op.DamageCount = MG_Pipe::kMGPImportPlaneWords;
op.Damage[MG_Pipe::kMGPImportPlaneCount] = static_cast<Int32>(layout.PlaneCount);
op.Damage[MG_Pipe::kMGPImportPlane0Offset] = static_cast<Int32>(layout.Offset[0]);
op.Damage[MG_Pipe::kMGPImportPlane1Offset] = static_cast<Int32>(layout.Offset[1]);
op.Damage[MG_Pipe::kMGPImportPlane0Pitch] = static_cast<Int32>(layout.Pitch[0]);
op.Damage[MG_Pipe::kMGPImportPlane1Pitch] = static_cast<Int32>(layout.Pitch[1]);
op.Damage[MG_Pipe::kMGPImportYuvColorSpace] = static_cast<Int32>(layout.ColorSpace);
op.Damage[MG_Pipe::kMGPImportYuvRange] = static_cast<Int32>(layout.Range);
op.Damage[MG_Pipe::kMGPImportYuvSiting] = static_cast<Int32>((layout.SitingX & 0xffffu) | (layout.SitingY << 16));
op.Damage[MG_Pipe::kMGPImportModifierLo] = static_cast<Int32>(static_cast<Uint32>(layout.Modifier));
op.Damage[MG_Pipe::kMGPImportModifierHi] = static_cast<Int32>(static_cast<Uint32>(layout.Modifier >> 32));
op.Damage[MG_Pipe::kMGPImportHasModifier] = layout.HasModifier ? 1 : 0;
MG_Pipe::MGPSharedImageReply reply{};
if (!session->EmitSharedImage(op, fd, &reply, nullptr)) return false;
if (outId != nullptr) *outId = reply.ImageId;
{
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
++m_sessionSharedImages[reply.ImageId];
}
return true;
}
Bool BackendObject_Remote::ReleaseSharedImage(Uint64 id) {
ClientSession* session = ClientSession::Active();
if (session == nullptr || !session->Started()) return false;
@@ -90,6 +90,8 @@ namespace MobileGL::MG_Remote::Client {
Bool AllocateSharedImage(Uint32 width, Uint32 height, Uint32 fourcc,
MG_Backend::SharedImageExport* out) override;
Bool ImportSharedImage(int fd, Uint32 width, Uint32 height, Uint32 fourcc, Uint64* outId) override;
Bool ImportSharedImagePlanes(int fd, Uint32 width, Uint32 height, Uint32 fourcc,
const MG_Backend::SharedImageImportLayout& layout, Uint64* outId) override;
Bool ReleaseSharedImage(Uint64 id) override;
Bool PresentToSharedImage(Uint64 id, const MG_Util::Damage::Region& region) override;
Bool QueryBufferAge(Bool damageRegionFollows, EGLint* age) override;
+41 -30
View File
@@ -2492,38 +2492,47 @@ namespace MobileGL::MG_Remote::Client {
// LastPublishedSeq by construction and a second wait would only add a park on a
// watermark that is already reached.
if (!m_runAheadArmed || !m_started) return;
const Uint64 target = m_producer.LastPublishedSeq();
if (target == 0) return;
if (m_producer.LastPublishedSeq() == 0) return;
const auto deadline = std::chrono::steady_clock::now() + std::chrono::milliseconds(kBarrierTimeoutMs);
const BarrierWaitScope waiting;
const Transport::SessionWait wait =
WaitForAppliedBudget(m_producer, *m_events, target, kBarrierTimeoutMs);
if (wait == Transport::SessionWait::ShutDown) {
// `dl`: the same two-events distinction the verb barrier makes. Only a peer that
// HUNG UP is a device loss; Stop() reaches this path too.
if (const Transport::Doorbell* bell = m_producer.SelfDoorbell();
bell != nullptr && bell->PeerHungUp()) {
LatchDeviceLost("a forced run-ahead wait woke on a peer that had hung up");
for (;;) {
const Uint64 target = m_producer.LastPublishedSeq();
const auto now = std::chrono::steady_clock::now();
const auto remaining = std::chrono::duration_cast<std::chrono::milliseconds>(deadline - now).count();
const Transport::SessionWait wait = now >= deadline
? Transport::SessionWait::TimedOut
: m_producer.WaitForAppliedOrEventBacklog(
target, static_cast<Uint32>(std::max<Int64>(remaining, 1)));
if (wait == Transport::SessionWait::ShutDown) {
// Only a peer that hung up is a device loss; Stop() reaches this path too.
if (const Transport::Doorbell* bell = m_producer.SelfDoorbell();
bell != nullptr && bell->PeerHungUp()) {
LatchDeviceLost("a forced run-ahead wait woke on a peer that had hung up");
}
if (!DeviceLost())
MGLOG_E("MG_Remote client: the forced wait for %s woke on a dead doorbell; the "
"server is gone", why);
return;
}
if (wait != Transport::SessionWait::Reached) {
SessionFail(MGFatalFamily::BarrierTimeout, "MGPipe: Fatal{BarrierTimeout, \"%s\"} - a forced run-ahead wait did not "
"reach appliedSeq %llu within %u ms. The forced waits of CONTRACT-P5E §2.5 "
"are the points at which the client must be in step again, so a wedge here "
"is a wedge in the queue and not a slow frame",
why, static_cast<unsigned long long>(target), kBarrierTimeoutMs);
}
// Drain even when only the event backlog woke us: the server may need that
// space before it can advance appliedSeq. SurfaceChanged callbacks and pin
// destruction can themselves publish records, so the original target alone
// no longer proves the applier is idle after these operations.
DrainEventRing(*m_events);
ReleaseFillPins();
const auto* progress = m_producer.Progress();
if (m_producer.LastPublishedSeq() == target && progress != nullptr &&
Transport::Watermark::Reached(progress->appliedSeq, target)) {
return;
}
// The doorbell died: teardown, not a fault - the same answer EmitAndWaitTails
// gives, and for the same reason.
if (!DeviceLost()) // see the barrier above: the latch already said it
MGLOG_E("MG_Remote client: the forced wait for %s woke on a dead doorbell; the "
"server is gone", why);
return;
}
if (wait != Transport::SessionWait::Reached) {
SessionFail(MGFatalFamily::BarrierTimeout, "MGPipe: Fatal{BarrierTimeout, \"%s\"} - a forced run-ahead wait did not "
"reach appliedSeq %llu within %u ms. The forced waits of CONTRACT-P5E §2.5 "
"are the points at which the client must be in step again, so a wedge here "
"is a wedge in the queue and not a slow frame",
why, static_cast<unsigned long long>(target), kBarrierTimeoutMs);
}
// EVERY WAIT EXIT DRAINS (§2.6). It is what the flow-control deadlock argument rests
// on: the server blocks only on a full event ring, the client blocks only on
// watermarks the server advances, and every client wait drains - so at most one side
// is parked at any instant.
DrainEventRing(*m_events);
ReleaseFillPins();
#else
(void)why;
#endif
@@ -2551,8 +2560,10 @@ namespace MobileGL::MG_Remote::Client {
// glFinish. Nothing goes on the wire - Flush and Finish have no record (ARCHITECTURE
// §11) - and under lockstep there was nothing to do, because the client had already
// waited out every command it issued. Under run-ahead "the commands issued so far have
// completed" is a promise the queue can break, so this is where it is kept.
// completed" is a promise the queue can break, so this is where it is kept - and
// "completed" includes the GPU, so a fence round trip follows the apply.
WaitForApplyToCatchUp("glFinish");
EmitFinishWait();
}
Uint64 ClientSession::AcquirePresentCredit() {
+17
View File
@@ -2085,6 +2085,23 @@ namespace MobileGL::MG_Remote::Client {
} // namespace
// glFinish's GPU half: a server fence created after everything the client issued, waited
// with GL_SYNC_FLUSH_COMMANDS_BIT, then destroyed. Waiting for the apply thread alone left
// the GPU queue unbounded behind a "finished" glFinish.
void EmitFinishWait() {
ClientSession* session = ClientSession::Active();
if (session == nullptr || !session->Started() || session->DeviceLost()) return;
const std::lock_guard<std::mutex> lock(g_fenceMutex);
const MG_Pipe::MGPipeHandle handle = MG_Pipe::MGPipeSlots().Allocate(MG_Pipe::MGPipeKind::Fence);
const MG_Pipe::MGPHandleOnly desc{handle, static_cast<Uint32>(MG_Pipe::MGPipeKind::Fence), 0};
session->EmitAndWait(MG_Pipe::MGPWireOp::FenceCreate, &desc, sizeof(desc), nullptr, 0, nullptr, 0, nullptr);
constexpr GLuint64 kFinishTimeoutNs = 10ull * 1000 * 1000 * 1000;
const MG_Pipe::MGPFenceWait request{handle, kFinishTimeoutNs, GL_SYNC_FLUSH_COMMANDS_BIT, 0};
(void)ReadFenceReply(*session, MG_Pipe::MGPWireOp::FenceWait, &request, sizeof(request));
session->EmitAndWait(MG_Pipe::MGPWireOp::FenceDestroy, &desc, sizeof(desc), nullptr, 0, nullptr, 0, nullptr);
MG_Pipe::MGPipeSlots().Free(MG_Pipe::MGPipeKind::Fence, handle);
}
void NoteFenceSignaledByServer(MG_Pipe::MGPipeHandle fence) {
RememberFenceSignaled(fence);
g_fenceServerReports.fetch_add(1, std::memory_order_relaxed);
+2
View File
@@ -291,6 +291,8 @@ namespace MobileGL::MG_Remote::Client {
//
// The event drain's half: the server reported `fence` signaled (kEventFenceSignaled). Every
// later glClientWaitSync / glGetSynciv(GL_SYNC_STATUS) on it answers without a round trip.
// glFinish's wait for the GPU (a server fence created, waited and destroyed).
void EmitFinishWait();
void NoteFenceSignaledByServer(MG_Pipe::MGPipeHandle fence);
// What the fence polls cost, for the lanes. `LocalAnswers` = polls answered with no record
+43
View File
@@ -1517,6 +1517,8 @@ namespace MobileGL::MG_Remote::Server {
reply.Format = image.Fourcc;
reply.Stride = image.Stride;
reply.Offset = image.Offset;
reply.Plane1Stride = image.Plane1Stride;
reply.Plane1Offset = image.Plane1Offset;
};
switch (op.Op) {
case MG_Pipe::kMGPSharedImageAllocate: {
@@ -1554,6 +1556,33 @@ namespace MobileGL::MG_Remote::Server {
return false;
}
SI::ImageRef image = SI::Identify(fd, why);
// YUV: the planes and colour hints ride in Damage (MGPImportPlaneWord). A buffer this
// server did not allocate is taken too, through the labelled CPU-copy fallback
// (SharedImageRegistry.h, Image::ForeignSource); RGBA ones never are.
const Bool yuv = SI::FourccIsYuv(op.Format) && op.DamageCount == MG_Pipe::kMGPImportPlaneWords;
if (image == nullptr && yuv) {
SI::PlaneLayout layout;
layout.Count = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportPlaneCount]);
layout.Offset[0] = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportPlane0Offset]);
layout.Offset[1] = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportPlane1Offset]);
layout.Pitch[0] = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportPlane0Pitch]);
layout.Pitch[1] = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportPlane1Pitch]);
// The copy reads rows: only a buffer said to be linear (or said nothing) is one.
const Uint64 modifier =
static_cast<Uint64>(static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportModifierLo])) |
(static_cast<Uint64>(static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportModifierHi])) << 32);
if (op.Damage[MG_Pipe::kMGPImportHasModifier] != 0 && modifier != SI::kModifierLinear &&
modifier != SI::kModifierInvalid) {
char text[96];
std::snprintf(text, sizeof(text), "; a foreign buffer of modifier 0x%016llx is not linear",
static_cast<unsigned long long>(modifier));
why += text;
} else {
std::string foreignWhy;
image = SI::ImportForeignYuv(fd, op.Width, op.Height, op.Format, layout, foreignWhy);
if (image == nullptr) why += "; as a foreign YUV buffer: " + foreignWhy;
}
}
#if !defined(_WIN32)
::close(fd);
#endif
@@ -1561,6 +1590,20 @@ namespace MobileGL::MG_Remote::Server {
MGLOG_W_ONCE("MG_Remote server: a dma-buf import was refused: %s", why.c_str());
return false;
}
if (SI::FourccIsYuv(image->Fourcc) != SI::FourccIsYuv(op.Format) ||
(SI::FourccIsYuv(op.Format) && image->Fourcc != op.Format)) {
MGLOG_W("MG_Remote server: a dma-buf import of image %llu (fourcc 0x%08x) as fourcc 0x%08x was refused",
static_cast<unsigned long long>(image->Id), image->Fourcc, op.Format);
return false;
}
if (yuv) {
SI::Yuv::Hints hints;
hints.ColorSpace = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportYuvColorSpace]);
hints.Range = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportYuvRange]);
hints.SitingX = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportYuvSiting]) & 0xffffu;
hints.SitingY = static_cast<Uint32>(op.Damage[MG_Pipe::kMGPImportYuvSiting]) >> 16;
SI::SetYuvHints(*image, hints);
}
if (image->Width != op.Width || image->Height != op.Height || !SI::FourccSupported(op.Format)) {
MGLOG_W("MG_Remote server: a dma-buf import of image %llu (%ux%u) as %ux%u fourcc 0x%08x was refused",
static_cast<unsigned long long>(image->Id), image->Width, image->Height, op.Width, op.Height,
+19 -7
View File
@@ -1085,6 +1085,7 @@ namespace MobileGL::MG_Remote::Server {
static std::uint64_t sFrameRecords = 0;
static std::uint64_t sFrameIndex = 0;
Uint64 applied = 0;
const bool timed = ServerFrameStatsEnabled();
for (;;) {
if (SessionLatched()) break;
if (m_windowRequests.load(std::memory_order_acquire) != 0) {
@@ -1093,15 +1094,25 @@ namespace MobileGL::MG_Remote::Server {
}
bool corrupt = false;
const bool popped = consumer.ApplyOne(
[this, &applier](const Transport::RingRecordView& record) {
const auto applyStarted = std::chrono::steady_clock::now();
applier.ApplyOne(record);
[this, &applier, timed](const Transport::RingRecordView& record) {
// Check after acquiring each record: a producer may publish more work
// while this drain is running, after the outer-loop reset check.
if (m_backend != nullptr && m_backend->LatchIfGpuFaulted()) return false;
if (SessionLatched()) return false;
// P65ServerFrame: timed around the ONE call that is the server's work, and
// summed across the drains a frame spans. See the block above DrainRing's loop.
sFrameApplyNs += static_cast<std::uint64_t>(
std::chrono::duration_cast<std::chrono::nanoseconds>(
std::chrono::steady_clock::now() - applyStarted)
.count());
// Only when the line is on: two clock reads per record were a measurable part
// of a record's cost.
if (timed) {
const auto applyStarted = std::chrono::steady_clock::now();
applier.ApplyOne(record);
sFrameApplyNs += static_cast<std::uint64_t>(
std::chrono::duration_cast<std::chrono::nanoseconds>(
std::chrono::steady_clock::now() - applyStarted)
.count());
} else {
applier.ApplyOne(record);
}
++sFrameRecords;
// P9 W3, F2: outside the apply timing above; 0 in every run that is not F2.
if (m_testLogFloodLines != 0) FloodLogForTest(m_testLogFloodLines);
@@ -1155,6 +1166,7 @@ namespace MobileGL::MG_Remote::Server {
// DrainedRecords() one behind appliedSeq on 1 run in ~40 under `-j 8`.
// Per-record, before the publish, it can never be behind.
m_drained.fetch_add(1, std::memory_order_acq_rel);
return true;
},
&corrupt);
if (corrupt) {
+10
View File
@@ -663,6 +663,16 @@ int RunSession(std::unique_ptr<SocketTransport> control, std::vector<std::uint8_
const char* forwarding = std::getenv("MOBILEGL_IPC_LOG_FORWARD");
if (tcp && (!forwarding || std::strcmp(forwarding, "0") != 0)) {
// F2 fault injection: bound the kernel queue as well as the log queue so a
// non-reading peer reliably blocks the sender, regardless of host autotuning.
if (const char* stalled = std::getenv("MOBILEGL_TEST_LOG_SMALL_SNDBUF");
stalled && std::strcmp(stalled, "1") == 0) {
const int bytes = 16 * 1024;
if (::setsockopt(control->StreamFd(), SOL_SOCKET, SO_SNDBUF, &bytes, sizeof(bytes)) != 0) {
WireLogError("MG_Remote server: could not set F2 control send buffer");
return endSession(70);
}
}
// P9 W3: the forward is a queue with a sender thread of its own, graded by severity
// (LogForward.h): a client that stops reading control no longer holds any logging thread -
// the apply thread above all - inside a socket write.
+382 -31
View File
@@ -19,10 +19,13 @@
#if defined(__ANDROID__)
#include <android/hardware_buffer.h>
#include <dlfcn.h>
#endif
#if !defined(_WIN32)
#include <fcntl.h>
#include <linux/dma-buf.h>
#include <poll.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <sys/socket.h>
#include <sys/stat.h>
@@ -43,6 +46,8 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
std::atomic<Uint64> nextId{1};
UnorderedMap<Uint64, std::weak_ptr<const Image>> byId;
std::map<std::pair<Uint64, Uint64>, std::weak_ptr<const Image>> byIdentity;
// Foreign YUV images, by the identity of the buffer they copy from.
std::map<std::pair<Uint64, Uint64>, std::weak_ptr<const Image>> foreignByIdentity;
// Set once two live images were seen with one identity: the kernel shares an inode
// between dma-bufs, so a descriptor can no longer say which image it is.
Bool identityAmbiguous = false;
@@ -107,14 +112,93 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
return true;
}
// The two AHardwareBuffer entry points newer than this library's minimum platform (29),
// resolved at run time. Null on an older platform: YUV images are then not allocatable.
struct AhbPlaneEntryPoints {
using LockPlanesFn = int (*)(AHardwareBuffer*, uint64_t, int32_t, const ARect*, AHardwareBuffer_Planes*);
using IsSupportedFn = int (*)(const AHardwareBuffer_Desc*);
LockPlanesFn LockPlanes = nullptr;
IsSupportedFn IsSupported = nullptr;
};
const AhbPlaneEntryPoints& AhbPlanes() {
static const AhbPlaneEntryPoints entry = [] {
AhbPlaneEntryPoints e;
void* lib = dlopen("libnativewindow.so", RTLD_NOW | RTLD_NOLOAD);
if (lib == nullptr) lib = dlopen("libnativewindow.so", RTLD_NOW);
if (lib != nullptr) {
e.LockPlanes =
reinterpret_cast<AhbPlaneEntryPoints::LockPlanesFn>(dlsym(lib, "AHardwareBuffer_lockPlanes"));
e.IsSupported =
reinterpret_cast<AhbPlaneEntryPoints::IsSupportedFn>(dlsym(lib, "AHardwareBuffer_isSupported"));
}
return e;
}();
return entry;
}
// A YUV image's buffer: sampled by the GPU, written by the CPU (a foreign image's copy). The
// CPU usage also keeps the layout one the platform can lock plane by plane, which is how
// every layout question about it is answered (AHardwareBuffer_lockPlanes) - never by
// assuming one.
constexpr uint64_t kYuvUsage = AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE |
AHARDWAREBUFFER_USAGE_CPU_WRITE_OFTEN | AHARDWAREBUFFER_USAGE_CPU_READ_RARELY;
uint32_t AhbFormatOf(Uint32 fourcc) {
if (fourcc == kFourccNv12) return AHARDWAREBUFFER_FORMAT_Y8Cb8Cr8_420;
if (fourcc == kFourccP010) return AHARDWAREBUFFER_FORMAT_YCbCr_P010;
return FourccIgnoresAlpha(fourcc) ? AHARDWAREBUFFER_FORMAT_R8G8B8X8_UNORM
: AHARDWAREBUFFER_FORMAT_R8G8B8A8_UNORM;
}
Bool YuvAllocatableNative(Uint32 fourcc) {
const AhbPlaneEntryPoints& entry = AhbPlanes();
if (entry.LockPlanes == nullptr || entry.IsSupported == nullptr) return false;
AHardwareBuffer_Desc desc{};
desc.width = 64;
desc.height = 64;
desc.layers = 1;
desc.format = AhbFormatOf(fourcc);
desc.usage = kYuvUsage;
return entry.IsSupported(&desc) == 1;
}
// The planes of a locked YUV buffer as the plane copy addresses them.
Bool LockYuvPlanes(AHardwareBuffer* ahb, Yuv::PlanarLayout& out, std::string& why) {
const AhbPlaneEntryPoints& entry = AhbPlanes();
if (entry.LockPlanes == nullptr) {
why = "AHardwareBuffer_lockPlanes is not available";
return false;
}
AHardwareBuffer_Planes planes{};
const int locked = entry.LockPlanes(ahb, AHARDWAREBUFFER_USAGE_CPU_WRITE_OFTEN, -1, nullptr, &planes);
if (locked != 0) {
why = Format("AHardwareBuffer_lockPlanes rc=%lld", locked);
return false;
}
if (planes.planeCount < 3) {
AHardwareBuffer_unlock(ahb, nullptr);
why = Format("the locked buffer has %lld planes, not 3", planes.planeCount);
return false;
}
const auto view = [](const AHardwareBuffer_Plane& plane) {
return Yuv::PlaneView{static_cast<Uint8*>(plane.data), static_cast<Int64>(plane.rowStride),
plane.pixelStride};
};
out.Y = view(planes.planes[0]);
out.Cb = view(planes.planes[1]);
out.Cr = view(planes.planes[2]);
return true;
}
Bool AllocateNative(Image& image, std::string& why) {
AHardwareBuffer_Desc desc{};
desc.width = image.Width;
desc.height = image.Height;
desc.layers = 1;
desc.format = FourccIgnoresAlpha(image.Fourcc) ? AHARDWAREBUFFER_FORMAT_R8G8B8X8_UNORM
: AHARDWAREBUFFER_FORMAT_R8G8B8A8_UNORM;
desc.usage = AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE | AHARDWAREBUFFER_USAGE_GPU_COLOR_OUTPUT;
desc.format = AhbFormatOf(image.Fourcc);
desc.usage = FourccIsYuv(image.Fourcc)
? kYuvUsage
: AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE | AHARDWAREBUFFER_USAGE_GPU_COLOR_OUTPUT;
AHardwareBuffer* ahb = nullptr;
const int allocated = AHardwareBuffer_allocate(&desc, &ahb);
if (allocated != 0 || ahb == nullptr) {
@@ -131,8 +215,32 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
image.Native = ahb;
image.Fd = fds[0];
for (SizeT i = 1; i < fds.size(); ++i) ::close(fds[i]);
image.Stride = described.stride * 4;
image.Offset = 0;
if (!FourccIsYuv(image.Fourcc)) {
image.Stride = described.stride * 4;
return true;
}
// Plane positions as the platform reports them when the buffer is locked: offsets from
// the luma plane's start, the Cb Cr plane at whichever of Cb and Cr comes first.
const Uint32 sampleBytes = FourccYuvSampleBytes(image.Fourcc);
image.Stride = described.stride * sampleBytes;
image.Plane1Stride = image.Stride;
image.Plane1Offset = image.Stride * image.Height;
Yuv::PlanarLayout planes;
std::string lockWhy;
if (LockYuvPlanes(ahb, planes, lockWhy)) {
const Uint8* chroma = std::min(planes.Cb.Data, planes.Cr.Data);
if (planes.Y.RowStride > 0 && chroma > planes.Y.Data) {
image.Stride = static_cast<Uint32>(planes.Y.RowStride);
image.Plane1Stride = static_cast<Uint32>(planes.Cb.RowStride);
image.Plane1Offset = static_cast<Uint32>(chroma - planes.Y.Data);
}
AHardwareBuffer_unlock(ahb, nullptr);
} else {
MGLOG_W_ONCE("MG_Remote server: a YUV shared image's planes could not be located (%s); its "
"plane offsets are nominal",
lockWhy.c_str());
}
return true;
}
@@ -143,15 +251,29 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
// A host has no AHardwareBuffer. A memfd stands in for the dma-buf so the registry, the
// descriptor round trip and identification can be exercised by the unit tests; no backend
// binds one.
Bool YuvAllocatableNative(Uint32) { return true; }
Bool AllocateNative(Image& image, std::string& why) {
image.Stride = image.Width * 4;
Uint64 bytes = 0;
image.Offset = 0;
if (FourccIsYuv(image.Fourcc)) {
// A tight semi-planar layout stands in for the platform's.
const Uint32 sampleBytes = FourccYuvSampleBytes(image.Fourcc);
image.Stride = image.Width * sampleBytes;
image.Plane1Stride = ((image.Width + 1) / 2) * 2 * sampleBytes;
image.Plane1Offset = image.Stride * image.Height;
bytes = static_cast<Uint64>(image.Plane1Offset) +
static_cast<Uint64>(image.Plane1Stride) * ((image.Height + 1) / 2);
} else {
image.Stride = image.Width * 4;
bytes = static_cast<Uint64>(image.Stride) * image.Height;
}
const int fd = ::memfd_create("mobilegl-shared-image", MFD_CLOEXEC);
if (fd < 0) {
why = Format("memfd_create errno=%lld", errno);
return false;
}
if (::ftruncate(fd, static_cast<off_t>(image.Stride) * image.Height) != 0) {
if (::ftruncate(fd, static_cast<off_t>(bytes)) != 0) {
why = Format("ftruncate errno=%lld", errno);
::close(fd);
return false;
@@ -162,6 +284,8 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
void ReleaseNative(void*) {}
#else
Bool YuvAllocatableNative(Uint32) { return false; }
Bool AllocateNative(Image&, std::string& why) {
why = "shared images need POSIX descriptors";
return false;
@@ -169,6 +293,66 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
void ReleaseNative(void*) {}
#endif
#if !defined(_WIN32)
// The planes of a YUV image as the CPU writes them, held until the guard goes: the locked
// AHardwareBuffer on Android, a mapping of the stand-in memfd on a host.
class YuvWriteAccess {
public:
YuvWriteAccess(const Image& image, std::string& why) : m_image(image) {
#if defined(__ANDROID__)
auto* ahb = static_cast<AHardwareBuffer*>(image.Native);
m_ok = ahb != nullptr && LockYuvPlanes(ahb, m_planes, why);
#else
const SizeT bytes = static_cast<SizeT>(image.Plane1Offset) +
static_cast<SizeT>(image.Plane1Stride) * ((image.Height + 1) / 2);
void* map = ::mmap(nullptr, bytes, PROT_READ | PROT_WRITE, MAP_SHARED, image.Fd, 0);
if (map == MAP_FAILED) {
why = Format("mmap of the stand-in buffer errno=%lld", errno);
return;
}
m_map = map;
m_mapSize = bytes;
const Uint32 sampleBytes = FourccYuvSampleBytes(image.Fourcc);
auto* base = static_cast<Uint8*>(map);
m_planes.Y = {base, image.Stride, sampleBytes};
m_planes.Cb = {base + image.Plane1Offset, image.Plane1Stride, 2 * sampleBytes};
m_planes.Cr = {base + image.Plane1Offset + sampleBytes, image.Plane1Stride, 2 * sampleBytes};
m_ok = true;
#endif
}
~YuvWriteAccess() {
#if defined(__ANDROID__)
if (m_ok) AHardwareBuffer_unlock(static_cast<AHardwareBuffer*>(m_image.Native), nullptr);
#else
if (m_map != nullptr) ::munmap(m_map, m_mapSize);
#endif
}
YuvWriteAccess(const YuvWriteAccess&) = delete;
YuvWriteAccess& operator=(const YuvWriteAccess&) = delete;
Bool Ok() const { return m_ok; }
const Yuv::PlanarLayout& Planes() const { return m_planes; }
private:
const Image& m_image;
Yuv::PlanarLayout m_planes;
Bool m_ok = false;
#if !defined(__ANDROID__)
void* m_map = nullptr;
SizeT m_mapSize = 0;
#endif
};
// DMA_BUF_IOCTL_SYNC around a CPU read of a foreign buffer: the producer's CPU-side writes
// are made visible (cache maintenance) and its implicit fences waited for. A buffer that
// does not take the ioctl (not a dma-buf: a host test's memfd) is read as it is.
void DmaBufSync(int fd, Bool start) {
dma_buf_sync sync{};
sync.flags = DMA_BUF_SYNC_READ | (start ? DMA_BUF_SYNC_START : DMA_BUF_SYNC_END);
while (::ioctl(fd, DMA_BUF_IOCTL_SYNC, &sync) != 0 && (errno == EINTR || errno == EAGAIN)) {
}
}
#endif
} // namespace
Image::~Image() {
@@ -181,8 +365,18 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
const auto key = std::make_pair(IdentityDev, IdentityIno);
if (auto it = reg.byIdentity.find(key); it != reg.byIdentity.end() && it->second.expired())
reg.byIdentity.erase(it);
if (Foreign != nullptr) {
const auto source = std::make_pair(Foreign->IdentityDev, Foreign->IdentityIno);
if (auto it = reg.foreignByIdentity.find(source);
it != reg.foreignByIdentity.end() && it->second.expired())
reg.foreignByIdentity.erase(it);
}
}
#if !defined(_WIN32)
if (Foreign != nullptr) {
if (Foreign->Map != nullptr) ::munmap(Foreign->Map, Foreign->MapSize);
if (Foreign->Fd >= 0) ::close(Foreign->Fd);
}
if (Fd >= 0) ::close(Fd);
if (WriteFence >= 0) ::close(WriteFence);
for (const int fence : ReadFences) ::close(fence);
@@ -190,49 +384,206 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
ReleaseNative(Native);
}
Bool FourccIsYuv(Uint32 fourcc) { return fourcc == kFourccNv12 || fourcc == kFourccP010; }
Uint32 FourccYuvSampleBytes(Uint32 fourcc) {
if (fourcc == kFourccNv12) return 1;
if (fourcc == kFourccP010) return 2;
return 0;
}
Bool YuvFourccAllocatable(Uint32 fourcc) {
if (!FourccIsYuv(fourcc)) return false;
static std::mutex mutex;
static std::map<Uint32, Bool> answers;
const std::lock_guard<std::mutex> lock(mutex);
const auto it = answers.find(fourcc);
if (it != answers.end()) return it->second;
const Bool answer = YuvAllocatableNative(fourcc);
answers.emplace(fourcc, answer);
return answer;
}
Bool FourccSupported(Uint32 fourcc) {
return fourcc == kFourccAbgr8888 || fourcc == kFourccXbgr8888 || fourcc == kFourccArgb8888 ||
fourcc == kFourccXrgb8888;
fourcc == kFourccXrgb8888 || YuvFourccAllocatable(fourcc);
}
Bool FourccIgnoresAlpha(Uint32 fourcc) { return fourcc == kFourccXbgr8888 || fourcc == kFourccXrgb8888; }
namespace {
// Allocate's body. `foreign`, when given, is on the image before anything can find it.
ImageRef AllocateImage(Uint32 width, Uint32 height, Uint32 fourcc, UniquePtr<Image::ForeignSource> foreign,
std::string& why) {
if (width == 0 || height == 0 || width > 16384 || height > 16384) {
why = Format("%lldx%lld is not an image size", width, height);
return nullptr;
}
if (!FourccSupported(fourcc)) {
why = Format("fourcc 0x%08llx is not a shared-image format", fourcc);
return nullptr;
}
Registry& reg = Reg();
auto image = std::make_shared<Image>();
image->Id = reg.nextId.fetch_add(1, std::memory_order_relaxed);
image->Width = width;
image->Height = height;
image->Fourcc = fourcc;
image->Foreign = std::move(foreign);
if (!AllocateNative(*image, why)) return nullptr;
#if !defined(_WIN32)
if (!IdentityOf(image->Fd, &image->IdentityDev, &image->IdentityIno)) {
why = Format("fstat on the exported descriptor errno=%lld", errno);
return nullptr;
}
#endif
const std::lock_guard<std::mutex> lock(reg.mutex);
reg.byId[image->Id] = image;
const auto key = std::make_pair(image->IdentityDev, image->IdentityIno);
if (auto it = reg.byIdentity.find(key); it != reg.byIdentity.end() && !it->second.expired()) {
if (!reg.identityAmbiguous) {
MGLOG_E("MG_Remote server: two live shared images have one dma-buf identity (dev %llu ino %llu); "
"this kernel does not keep dma-buf inodes unique, so a returned descriptor cannot name its "
"image and every shared-image import is refused from here on",
static_cast<unsigned long long>(key.first), static_cast<unsigned long long>(key.second));
}
reg.identityAmbiguous = true;
}
reg.byIdentity[key] = image;
if (image->Foreign != nullptr) {
reg.foreignByIdentity[std::make_pair(image->Foreign->IdentityDev, image->Foreign->IdentityIno)] =
image;
}
return image;
}
} // namespace
ImageRef Allocate(Uint32 width, Uint32 height, Uint32 fourcc, std::string& why) {
return AllocateImage(width, height, fourcc, nullptr, why);
}
ImageRef ImportForeignYuv(int fd, Uint32 width, Uint32 height, Uint32 fourcc, const PlaneLayout& layout,
std::string& why) {
#if defined(_WIN32)
(void)fd, (void)width, (void)height, (void)fourcc, (void)layout;
why = "shared images need POSIX descriptors";
return nullptr;
#else
if (!FourccIsYuv(fourcc) || !YuvFourccAllocatable(fourcc)) {
why = Format("fourcc 0x%08llx is not a YUV format this server can hold", fourcc);
return nullptr;
}
if (width == 0 || height == 0 || width > 16384 || height > 16384) {
why = Format("%lldx%lld is not an image size", width, height);
return nullptr;
}
if (!FourccSupported(fourcc)) {
why = Format("fourcc 0x%08llx is not a shared-image format", fourcc);
if (layout.Count != 2) {
why = Format("a YUV import needs 2 planes, not %lld", layout.Count);
return nullptr;
}
Uint64 dev = 0, ino = 0;
if (!IdentityOf(fd, &dev, &ino)) {
why = Format("fstat on the imported descriptor errno=%lld", errno);
return nullptr;
}
Registry& reg = Reg();
auto image = std::make_shared<Image>();
image->Id = reg.nextId.fetch_add(1, std::memory_order_relaxed);
image->Width = width;
image->Height = height;
image->Fourcc = fourcc;
if (!AllocateNative(*image, why)) return nullptr;
#if !defined(_WIN32)
if (!IdentityOf(image->Fd, &image->IdentityDev, &image->IdentityIno)) {
why = Format("fstat on the exported descriptor errno=%lld", errno);
{
const std::lock_guard<std::mutex> lock(reg.mutex);
const auto it = reg.foreignByIdentity.find(std::make_pair(dev, ino));
ImageRef existing = it != reg.foreignByIdentity.end() ? it->second.lock() : nullptr;
if (existing != nullptr && existing->Width == width && existing->Height == height &&
existing->Fourcc == fourcc && existing->Foreign->Layout.Offset[0] == layout.Offset[0] &&
existing->Foreign->Layout.Offset[1] == layout.Offset[1] &&
existing->Foreign->Layout.Pitch[0] == layout.Pitch[0] &&
existing->Foreign->Layout.Pitch[1] == layout.Pitch[1])
return existing;
}
// The planes must fit the buffer: every row the copy reads is inside the mapping.
const Uint32 sampleBytes = FourccYuvSampleBytes(fourcc);
const off_t end = ::lseek(fd, 0, SEEK_END);
if (end <= 0) {
why = Format("the buffer's size is unknown (lseek errno=%lld)", errno);
return nullptr;
}
#endif
const std::lock_guard<std::mutex> lock(reg.mutex);
reg.byId[image->Id] = image;
const auto key = std::make_pair(image->IdentityDev, image->IdentityIno);
if (auto it = reg.byIdentity.find(key); it != reg.byIdentity.end() && !it->second.expired()) {
if (!reg.identityAmbiguous) {
MGLOG_E("MG_Remote server: two live shared images have one dma-buf identity (dev %llu ino %llu); "
"this kernel does not keep dma-buf inodes unique, so a returned descriptor cannot name its "
"image and every shared-image import is refused from here on",
static_cast<unsigned long long>(key.first), static_cast<unsigned long long>(key.second));
}
reg.identityAmbiguous = true;
const SizeT size = static_cast<SizeT>(end);
const Uint32 chromaHeight = (height + 1) / 2;
const Uint64 chromaRow = static_cast<Uint64>((width + 1) / 2) * 2 * sampleBytes;
const Uint64 lumaEnd = static_cast<Uint64>(layout.Offset[0]) +
static_cast<Uint64>(layout.Pitch[0]) * (height - 1) +
static_cast<Uint64>(width) * sampleBytes;
const Uint64 chromaEnd =
static_cast<Uint64>(layout.Offset[1]) + static_cast<Uint64>(layout.Pitch[1]) * (chromaHeight - 1) + chromaRow;
if (layout.Pitch[0] < width * sampleBytes || layout.Pitch[1] < chromaRow || lumaEnd > size ||
chromaEnd > size) {
why = Format("its planes do not fit the %lld-byte buffer", static_cast<long long>(size));
return nullptr;
}
reg.byIdentity[key] = image;
void* map = ::mmap(nullptr, size, PROT_READ, MAP_SHARED, fd, 0);
if (map == MAP_FAILED) {
why = Format("the buffer cannot be mapped for reading (errno=%lld)", errno);
return nullptr;
}
auto foreign = MakeUnique<Image::ForeignSource>();
foreign->Fd = ::fcntl(fd, F_DUPFD_CLOEXEC, 0);
foreign->Layout = layout;
foreign->Map = map;
foreign->MapSize = size;
foreign->IdentityDev = dev;
foreign->IdentityIno = ino;
if (foreign->Fd < 0) {
::munmap(map, size);
why = Format("dup errno=%lld", errno);
return nullptr;
}
// A failed allocation drops the half-built image, and ~Image unmaps and closes the source.
ImageRef image = AllocateImage(width, height, fourcc, std::move(foreign), why);
if (image == nullptr) return nullptr;
MGLOG_I("MG_Remote server: a foreign %ux%u %s dma-buf is sampled through image %llu, refreshed by a CPU "
"copy (the fallback for a buffer this platform cannot import)",
width, height, fourcc == kFourccNv12 ? "NV12" : "P010", static_cast<unsigned long long>(image->Id));
return image;
#endif
}
Bool RefreshForeign(const Image& image, std::string& why) {
#if defined(_WIN32)
(void)image;
why = "shared images need POSIX descriptors";
return false;
#else
if (image.Foreign == nullptr) return true;
Image::ForeignSource& source = *image.Foreign;
// WRITE AFTER READ: what a frame still samples of the old content finishes first. The copy
// is the CPU's, so this is a CPU wait - bounded, like every wait on another session.
if (!WaitForReads(image, 1000)) {
MGLOG_W_ONCE("MG_Remote server: a foreign image's readers did not finish within 1 s; refreshed anyway");
}
const std::lock_guard<std::mutex> lock(source.Mutex);
YuvWriteAccess access(image, why);
if (!access.Ok()) return false;
const auto* base = static_cast<const Uint8*>(source.Map);
DmaBufSync(source.Fd, true);
const Bool copied = Yuv::CopySemiPlanar(base + source.Layout.Offset[0], source.Layout.Pitch[0],
base + source.Layout.Offset[1], source.Layout.Pitch[1], image.Width,
image.Height, FourccYuvSampleBytes(image.Fourcc), access.Planes());
DmaBufSync(source.Fd, false);
if (!copied) {
why = "the locked buffer's planes cannot hold the samples";
return false;
}
++source.Copies;
return true;
#endif
}
void SetYuvHints(const Image& image, const Yuv::Hints& hints) {
const std::lock_guard<std::mutex> lock(image.SyncMutex);
image.YuvHints = hints;
}
Yuv::Hints GetYuvHints(const Image& image) {
const std::lock_guard<std::mutex> lock(image.SyncMutex);
return image.YuvHints;
}
ImageRef Identify(int fd, std::string& why) {
@@ -25,6 +25,8 @@
#pragma once
#include <Includes.h>
#include "SharedImageYuv.h"
#include <mutex>
#include <string>
@@ -44,6 +46,11 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
// client renders as red samples as red, whichever of the four names it allocated under.
inline constexpr Uint32 kFourccArgb8888 = FourCC('A', 'R', '2', '4');
inline constexpr Uint32 kFourccXrgb8888 = FourCC('X', 'R', '2', '4');
// YUV 4:2:0, two planes: Y, then Cb Cr interleaved at half resolution (8-bit samples; P010 the
// same with 10 significant bits in the high end of 16). A YUV image is never rendered to: it is
// SAMPLED, converted into RGBA by whoever binds it (SharedImageYuv.h).
inline constexpr Uint32 kFourccNv12 = FourCC('N', 'V', '1', '2');
inline constexpr Uint32 kFourccP010 = FourCC('P', '0', '1', '0');
// "The layout is the allocator's": what an AHardwareBuffer is to anybody but its allocator.
inline constexpr Uint64 kModifierInvalid = 0x00ffffffffffffffull;
inline constexpr Uint64 kModifierLinear = 0;
@@ -52,6 +59,20 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
Bool FourccSupported(Uint32 fourcc);
// The X formats: alpha reads as 1 whatever is stored.
Bool FourccIgnoresAlpha(Uint32 fourcc);
// NV12 / P010.
Bool FourccIsYuv(Uint32 fourcc);
// Bytes of one sample of a YUV fourcc (1, or 2 for P010); 0 for anything else.
Uint32 FourccYuvSampleBytes(Uint32 fourcc);
// Whether this server can allocate images of a YUV fourcc (the platform's allocator takes the
// format for sampling). Asked once per format, cached.
Bool YuvFourccAllocatable(Uint32 fourcc);
// Where an imported dma-buf's planes are: byte offset and row pitch of each.
struct PlaneLayout {
Uint32 Count = 1;
Uint32 Offset[2] = {0, 0};
Uint32 Pitch[2] = {0, 0};
};
struct Image {
Uint64 Id = 0;
@@ -61,6 +82,11 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
Uint32 Stride = 0; // bytes
Uint32 Offset = 0;
Uint64 Modifier = kModifierInvalid;
// A YUV image's second plane (Cb Cr): same descriptor, this offset and pitch. The layout is
// still the allocator's - these are what it reported when the buffer was locked, so a
// client can name the plane, not a promise that the descriptor maps that way.
Uint32 Plane1Stride = 0;
Uint32 Plane1Offset = 0;
// AHardwareBuffer* on Android; null on a host, where the memory is a memfd (unit tests).
void* Native = nullptr;
// The descriptor clients receive duplicates of. Owned.
@@ -77,6 +103,31 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
mutable Uint64 WriteGeneration = 0;
mutable Vector<int> ReadFences;
// FOREIGN YUV - THE CPU-COPY FALLBACK. A YUV dma-buf this server did not allocate (a video
// decoder's frame) cannot be imported vendor-neutrally: neither the platform's EGL nor its
// Vulkan takes a raw dma-buf. Such an import becomes a YUV image of this server's (Native,
// an AHardwareBuffer the CPU can write) whose content is COPIED from a CPU mapping of the
// foreign buffer (RefreshForeign) at a reader's first use of it in each of its frames - the
// moments the producer's content can have moved for that reader. Every other image is
// zero-copy; this one costs a frame's worth of memcpy per refresh, and its log says so.
struct ForeignSource {
int Fd = -1; // a duplicate, owned
PlaneLayout Layout;
void* Map = nullptr;
SizeT MapSize = 0;
std::mutex Mutex; // one refresh at a time
Uint64 Copies = 0;
// The foreign buffer's identity, under which a later import of it finds this image.
Uint64 IdentityDev = 0;
Uint64 IdentityIno = 0;
};
UniquePtr<ForeignSource> Foreign;
Bool IsForeign() const { return Foreign != nullptr; }
// How the samples of a YUV image convert to RGB: the EGL_EXT_image_dma_buf_import hints
// of its last import. Guarded by SyncMutex.
mutable Yuv::Hints YuvHints;
Image() = default;
Image(const Image&) = delete;
Image& operator=(const Image&) = delete;
@@ -91,6 +142,22 @@ namespace MobileGL::MG_Remote::Server::SharedImages {
// this server never exported, or one whose identity the kernel does not keep unique.
ImageRef Identify(int fd, std::string& why);
// A YUV import of a dma-buf this server did not export (Image::ForeignSource): the live
// foreign image already made from the same buffer with the same shape, or a new one, which
// keeps a duplicate of `fd`. Null with `why` when the buffer cannot be mapped, its planes do
// not fit in it, or no YUV image can be allocated to copy it into.
ImageRef ImportForeignYuv(int fd, Uint32 width, Uint32 height, Uint32 fourcc, const PlaneLayout& layout,
std::string& why);
// A foreign image's content brought up to date from its source (the copy). Waits for the
// image's pending reads first, so a frame still sampling the old content keeps it. True for an
// image that is not foreign: there is nothing to do.
Bool RefreshForeign(const Image& image, std::string& why);
// The colour hints a YUV image is converted with.
void SetYuvHints(const Image& image, const Yuv::Hints& hints);
Yuv::Hints GetYuvHints(const Image& image);
// The live image with this id, or null.
ImageRef Find(Uint64 id);
+158
View File
@@ -0,0 +1,158 @@
// MobileGL - MobileGL/MG_Remote/Server/SharedImageYuv.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
// YUV SHARED IMAGES: the pure parts - plane copies and the YCbCr -> RGB conversion an image's
// colour hints select - kept free of Android and of any backend so the host tests can check them.
//
// A YUV image is sampled as RGBA: whoever binds it to a texture gets the image CONVERTED into the
// texture's own RGBA8 storage at its first use in each of the reader's frames (the backends do
// that on the GPU, from the image's AHardwareBuffer). The conversion is the one the EGLImage's
// EGL_EXT_image_dma_buf_import hints name - colour space, sample range and chroma siting - with
// that extension's defaults (ITU-R BT.601, narrow range, siting 0) where a hint is absent.
#pragma once
#include <Includes.h>
#include <cstring>
namespace MobileGL::MG_Remote::Server::SharedImages::Yuv {
// EGL_EXT_image_dma_buf_import's hint values.
inline constexpr Uint32 kEglItuRec601 = 0x327F;
inline constexpr Uint32 kEglItuRec709 = 0x3280;
inline constexpr Uint32 kEglItuRec2020 = 0x3281;
inline constexpr Uint32 kEglYuvFullRange = 0x3282;
inline constexpr Uint32 kEglYuvNarrowRange = 0x3283;
inline constexpr Uint32 kEglYuvChromaSiting0 = 0x3284;
inline constexpr Uint32 kEglYuvChromaSiting0_5 = 0x3285;
// What an import said about its samples. 0 = not given (the extension's default applies).
struct Hints {
Uint32 ColorSpace = 0;
Uint32 Range = 0;
Uint32 SitingX = 0;
Uint32 SitingY = 0;
Uint32 EffectiveColorSpace() const {
return ColorSpace == kEglItuRec709 || ColorSpace == kEglItuRec2020 ? ColorSpace : kEglItuRec601;
}
Bool FullRange() const { return Range == kEglYuvFullRange; }
Bool MidpointX() const { return SitingX == kEglYuvChromaSiting0_5; }
Bool MidpointY() const { return SitingY == kEglYuvChromaSiting0_5; }
Bool operator==(const Hints& other) const {
return ColorSpace == other.ColorSpace && Range == other.Range && SitingX == other.SitingX &&
SitingY == other.SitingY;
}
};
// rgb = Matrix * (ycbcr - Offset), every value normalized to [0, 1] as a UNORM texture of the
// image's sample size returns it. Matrix is COLUMN-major (a GLSL mat3 uniform as is).
struct Conversion {
float Matrix[9];
float Offset[3];
};
// `sampleMax` is what a full-scale code reads as before normalization: 255 for 8-bit samples,
// 65535/64 for P010 (10 bits in the high end of 16, read as UNORM16).
inline Conversion ConversionFor(const Hints& hints, double sampleMax) {
double kr = 0.299, kb = 0.114;
if (hints.EffectiveColorSpace() == kEglItuRec709) {
kr = 0.2126;
kb = 0.0722;
} else if (hints.EffectiveColorSpace() == kEglItuRec2020) {
kr = 0.2627;
kb = 0.0593;
}
const double kg = 1.0 - kr - kb;
// Code values of an 8-bit-equivalent scale, multiplied up for deeper samples.
const double scale = sampleMax / 255.0;
double yOffset = 0.0, yScale = 1.0, cScale = 1.0;
const double cOffset = 128.0 * scale / sampleMax;
if (!hints.FullRange()) {
yOffset = 16.0 * scale / sampleMax;
yScale = sampleMax / (219.0 * scale);
cScale = sampleMax / (224.0 * scale);
}
Conversion c{};
// Columns: Y, Cb, Cr.
const double m[9] = {
yScale,
yScale,
yScale,
0.0,
-cScale * 2.0 * kb * (1.0 - kb) / kg,
cScale * 2.0 * (1.0 - kb),
cScale * 2.0 * (1.0 - kr),
-cScale * 2.0 * kr * (1.0 - kr) / kg,
0.0,
};
for (SizeT i = 0; i < 9; ++i) c.Matrix[i] = static_cast<float>(m[i]);
c.Offset[0] = static_cast<float>(yOffset);
c.Offset[1] = static_cast<float>(cOffset);
c.Offset[2] = static_cast<float>(cOffset);
return c;
}
// One component plane of a destination: where (x, y)'s sample of it lives is
// Data + y * RowStride + x * PixelStride.
struct PlaneView {
Uint8* Data = nullptr;
Int64 RowStride = 0;
Uint32 PixelStride = 0;
};
struct PlanarLayout {
PlaneView Y, Cb, Cr;
};
// A two-plane semi-planar source (NV12: Y, then Cb Cr interleaved at half resolution; P010
// the same with 16-bit samples, `sampleBytes` = 2) into any destination layout: semi-planar
// Cb-first rows are copied as they are, everything else (Cr first, fully planar) sample by
// sample. False when the destination cannot hold the samples (null planes, a pixel stride
// smaller than a sample).
inline Bool CopySemiPlanar(const Uint8* srcY, Int64 srcYPitch, const Uint8* srcCbCr, Int64 srcCbCrPitch,
Uint32 width, Uint32 height, Uint32 sampleBytes, const PlanarLayout& dst) {
if (srcY == nullptr || srcCbCr == nullptr || dst.Y.Data == nullptr || dst.Cb.Data == nullptr ||
dst.Cr.Data == nullptr || (sampleBytes != 1 && sampleBytes != 2) || dst.Y.PixelStride < sampleBytes ||
dst.Cb.PixelStride < sampleBytes || dst.Cr.PixelStride < sampleBytes)
return false;
const Uint32 chromaWidth = (width + 1) / 2;
const Uint32 chromaHeight = (height + 1) / 2;
// Luma.
for (Uint32 y = 0; y < height; ++y) {
const Uint8* from = srcY + static_cast<Int64>(y) * srcYPitch;
Uint8* to = dst.Y.Data + static_cast<Int64>(y) * dst.Y.RowStride;
if (dst.Y.PixelStride == sampleBytes) {
std::memcpy(to, from, static_cast<SizeT>(width) * sampleBytes);
} else {
for (Uint32 x = 0; x < width; ++x)
std::memcpy(to + static_cast<SizeT>(x) * dst.Y.PixelStride, from + static_cast<SizeT>(x) * sampleBytes,
sampleBytes);
}
}
// Chroma.
const Bool sameInterleave = dst.Cb.PixelStride == 2 * sampleBytes && dst.Cr.PixelStride == 2 * sampleBytes &&
dst.Cr.Data == dst.Cb.Data + sampleBytes && dst.Cr.RowStride == dst.Cb.RowStride;
for (Uint32 y = 0; y < chromaHeight; ++y) {
const Uint8* from = srcCbCr + static_cast<Int64>(y) * srcCbCrPitch;
if (sameInterleave) {
std::memcpy(dst.Cb.Data + static_cast<Int64>(y) * dst.Cb.RowStride, from,
static_cast<SizeT>(chromaWidth) * 2 * sampleBytes);
continue;
}
Uint8* cb = dst.Cb.Data + static_cast<Int64>(y) * dst.Cb.RowStride;
Uint8* cr = dst.Cr.Data + static_cast<Int64>(y) * dst.Cr.RowStride;
for (Uint32 x = 0; x < chromaWidth; ++x) {
const Uint8* pair = from + static_cast<SizeT>(x) * 2 * sampleBytes;
std::memcpy(cb + static_cast<SizeT>(x) * dst.Cb.PixelStride, pair, sampleBytes);
std::memcpy(cr + static_cast<SizeT>(x) * dst.Cr.PixelStride, pair + sampleBytes, sampleBytes);
}
}
return true;
}
} // namespace MobileGL::MG_Remote::Server::SharedImages::Yuv
+4 -1
View File
@@ -274,7 +274,10 @@ namespace MobileGL::MG_Remote::Transport {
for (;;) {
// Announce, FENCE, then re-test. The fence is the mechanism -
// see the file header - so setting the flag itself is relaxed.
parked.store(1, std::memory_order_relaxed);
// Release, so a notifier that acquires this 1 also sees what the waiter wrote
// before parking (RingControl::producerWakeSeq). The fence below is still what
// closes the lost-wakeup window.
parked.store(1, std::memory_order_release);
std::atomic_thread_fence(std::memory_order_seq_cst);
if (ready()) {
parked.store(0, std::memory_order_relaxed);
+24 -1
View File
@@ -497,11 +497,24 @@ namespace MobileGL::MG_Remote::Transport {
return m_selfBell->Dead() ? SessionWait::ShutDown : SessionWait::TimedOut;
}
// The wake target for the duration of one wait; 0 again afterwards, so a later wait that
// needs any progress (or one outside this class on the same flag) is rung as before.
namespace {
struct ScopedWakeTarget {
std::atomic<std::uint64_t>& target;
ScopedWakeTarget(std::atomic<std::uint64_t>& t, std::uint64_t seq) : target(t) {
target.store(seq, std::memory_order_seq_cst);
}
~ScopedWakeTarget() { target.store(0, std::memory_order_seq_cst); }
};
} // namespace
SessionWait SessionProducer::WaitForApplied(std::uint64_t seq, std::uint32_t timeoutMs) {
if (!Valid()) {
return SessionWait::TimedOut;
}
RingControl* control = m_control;
const ScopedWakeTarget wake(control->producerWakeSeq, seq);
return Park([control, seq] { return Watermark::Reached(control->Progress.appliedSeq, seq); },
timeoutMs);
}
@@ -527,6 +540,7 @@ namespace MobileGL::MG_Remote::Transport {
return SessionWait::TimedOut;
}
RingControl* control = m_control;
const ScopedWakeTarget wake(control->producerWakeSeq, seq);
return Park(
[control, seq] {
return Watermark::Reached(control->Progress.appliedSeq, seq) ||
@@ -670,7 +684,16 @@ namespace MobileGL::MG_Remote::Transport {
void SessionConsumer::NotifyClient() {
if (m_link) m_link->ProgressChanged();
if (m_control != nullptr && m_peerBell != nullptr) {
NotifyIfParked(*m_peerBell, m_control->producerParked);
// NotifyIfParked, narrowed by the client's wake target: a client parked for an
// applied sequence the server has not reached yet is not rung (unless the event
// ring filled, which that wait also wakes for).
std::atomic_thread_fence(std::memory_order_seq_cst);
if (m_control->producerParked.load(std::memory_order_acquire) == 0) return;
const std::uint64_t target = m_control->producerWakeSeq.load(std::memory_order_seq_cst);
if (target != 0 && !Watermark::Reached(m_control->Progress.appliedSeq, target) &&
m_control->eventRingFull.load(std::memory_order_acquire) == 0)
return;
m_peerBell->Notify();
}
}
+14
View File
@@ -164,6 +164,16 @@ namespace MobileGL::MG_Remote::Transport {
std::atomic<std::uint32_t> producerParked; // client asleep, server must ring
std::atomic<std::uint32_t> eventRingFull; // SEG_EVENT full, server stopped applying
std::atomic<std::uint32_t> eventDropped; // dropped lossy events
// ---- the parked producer's wake target -------------------------------
//
// What a parked client is waiting for, so the server rings only when it is there: 0 =
// any progress (present credit, ring space, events), otherwise the appliedSeq it needs.
// Without it every record the server applied while the client waited for a later one
// was a socket write and a futile wake-up. Written by the client before it parks (and
// reset to 0 after), read by the server only after it saw producerParked set; on a link
// without this shared page it stays 0, which is the old ring-on-any-progress rule.
alignas(64) std::atomic<std::uint64_t> producerWakeSeq;
};
static_assert(sizeof(RingControl) == 4096, "RingControl must be exactly one page");
@@ -232,6 +242,10 @@ namespace MobileGL::MG_Remote::Transport {
// The page's real content must fit the page it claims. sizeof(RingControl) == 4096
// above cannot catch a member that grew, because alignas(4096) forces that number;
// the struct's content ends far short of it and the tail is padding.
static_assert(offsetof(RingControl, producerWakeSeq) / 64 != offsetof(RingControl, serverEpoch) / 64,
"the wake target has a line of its own");
static_assert(offsetof(RingControl, producerWakeSeq) + sizeof(std::uint64_t) <= 4096,
"RingControl's members must fit the page it claims to be");
static_assert(offsetof(RingControl, eventDropped) + sizeof(std::uint32_t) <= 4096,
"RingControl's members must fit the page it claims to be");
static_assert(alignof(RingControl) == 4096, "RingControl must be page aligned");
+9 -1
View File
@@ -73,6 +73,7 @@
#include <atomic>
#include <cstdint>
#include <type_traits>
namespace MobileGL::MG_Remote::Transport {
@@ -462,7 +463,14 @@ namespace MobileGL::MG_Remote::Transport {
if (!m_cmd->Pop(view, outCorrupt)) {
return false;
}
apply(view);
// A terminal rejection after Pop must not acknowledge an unapplied record.
// A bool callback returning false ends this consumer's session; it cannot retry
// the popped record. Existing void callbacks retain their original semantics.
if constexpr (std::is_same_v<std::invoke_result_t<Apply, const RingRecordView&>, bool>) {
if (!apply(view)) return false;
} else {
apply(view);
}
++m_appliedSeq;
Watermark::AdvanceApplied(*m_control, m_appliedSeq);
m_cmd->PublishApplied();
+95 -37
View File
@@ -1874,9 +1874,14 @@ namespace MobileGL {
}
Bool EGLContext::IsDmaBufFourccSupported(Uint32 fourcc) {
// The four orders a shared image can be allocated under (SharedImageApi.h).
// The four orders a shared image can be allocated under (SharedImageApi.h), and the
// two YUV ones that are sampled only.
return fourcc == kDrmFourccAbgr8888 || fourcc == kDrmFourccXbgr8888 || fourcc == kDrmFourccArgb8888 ||
fourcc == kDrmFourccXrgb8888;
fourcc == kDrmFourccXrgb8888 || IsDmaBufFourccYuv(fourcc);
}
Bool EGLContext::IsDmaBufFourccYuv(Uint32 fourcc) {
return fourcc == kDrmFourccNv12 || fourcc == kDrmFourccP010;
}
Bool EGLContext::PrepareDmaBufImport(EGLDisplayHandle display, EGLContextHandle context,
@@ -1899,9 +1904,11 @@ namespace MobileGL {
}
DmaBufImportAttribs attribs;
Bool hasWidth = false, hasHeight = false, hasFourcc = false, hasFd = false, hasOffset = false,
hasPitch = false, hasModifierLo = false, hasModifierHi = false, otherPlane = false;
Uint64 modifierLo = 0, modifierHi = 0;
Bool hasWidth = false, hasHeight = false, hasFourcc = false;
// Per plane: fd, offset, pitch, modifier lo, modifier hi.
Bool has[2][5] = {};
Uint64 modifierLo[2] = {0, 0}, modifierHi[2] = {0, 0};
Bool laterPlane = false;
for (SizeT i = 0; attribList != nullptr && attribList[i] != EGL_NONE; i += 2) {
const EGLAttrib value = attribList[i + 1];
switch (attribList[i]) {
@@ -1919,29 +1926,44 @@ namespace MobileGL {
break;
case EGL_DMA_BUF_PLANE0_FD_EXT:
attribs.Fd = static_cast<int>(value);
hasFd = true;
has[0][0] = true;
break;
case EGL_DMA_BUF_PLANE0_OFFSET_EXT:
attribs.Offset = static_cast<EGLint>(value);
hasOffset = true;
has[0][1] = true;
break;
case EGL_DMA_BUF_PLANE0_PITCH_EXT:
attribs.Pitch = static_cast<EGLint>(value);
hasPitch = true;
has[0][2] = true;
break;
case EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT:
modifierLo = static_cast<Uint32>(value);
hasModifierLo = true;
modifierLo[0] = static_cast<Uint32>(value);
has[0][3] = true;
break;
case EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT:
modifierHi = static_cast<Uint32>(value);
hasModifierHi = true;
modifierHi[0] = static_cast<Uint32>(value);
has[0][4] = true;
break;
case EGL_DMA_BUF_PLANE1_FD_EXT:
attribs.Plane1Fd = static_cast<int>(value);
has[1][0] = true;
break;
case EGL_DMA_BUF_PLANE1_OFFSET_EXT:
attribs.Plane1Offset = static_cast<EGLint>(value);
has[1][1] = true;
break;
case EGL_DMA_BUF_PLANE1_PITCH_EXT:
attribs.Plane1Pitch = static_cast<EGLint>(value);
has[1][2] = true;
break;
case EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT:
modifierLo[1] = static_cast<Uint32>(value);
has[1][3] = true;
break;
case EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT:
modifierHi[1] = static_cast<Uint32>(value);
has[1][4] = true;
break;
case EGL_DMA_BUF_PLANE2_FD_EXT:
case EGL_DMA_BUF_PLANE2_OFFSET_EXT:
case EGL_DMA_BUF_PLANE2_PITCH_EXT:
@@ -1952,14 +1974,22 @@ namespace MobileGL {
case EGL_DMA_BUF_PLANE3_PITCH_EXT:
case EGL_DMA_BUF_PLANE3_MODIFIER_LO_EXT:
case EGL_DMA_BUF_PLANE3_MODIFIER_HI_EXT:
otherPlane = true;
laterPlane = true;
break;
// Hints for YUV formats, and EGL_KHR_image_base's own attribute: meaningless
// for a single-plane RGB image, accepted and ignored.
// The YUV hints: kept for a YUV import, which is converted with them; meaningless
// for an RGB one and ignored there. EGL_KHR_image_base's own attribute likewise.
case EGL_YUV_COLOR_SPACE_HINT_EXT:
attribs.ColorSpace = static_cast<Uint32>(value);
break;
case EGL_SAMPLE_RANGE_HINT_EXT:
attribs.Range = static_cast<Uint32>(value);
break;
case EGL_YUV_CHROMA_HORIZONTAL_SITING_HINT_EXT:
attribs.SitingX = static_cast<Uint32>(value);
break;
case EGL_YUV_CHROMA_VERTICAL_SITING_HINT_EXT:
attribs.SitingY = static_cast<Uint32>(value);
break;
case EGL_IMAGE_PRESERVED_KHR:
break;
default:
@@ -1967,8 +1997,8 @@ namespace MobileGL {
return false;
}
}
if (!hasWidth || !hasHeight || !hasFourcc || !hasFd || !hasOffset || !hasPitch ||
hasModifierLo != hasModifierHi) {
if (!hasWidth || !hasHeight || !hasFourcc || !has[0][0] || !has[0][1] || !has[0][2] ||
has[0][3] != has[0][4]) {
SetError(EGL_BAD_PARAMETER);
return false;
}
@@ -1976,23 +2006,37 @@ namespace MobileGL {
SetError(EGL_BAD_MATCH);
return false;
}
// Every supported format has exactly one plane.
if (otherPlane) {
const Bool yuv = IsDmaBufFourccYuv(attribs.Fourcc);
attribs.PlaneCount = yuv ? 2u : 1u;
const Bool anyPlane1 = has[1][0] || has[1][1] || has[1][2] || has[1][3] || has[1][4];
// Planes the format does not have are an attribute error; planes it has but the
// list leaves out (or half a modifier) a parameter error.
if (laterPlane || (!yuv && anyPlane1)) {
SetError(EGL_BAD_ATTRIBUTE);
return false;
}
if (attribs.Width <= 0 || attribs.Height <= 0 || attribs.Fd < 0) {
if (yuv && (!has[1][0] || !has[1][1] || !has[1][2] || has[1][3] != has[1][4] || has[1][3] != has[0][3])) {
SetError(EGL_BAD_PARAMETER);
return false;
}
if (attribs.Offset < 0 || attribs.Pitch <= 0) {
if (attribs.Width <= 0 || attribs.Height <= 0 || attribs.Fd < 0 || (yuv && attribs.Plane1Fd < 0)) {
SetError(EGL_BAD_PARAMETER);
return false;
}
if (attribs.Offset < 0 || attribs.Pitch <= 0 || (yuv && (attribs.Plane1Offset < 0 || attribs.Plane1Pitch <= 0))) {
SetError(EGL_BAD_ACCESS);
return false;
}
// Any modifier is taken: the image is identified by its descriptor, never read
// through the layout a modifier describes.
attribs.HasModifier = hasModifierLo;
attribs.Modifier = (modifierHi << 32) | modifierLo;
// Every plane names the one modifier of the buffer.
if (yuv && has[0][3] && (modifierLo[0] != modifierLo[1] || modifierHi[0] != modifierHi[1])) {
SetError(EGL_BAD_MATCH);
return false;
}
// An RGBA image is identified by its descriptor, never read through the layout a
// modifier describes, so any modifier is taken; a YUV one may be a buffer the server
// copies (a foreign one), which it reads as linear rows - the server decides.
attribs.HasModifier = has[0][3];
attribs.Modifier = (modifierHi[0] << 32) | modifierLo[0];
if (out != nullptr) *out = attribs;
return true;
}
@@ -2034,8 +2078,8 @@ namespace MobileGL {
return orphaned;
}
Bool EGLContext::QueryDmaBufFormats(EGLDisplayHandle display, Bool available, EGLint maxFormats,
EGLint* formats, EGLint* numFormats) {
Bool EGLContext::QueryDmaBufFormats(EGLDisplayHandle display, Bool available, Bool yuvAvailable,
EGLint maxFormats, EGLint* formats, EGLint* numFormats) {
const std::lock_guard<std::recursive_mutex> lock(m_mutex);
if (!ValidateDisplay(display)) {
SetError(EGL_BAD_DISPLAY);
@@ -2050,8 +2094,10 @@ namespace MobileGL {
return false;
}
static constexpr Uint32 kFormats[] = {kDrmFourccAbgr8888, kDrmFourccXbgr8888, kDrmFourccArgb8888,
kDrmFourccXrgb8888};
const EGLint total = available ? static_cast<EGLint>(std::size(kFormats)) : 0;
kDrmFourccXrgb8888, kDrmFourccNv12, kDrmFourccP010};
constexpr EGLint kRgbCount = 4;
const EGLint total =
!available ? 0 : (yuvAvailable ? static_cast<EGLint>(std::size(kFormats)) : kRgbCount);
if (maxFormats == 0) {
*numFormats = total;
return true;
@@ -2062,11 +2108,10 @@ namespace MobileGL {
return true;
}
Bool EGLContext::QueryDmaBufModifiers(EGLDisplayHandle display, Bool available, EGLint format,
EGLint maxModifiers, Uint64* modifiers,
Bool EGLContext::QueryDmaBufModifiers(EGLDisplayHandle display, Bool available, Bool yuvAvailable,
EGLint format, EGLint maxModifiers, Uint64* modifiers,
EGLBoolean* externalOnly, EGLint* numModifiers) {
const std::lock_guard<std::recursive_mutex> lock(m_mutex);
(void)externalOnly;
if (!ValidateDisplay(display)) {
SetError(EGL_BAD_DISPLAY);
return false;
@@ -2075,14 +2120,27 @@ namespace MobileGL {
SetError(EGL_NOT_INITIALIZED);
return false;
}
if (!available || !IsDmaBufFourccSupported(static_cast<Uint32>(format)) || maxModifiers < 0 ||
numModifiers == nullptr || (maxModifiers > 0 && modifiers == nullptr)) {
const Bool yuv = IsDmaBufFourccYuv(static_cast<Uint32>(format));
if (!available || !IsDmaBufFourccSupported(static_cast<Uint32>(format)) || (yuv && !yuvAvailable) ||
maxModifiers < 0 || numModifiers == nullptr || (maxModifiers > 0 && modifiers == nullptr)) {
SetError(EGL_BAD_PARAMETER);
return false;
}
// No explicit modifiers: an image's layout is its allocator's, which is what an
// import without modifier attributes (the implicit modifier) means.
*numModifiers = 0;
if (!yuv) {
// No explicit modifiers: an image's layout is its allocator's, which is what an
// import without modifier attributes (the implicit modifier) means.
*numModifiers = 0;
return true;
}
// A YUV buffer is taken as linear rows when it is not one of the server's own (which
// are identified, whatever modifier is named), and only ever sampled: external only.
if (maxModifiers == 0) {
*numModifiers = 1;
return true;
}
modifiers[0] = 0; // DRM_FORMAT_MOD_LINEAR
if (externalOnly != nullptr) externalOnly[0] = EGL_TRUE;
*numModifiers = 1;
return true;
}
} // namespace EGLState
+22 -6
View File
@@ -200,6 +200,15 @@ namespace MobileGL {
EGLint Pitch = 0;
Bool HasModifier = false;
Uint64 Modifier = 0;
// YUV (two planes): the second plane, and the colour hints (0 = not given).
Uint32 PlaneCount = 1;
int Plane1Fd = -1;
EGLint Plane1Offset = 0;
EGLint Plane1Pitch = 0;
Uint32 ColorSpace = 0;
Uint32 Range = 0;
Uint32 SitingX = 0;
Uint32 SitingY = 0;
};
struct SharedImageInfo {
Uint64 Id = 0;
@@ -215,11 +224,16 @@ namespace MobileGL {
static constexpr Uint32 kDrmFourccXbgr8888 = 0x34324258u; // 'XB24'
static constexpr Uint32 kDrmFourccArgb8888 = 0x34325241u; // 'AR24'
static constexpr Uint32 kDrmFourccXrgb8888 = 0x34325258u; // 'XR24'
// Two-plane YUV 4:2:0, sampled only (through GL_OES_EGL_image_external, or as RGBA).
static constexpr Uint32 kDrmFourccNv12 = 0x3231564Eu; // 'NV12'
static constexpr Uint32 kDrmFourccP010 = 0x30313050u; // 'P010'
static Bool IsDmaBufFourccSupported(Uint32 fourcc);
static Bool IsDmaBufFourccYuv(Uint32 fourcc);
// The checks eglCreateImage owes an EGL_LINUX_DMA_BUF_EXT target before anything
// is imported: the display, a null context and buffer, and a complete single-plane
// attribute list of a supported format. False with the EGL error set.
// is imported: the display, a null context and buffer, and a complete attribute
// list of a supported format - one plane, or two for YUV. False with the EGL error
// set. Whether the YUV formats can be served at all is the caller's question.
Bool PrepareDmaBufImport(EGLDisplayHandle display, EGLContextHandle context, EGLClientBuffer buffer,
const EGLAttrib* attribList, DmaBufImportAttribs* out);
// An EGLImage naming an imported shared image.
@@ -230,10 +244,12 @@ namespace MobileGL {
// Shared images whose EGLImages went with an eglTerminate: the caller releases them.
Vector<Uint64> TakeOrphanedSharedImages();
// eglQueryDmaBufFormatsEXT / eglQueryDmaBufModifiersEXT. `available` is whether the
// backend has shared images at all; without them the lists are empty.
Bool QueryDmaBufFormats(EGLDisplayHandle display, Bool available, EGLint maxFormats, EGLint* formats,
EGLint* numFormats);
Bool QueryDmaBufModifiers(EGLDisplayHandle display, Bool available, EGLint format,
// backend has shared images at all; without them the lists are empty. `yuvAvailable`
// adds the YUV formats, which are external-only (OES_EGL_image_external) and LINEAR
// as far as an explicit modifier goes.
Bool QueryDmaBufFormats(EGLDisplayHandle display, Bool available, Bool yuvAvailable,
EGLint maxFormats, EGLint* formats, EGLint* numFormats);
Bool QueryDmaBufModifiers(EGLDisplayHandle display, Bool available, Bool yuvAvailable, EGLint format,
EGLint maxModifiers, Uint64* modifiers, EGLBoolean* externalOnly,
EGLint* numModifiers);
@@ -1875,8 +1875,25 @@ namespace MobileGL::MG_State::GLState {
artifacts.uniformReflection.reserve(static_cast<SizeT>(uniformCount));
artifacts.uniformIndexByName.clear();
artifacts.uniformIndexByName.reserve(static_cast<SizeT>(uniformCount));
// GL_OES_EGL_image_external: the uniforms a stage declared samplerExternalOES were
// compiled as sampler2D (LowerExternalSamplers) and are reported with their own type,
// which is what sends their units to the external binding point.
Vector<String> externalSamplers;
for (const LinkShaderInput& shader : in.shaders) {
if (shader.compiled == nullptr) continue;
for (String& name :
MG_Util::ShaderTranspiler::ExternalSamplerUniformNames(CompiledArtifacts(shader.compiled).preprocessedSource))
externalSamplers.push_back(Move(name));
}
const auto isExternalSampler = [&externalSamplers](const String& reflected) {
if (externalSamplers.empty()) return false;
const String base = reflected.substr(0, reflected.find('['));
return std::find(externalSamplers.begin(), externalSamplers.end(), base) != externalSamplers.end();
};
for (Int i = 0; i < uniformCount; ++i) {
ProgramObject::UniformReflection record = MakeResourceReflection(program.getUniform(i));
if (record.glDefineType == GL_SAMPLER_2D && isExternalSampler(record.name))
record.glDefineType = GL_SAMPLER_EXTERNAL_OES;
// A block-level layout(row_major)/(column_major) that the member did not inherit
// in its own qualifier. Resolved once HERE rather than at every GL_UNIFORM_* query,
// which is what the getUniformBlock() fallback in the old accessors was doing.
@@ -43,6 +43,8 @@ namespace MobileGL::MG_State::GLState {
return MakeShared<TextureObjectCubeMapArray>(index);
case TextureTarget::Texture2DMultisampleArray:
return MakeShared<TextureObject2DMultisampleArray>(index);
case TextureTarget::External:
return MakeShared<TextureObjectExternal>(index);
default:
MOBILEGL_ASSERT(false, "Unimplemented texture type when creating texture object!: %d", (int)target);
return nullptr;
@@ -185,4 +187,26 @@ namespace MobileGL::MG_State::GLState {
Bool ShareGroupState::ValidateRenderbufferObject(Uint index) const {
return m_renderbufferObjects.find(index) != m_renderbufferObjects.end();
}
UniquePtr<BufferObject> ShareGroupState::TakeClientArrayBuffer() {
const std::lock_guard<std::mutex> lock(m_clientArrayBuffersMutex);
if (m_clientArrayBuffers.empty()) return nullptr;
UniquePtr<BufferObject> buffer = std::move(m_clientArrayBuffers.back());
m_clientArrayBuffers.pop_back();
return buffer;
}
void ShareGroupState::ReturnClientArrayBuffer(UniquePtr<BufferObject> buffer) {
// Enough for a draw's every attribute plus its indices; more is never in use at once.
constexpr SizeT kKeptClientArrayBuffers = 48;
if (!buffer) return;
{
const std::lock_guard<std::mutex> lock(m_clientArrayBuffersMutex);
if (m_clientArrayBuffers.size() < kKeptClientArrayBuffers) {
m_clientArrayBuffers.push_back(std::move(buffer));
return;
}
}
// A surplus buffer is destroyed outside the lock (its destructor emits).
}
} // namespace MobileGL::MG_State::GLState
@@ -8,6 +8,7 @@
#pragma once
#include <Includes.h>
#include <mutex>
#include <MG_Util/Miscellany/IndexGenerator.h>
#include "BufferState/BufferObject.h"
#include "ProgramState/ProgramState.h"
@@ -67,6 +68,12 @@ namespace MobileGL::MG_State::GLState {
ProgramState& Programs() { return m_programState; }
const ProgramState& Programs() const { return m_programState; }
// Unnamed buffers that carried a draw's client-memory arrays, kept for the next such draw
// so it respecifies one of them (a non-blocking record) instead of creating a resource
// (a round trip) and destroying it again. Buffers are share-group objects, hence here.
UniquePtr<BufferObject> TakeClientArrayBuffer();
void ReturnClientArrayBuffer(UniquePtr<BufferObject> buffer);
private:
Uint64 m_token = 0;
UnorderedMap<Uint, SharedPtr<BufferObject>> m_bufferObjects;
@@ -77,5 +84,7 @@ namespace MobileGL::MG_State::GLState {
IndexGenerator<Uint> m_renderbufferNames{1024, 1};
SamplerState m_samplerState;
ProgramState m_programState;
std::mutex m_clientArrayBuffersMutex;
Vector<UniquePtr<BufferObject>> m_clientArrayBuffers;
};
} // namespace MobileGL::MG_State::GLState
@@ -21,6 +21,10 @@ namespace MobileGL {
Texture2DArray,
TextureCubeMapArray,
Texture2DMultisampleArray,
// GL_OES_EGL_image_external: a 2D texture whose only storage is an EGLImage (a YUV one is
// converted to RGBA by the backend), with its own binding point on every unit. Below the
// frontend it IS a 2D texture: the pipe and the backends never see this target.
External,
TextureTargetCount,
Unknown = -1
};
@@ -127,7 +127,9 @@ namespace MobileGL {
TextureObjectBase::TextureObjectBase(TextureTarget target, Uint externalIndex)
: m_externalIndex(externalIndex), m_lifetimeId(AllocateLifetimeId()), m_target(target) {
m_sampler = MakeShared<SamplerObject>(0);
if (target == TextureTarget::TextureRectangle) {
if (target == TextureTarget::TextureRectangle || target == TextureTarget::External) {
// (An external texture, GL_OES_EGL_image_external, starts the same way: LINEAR
// and CLAMP_TO_EDGE, and it has no mip chain either.)
// A rectangle texture has no mip chain, so its initial sampler state is not
// the shared one: TEXTURE_MIN_FILTER is LINEAR and TEXTURE_WRAP_S/T are
// CLAMP_TO_EDGE (GL 4.6 core table 23.15). Leaving the 2D default of
@@ -53,6 +53,10 @@ namespace MobileGL {
STUB_TEXTURE_OBJECT_CLASS_DEFINITION(TextureObject2DMultisampleArray,
TextureTarget::Texture2DMultisampleArray,
{TextureUploadTarget::Texture2DMultisampleArray});
// GL_OES_EGL_image_external: level 0 is defined as a 2D image (by glEGLImageTargetTexture2DOES
// only - glTexImage2D does not take the target).
STUB_TEXTURE_OBJECT_CLASS_DEFINITION(TextureObjectExternal, TextureTarget::External,
{TextureUploadTarget::Texture2D});
#undef STUB_TEXTURE_OBJECT_CLASS_DEFINITION
} // namespace GLState
@@ -1590,6 +1590,19 @@ TEST(ComputeWorkGroupCapabilities, TakesEveryAxisFromTheIndexedQuery) {
}
namespace {
const char* g_damageExtensions = nullptr;
EGLDisplay g_damageDisplay = EGL_NO_DISPLAY;
bool g_damageExtAvailable = false;
bool g_damageKhrAvailable = false;
std::vector<std::string> g_damageQueries;
EGLDisplay FakeDamageDisplay() { return g_damageDisplay; }
const char* FakeDamageQueryString(EGLDisplay display, EGLint name) {
EXPECT_EQ(display, g_damageDisplay);
EXPECT_EQ(name, EGL_EXTENSIONS);
return g_damageExtensions;
}
EGLBoolean FakeSwapBuffersWithDamage(EGLDisplay dpy, EGLSurface surface, const EGLint* rects, EGLint nRects) {
(void)dpy;
(void)surface;
@@ -1597,6 +1610,14 @@ namespace {
(void)nRects;
return EGL_TRUE;
}
__eglMustCastToProperFunctionPointerType FakeDamageGetProcAddress(const char* name) {
g_damageQueries.emplace_back(name);
const bool available = std::string(name) == "eglSwapBuffersWithDamageEXT"
? g_damageExtAvailable : g_damageKhrAvailable;
return available ? reinterpret_cast<__eglMustCastToProperFunctionPointerType>(&FakeSwapBuffersWithDamage)
: nullptr;
}
} // namespace
// EXT -> KHR -> nothing, the whole policy behind the loader's single damage-present slot.
@@ -1609,13 +1630,46 @@ namespace {
// call to either spelling is defined to be equivalent to that anyway.
TEST(SwapBuffersWithDamageResolution, PrefersExtThenKhrThenAnswersNothing) {
using namespace MobileGL;
MG_External::EGL::eglSwapBuffersWithDamageEXT_PTR ext = &FakeSwapBuffersWithDamage;
MG_External::EGL::eglSwapBuffersWithDamageKHR_PTR khr = &FakeSwapBuffersWithDamage;
MG_External::EGLFunctionsTable funcs{};
funcs.eglGetCurrentDisplay = &FakeDamageDisplay;
funcs.eglQueryString = &FakeDamageQueryString;
funcs.eglGetProcAddress = &FakeDamageGetProcAddress;
g_damageDisplay = reinterpret_cast<EGLDisplay>(1);
g_damageExtensions = "EGL_EXT_swap_buffers_with_damage EGL_KHR_swap_buffers_with_damage";
g_damageExtAvailable = g_damageKhrAvailable = true;
g_damageQueries.clear();
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(funcs), &FakeSwapBuffersWithDamage);
EXPECT_EQ(g_damageQueries, (std::vector<std::string>{"eglSwapBuffersWithDamageEXT"}));
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(ext, khr), ext)
<< "a driver that exports both spellings is answered with the EXT one";
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(nullptr, khr), khr)
<< "the KHR spelling alone must still fill the slot";
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(nullptr, nullptr), nullptr)
<< "neither spelling stays null, so the caller keeps eglSwapBuffers";
g_damageExtAvailable = false;
g_damageQueries.clear();
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(funcs), &FakeSwapBuffersWithDamage);
EXPECT_EQ(g_damageQueries, (std::vector<std::string>{"eglSwapBuffersWithDamageEXT", "eglSwapBuffersWithDamageKHR"}));
g_damageExtensions = "EGL_KHR_swap_buffers_with_damage";
g_damageQueries.clear();
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(funcs), &FakeSwapBuffersWithDamage);
EXPECT_EQ(g_damageQueries, (std::vector<std::string>{"eglSwapBuffersWithDamageKHR"}));
g_damageKhrAvailable = false;
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(funcs), nullptr);
g_damageExtensions = "EGL_EXT_swap_buffers_with_damage";
g_damageQueries.clear();
g_damageKhrAvailable = true;
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(funcs), nullptr);
EXPECT_EQ(g_damageQueries, (std::vector<std::string>{"eglSwapBuffersWithDamageEXT"}));
g_damageExtAvailable = true;
for (const char* extensions : {"", "EGL_EXT_swap_buffers_with_damage_extra", static_cast<const char*>(nullptr)}) {
g_damageExtensions = extensions;
g_damageQueries.clear();
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(funcs), nullptr);
EXPECT_TRUE(g_damageQueries.empty());
}
g_damageExtensions = "EGL_EXT_swap_buffers_with_damage EGL_KHR_swap_buffers_with_damage";
g_damageDisplay = EGL_NO_DISPLAY;
g_damageQueries.clear();
EXPECT_EQ(MG_Util::BackendLoader::ResolveSwapBuffersWithDamage(funcs), nullptr);
EXPECT_TRUE(g_damageQueries.empty());
}
+111 -8
View File
@@ -405,40 +405,143 @@ TEST(EGLStateDmaBufImport, SharedImageLifetime) {
TEST(EGLStateDmaBufImport, FormatAndModifierQueries) {
auto fixture = CreateFixture();
EGLint count = -1;
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, 0, nullptr, &count));
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, false, 0, nullptr, &count));
EXPECT_EQ(count, 4);
EGLint formats[4] = {};
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, 4, formats, &count));
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, false, 4, formats, &count));
ASSERT_EQ(count, 4);
EXPECT_EQ(formats[0], static_cast<EGLint>(kAbgr8888));
EXPECT_EQ(formats[1], static_cast<EGLint>(kXbgr8888));
EXPECT_EQ(formats[2], static_cast<EGLint>(kArgb8888));
EXPECT_EQ(formats[3], static_cast<EGLint>(kXrgb8888));
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, 1, formats, &count));
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, false, 1, formats, &count));
EXPECT_EQ(count, 1);
// Without shared images there is nothing to import.
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, false, 0, nullptr, &count));
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, false, false, 0, nullptr, &count));
EXPECT_EQ(count, 0);
EXPECT_FALSE(fixture->State.QueryDmaBufFormats(fixture->Display, true, -1, formats, &count));
EXPECT_FALSE(fixture->State.QueryDmaBufFormats(fixture->Display, true, false, -1, formats, &count));
EXPECT_EQ(fixture->State.ConsumeError(), EGL_BAD_PARAMETER);
// No explicit modifiers: importers fall back to the implicit one.
uint64_t modifiers[2] = {};
EGLBoolean externalOnly[2] = {};
count = -1;
ASSERT_TRUE(fixture->State.QueryDmaBufModifiers(fixture->Display, true, static_cast<EGLint>(kAbgr8888), 2,
ASSERT_TRUE(fixture->State.QueryDmaBufModifiers(fixture->Display, true, false, static_cast<EGLint>(kAbgr8888), 2,
modifiers, externalOnly, &count));
EXPECT_EQ(count, 0);
EXPECT_FALSE(fixture->State.QueryDmaBufModifiers(fixture->Display, true, static_cast<EGLint>(kRgb565), 0,
EXPECT_FALSE(fixture->State.QueryDmaBufModifiers(fixture->Display, true, false, static_cast<EGLint>(kRgb565), 0,
nullptr, nullptr, &count));
EXPECT_EQ(fixture->State.ConsumeError(), EGL_BAD_PARAMETER);
ASSERT_TRUE(fixture->State.TerminateDisplay(fixture->Display));
EXPECT_FALSE(fixture->State.QueryDmaBufFormats(fixture->Display, true, 0, nullptr, &count));
EXPECT_FALSE(fixture->State.QueryDmaBufFormats(fixture->Display, true, false, 0, nullptr, &count));
EXPECT_EQ(fixture->State.ConsumeError(), EGL_NOT_INITIALIZED);
}
namespace {
constexpr EGLAttrib kNv12 = StateContext::kDrmFourccNv12;
constexpr EGLAttrib kP010 = StateContext::kDrmFourccP010;
// A complete NV12 import as a video decoder exports one: both planes in one buffer.
std::vector<EGLAttrib> Nv12Attribs(EGLAttrib fourcc = kNv12) {
return {EGL_WIDTH,
1280,
EGL_HEIGHT,
720,
EGL_LINUX_DRM_FOURCC_EXT,
fourcc,
EGL_DMA_BUF_PLANE0_FD_EXT,
7,
EGL_DMA_BUF_PLANE0_OFFSET_EXT,
0,
EGL_DMA_BUF_PLANE0_PITCH_EXT,
1280,
EGL_DMA_BUF_PLANE1_FD_EXT,
9,
EGL_DMA_BUF_PLANE1_OFFSET_EXT,
942080,
EGL_DMA_BUF_PLANE1_PITCH_EXT,
1280,
EGL_NONE};
}
} // namespace
TEST(EGLStateDmaBufImport, ParsesATwoPlaneYuvImportWithItsHints) {
auto fixture = CreateFixture();
auto attribs = WithAttrib(Nv12Attribs(), EGL_YUV_COLOR_SPACE_HINT_EXT, EGL_ITU_REC709_EXT);
attribs = WithAttrib(attribs, EGL_SAMPLE_RANGE_HINT_EXT, EGL_YUV_NARROW_RANGE_EXT);
attribs = WithAttrib(attribs, EGL_YUV_CHROMA_HORIZONTAL_SITING_HINT_EXT, EGL_YUV_CHROMA_SITING_0_5_EXT);
StateContext::DmaBufImportAttribs parsed;
ASSERT_TRUE(fixture->State.PrepareDmaBufImport(fixture->Display, EGL_NO_CONTEXT, nullptr, attribs.data(), &parsed));
EXPECT_EQ(parsed.PlaneCount, 2u);
EXPECT_EQ(parsed.Fd, 7);
EXPECT_EQ(parsed.Plane1Fd, 9);
EXPECT_EQ(parsed.Plane1Offset, 942080);
EXPECT_EQ(parsed.Plane1Pitch, 1280);
EXPECT_EQ(parsed.ColorSpace, static_cast<uint32_t>(EGL_ITU_REC709_EXT));
EXPECT_EQ(parsed.Range, static_cast<uint32_t>(EGL_YUV_NARROW_RANGE_EXT));
EXPECT_EQ(parsed.SitingX, static_cast<uint32_t>(EGL_YUV_CHROMA_SITING_0_5_EXT));
EXPECT_EQ(parsed.SitingY, 0u);
EXPECT_FALSE(parsed.HasModifier);
EXPECT_EQ(PrepareError(*fixture, Nv12Attribs(kP010)), EGL_SUCCESS);
// One modifier for the buffer: the same on both planes, or none on either.
auto linear = WithAttrib(WithAttrib(Nv12Attribs(), EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT, 0),
EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT, 0);
EXPECT_EQ(PrepareError(*fixture, linear), EGL_BAD_PARAMETER);
linear = WithAttrib(WithAttrib(linear, EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT, 0), EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT, 0);
EXPECT_EQ(PrepareError(*fixture, linear), EGL_SUCCESS);
auto mixed = WithAttrib(WithAttrib(WithoutAttrib(WithoutAttrib(linear, EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT),
EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT),
EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT, 1),
EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT, 0);
EXPECT_EQ(PrepareError(*fixture, mixed), EGL_BAD_MATCH);
}
TEST(EGLStateDmaBufImport, RefusesAnIncompleteYuvImport) {
auto fixture = CreateFixture();
for (const EGLAttrib required :
{EGL_DMA_BUF_PLANE1_FD_EXT, EGL_DMA_BUF_PLANE1_OFFSET_EXT, EGL_DMA_BUF_PLANE1_PITCH_EXT}) {
EXPECT_EQ(PrepareError(*fixture, WithoutAttrib(Nv12Attribs(), required)), EGL_BAD_PARAMETER)
<< "missing 0x" << std::hex << required;
}
// NV12 has two planes, not three.
EXPECT_EQ(PrepareError(*fixture, WithAttrib(Nv12Attribs(), EGL_DMA_BUF_PLANE2_FD_EXT, 7)), EGL_BAD_ATTRIBUTE);
auto badPitch = WithAttrib(WithoutAttrib(Nv12Attribs(), EGL_DMA_BUF_PLANE1_PITCH_EXT), EGL_DMA_BUF_PLANE1_PITCH_EXT, 0);
EXPECT_EQ(PrepareError(*fixture, badPitch), EGL_BAD_ACCESS);
}
TEST(EGLStateDmaBufImport, YuvFormatsAreListedOnlyWhenTheServerHoldsThem) {
auto fixture = CreateFixture();
EGLint count = -1;
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, false, 0, nullptr, &count));
EXPECT_EQ(count, 4);
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, true, 0, nullptr, &count));
EXPECT_EQ(count, 6);
EGLint formats[6] = {};
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, true, true, 6, formats, &count));
EXPECT_EQ(formats[4], static_cast<EGLint>(kNv12));
EXPECT_EQ(formats[5], static_cast<EGLint>(kP010));
ASSERT_TRUE(fixture->State.QueryDmaBufFormats(fixture->Display, false, true, 0, nullptr, &count));
EXPECT_EQ(count, 0);
// A YUV format: LINEAR, external only.
uint64_t modifiers[2] = {99, 99};
EGLBoolean externalOnly[2] = {EGL_FALSE, EGL_FALSE};
ASSERT_TRUE(fixture->State.QueryDmaBufModifiers(fixture->Display, true, true, static_cast<EGLint>(kNv12), 0,
nullptr, nullptr, &count));
EXPECT_EQ(count, 1);
ASSERT_TRUE(fixture->State.QueryDmaBufModifiers(fixture->Display, true, true, static_cast<EGLint>(kNv12), 2,
modifiers, externalOnly, &count));
EXPECT_EQ(count, 1);
EXPECT_EQ(modifiers[0], 0u);
EXPECT_EQ(externalOnly[0], EGL_TRUE);
EXPECT_FALSE(fixture->State.QueryDmaBufModifiers(fixture->Display, true, false, static_cast<EGLint>(kNv12), 0,
nullptr, nullptr, &count));
EXPECT_EQ(fixture->State.ConsumeError(), EGL_BAD_PARAMETER);
}
// EGL_EXT_create_context_robustness / EGL 1.5: the reset notification behavior a context is
// created with, under both names of the attribute (the extension's, which Chrome, KWin and Qt
// pass, and EGL 1.5's, which ANGLE's GL-on-EGL backend passes for its own native context), the
@@ -36,7 +36,8 @@ namespace {
bool Supported(uint32_t fourcc) {
return fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_ABGR8888 || fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_XBGR8888 ||
fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_ARGB8888 || fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_XRGB8888;
fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_ARGB8888 || fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_XRGB8888 ||
fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_NV12;
}
} // namespace
@@ -44,7 +45,13 @@ FAKE_EXPORT uint32_t mobilegl_shared_image_abi_version(void) { return MOBILEGL_S
FAKE_EXPORT int mobilegl_server_available(void) { return 1; }
FAKE_EXPORT int mobilegl_shared_image_formats(uint32_t*, uint32_t) { return 4; }
FAKE_EXPORT int mobilegl_shared_image_formats(uint32_t* formats, uint32_t capacity) {
const uint32_t all[] = {MOBILEGL_SHARED_IMAGE_FOURCC_ABGR8888, MOBILEGL_SHARED_IMAGE_FOURCC_XBGR8888,
MOBILEGL_SHARED_IMAGE_FOURCC_ARGB8888, MOBILEGL_SHARED_IMAGE_FOURCC_XRGB8888,
MOBILEGL_SHARED_IMAGE_FOURCC_NV12};
for (uint32_t i = 0; i < 5 && i < capacity; ++i) formats[i] = all[i];
return 5;
}
FAKE_EXPORT int mobilegl_shared_image_allocate(uint32_t width, uint32_t height, uint32_t fourcc,
struct mobilegl_shared_image* out) {
@@ -64,6 +71,11 @@ FAKE_EXPORT int mobilegl_shared_image_allocate(uint32_t width, uint32_t height,
out->stride = width * 4;
out->offset = 0;
out->modifier = 0x00ffffffffffffffull;
if (fourcc == MOBILEGL_SHARED_IMAGE_FOURCC_NV12) {
out->stride = width;
out->plane1_stride = width;
out->plane1_offset = width * height;
}
return 0;
}
@@ -103,3 +115,10 @@ FAKE_EXPORT int fake_shared_image_imports(void) {
const std::lock_guard<std::mutex> lock(g_mutex);
return g_imports;
}
// A YUV import names an image by its descriptor as an RGBA one does; the planes are not looked at.
FAKE_EXPORT int mobilegl_shared_image_import_planes(int fd, uint32_t width, uint32_t height, uint32_t fourcc,
const struct mobilegl_shared_image_planes* planes, uint64_t* out_id) {
if (planes == nullptr || planes->plane_count != 2 || fourcc != MOBILEGL_SHARED_IMAGE_FOURCC_NV12) return -EINVAL;
return mobilegl_shared_image_import(fd, width, height, fourcc, out_id);
}
+56
View File
@@ -180,3 +180,59 @@ TEST_F(GbmGlamor, AForeignDmaBufIsRefused) {
EXPECT_EQ(gbm_bo_import(m_gbm, GBM_BO_IMPORT_FD, &data, GBM_BO_USE_RENDERING), nullptr);
::close(foreign);
}
// YUV buffers (ABI 2): NV12 where the library lists it, two planes in one descriptor, sampled only.
TEST_F(GbmGlamor, AnNv12BufferHasTwoPlanesInOneDescriptor) {
EXPECT_EQ(gbm_device_is_format_supported(m_gbm, GBM_FORMAT_NV12, 0), 1);
EXPECT_EQ(gbm_device_is_format_supported(m_gbm, GBM_FORMAT_NV12, GBM_BO_USE_RENDERING), 0);
EXPECT_EQ(gbm_device_is_format_supported(m_gbm, 0x30313050u /* P010 */, 0), 0) << "the library does not list P010";
EXPECT_EQ(gbm_device_get_format_modifier_plane_count(m_gbm, GBM_FORMAT_NV12, DRM_FORMAT_MOD_INVALID), 2);
EXPECT_EQ(gbm_bo_create(m_gbm, 64, 32, GBM_FORMAT_NV12, GBM_BO_USE_RENDERING), nullptr);
gbm_bo* bo = gbm_bo_create(m_gbm, 64, 32, GBM_FORMAT_NV12, 0);
ASSERT_NE(bo, nullptr);
EXPECT_EQ(gbm_bo_get_plane_count(bo), 2);
EXPECT_EQ(gbm_bo_get_stride_for_plane(bo, 0), 64u);
EXPECT_EQ(gbm_bo_get_stride_for_plane(bo, 1), 64u);
EXPECT_EQ(gbm_bo_get_offset(bo, 1), 64u * 32u);
EXPECT_EQ(gbm_bo_get_handle_for_plane(bo, 1).u32, gbm_bo_get_handle(bo).u32);
const int fd = gbm_bo_get_fd(bo);
const int plane1 = gbm_bo_get_fd_for_plane(bo, 1);
ASSERT_GE(fd, 0);
ASSERT_GE(plane1, 0);
// Imported back with both planes, through one descriptor or two of the same buffer.
gbm_import_fd_modifier_data data{};
data.width = 64;
data.height = 32;
data.format = GBM_FORMAT_NV12;
data.num_fds = 2;
data.fds[0] = fd;
data.fds[1] = plane1;
data.strides[0] = data.strides[1] = 64;
data.offsets[1] = 64 * 32;
data.modifier = DRM_FORMAT_MOD_INVALID;
gbm_bo* imported = gbm_bo_import(m_gbm, GBM_BO_IMPORT_FD_MODIFIER, &data, 0);
ASSERT_NE(imported, nullptr);
EXPECT_EQ(gbm_bo_get_plane_count(imported), 2);
EXPECT_EQ(gbm_bo_get_offset(imported, 1), 64u * 32u);
// Planes in two different buffers, and a one-plane import that cannot say where plane 1 is.
const int other = ::memfd_create("other", MFD_CLOEXEC);
data.fds[1] = other;
EXPECT_EQ(gbm_bo_import(m_gbm, GBM_BO_IMPORT_FD_MODIFIER, &data, 0), nullptr);
gbm_import_fd_data plain{};
plain.fd = fd;
plain.width = 64;
plain.height = 32;
plain.stride = 64;
plain.format = GBM_FORMAT_NV12;
EXPECT_EQ(gbm_bo_import(m_gbm, GBM_BO_IMPORT_FD, &plain, 0), nullptr);
::close(other);
gbm_bo_destroy(imported);
gbm_bo_destroy(bo);
::close(fd);
::close(plane1);
EXPECT_EQ(m_live(), m_liveAtStart) << "every reference was dropped";
}
+1 -1
View File
@@ -704,7 +704,7 @@ TEST(PipeCatalogue, LateArrivalsAreAppendedWithoutRenumbering) {
// region (kMGPMaxDamageRects rectangles), as the present record carries its frame's damage.
EXPECT_EQ(sizeof(MGPSharedImageOp), 296u);
EXPECT_EQ(sizeof(MGPPresent), 272u);
EXPECT_EQ(sizeof(MGPSharedImageReply), 40u);
EXPECT_EQ(sizeof(MGPSharedImageReply), 48u);
EXPECT_EQ(sizeof(MGPSharedImageFdOffer), 32u);
EXPECT_EQ(sizeof(MGPHandleOnly), 16u);
// rv's two (§5.3/§7.6): the residual-value POD - 2 + 15 Uint32s, 2 Uint8s and 2 pad bytes,
@@ -384,6 +384,13 @@ TEST(TextureEmit, EveryTextureTargetMapsToItsOwnResourceTarget) {
<< "TextureTarget " << i << " maps onto the BUFFER row, which the ack predicate reads";
EXPECT_NE(resourceTarget, static_cast<Uint32>(MGPipeResourceTarget::Renderbuffer))
<< "TextureTarget " << i << " maps onto the RENDERBUFFER row";
// The one deliberate fold: an external texture (GL_OES_EGL_image_external) is a 2D texture
// below the frontend.
if (target == TextureTarget::External) {
EXPECT_EQ(resourceTarget, static_cast<Uint32>(MGPipeResourceTarget::Tex2D));
seen.push_back(resourceTarget);
continue;
}
for (const Uint32 previous : seen) {
EXPECT_NE(previous, resourceTarget)
<< "TextureTarget " << i << " shares its resource target with an earlier one";
+99
View File
@@ -8,6 +8,7 @@
#include <gtest/gtest.h>
#include <spirv_reflect.h>
#include <cstdio>
#include <cstring>
#include <utility>
#include <vector>
@@ -5276,3 +5277,101 @@ TEST_F(ProgramTest, ASpirvModulesXfbDecorationsDoNotSurviveIntoTheTranslatedSour
DrainProgramTestErrors();
}
// GL_OES_EGL_image_external. A samplerExternalOES uniform is compiled as a sampler2D (the desktop
// profile the shader is compiled under has no external overloads; below the frontend an external
// texture is a 2D one) and reported with its own type, which is what sends its unit to the
// external binding point.
namespace {
GLuint LinkExternalSamplerProgram(const char* vertexSource, const char* fragmentSource, String& log) {
char info[2048] = "";
const GLuint vs = CreateShader(GL_VERTEX_SHADER);
ShaderSource(vs, 1, &vertexSource, nullptr);
CompileShader(vs);
const GLuint fs = CreateShader(GL_FRAGMENT_SHADER);
ShaderSource(fs, 1, &fragmentSource, nullptr);
CompileShader(fs);
GLint compiled = GL_FALSE;
GetShaderiv(fs, GL_COMPILE_STATUS, &compiled);
GetShaderInfoLog(fs, sizeof(info), nullptr, info);
log = info;
if (compiled != GL_TRUE) return 0;
const GLuint program = CreateProgram();
AttachShader(program, vs);
AttachShader(program, fs);
LinkProgram(program);
GLint linked = GL_FALSE;
GetProgramiv(program, GL_LINK_STATUS, &linked);
GetProgramInfoLog(program, sizeof(info), nullptr, info);
log += info;
return linked == GL_TRUE ? program : 0;
}
GLenum UniformTypeByName(GLuint program, const char* wanted) {
GLint count = 0;
GetProgramiv(program, GL_ACTIVE_UNIFORMS, &count);
for (GLint i = 0; i < count; ++i) {
char name[64] = "";
GLsizei length = 0;
GLint size = 0;
GLenum type = 0;
GetActiveUniform(program, static_cast<GLuint>(i), sizeof(name), &length, &size, &type, name);
if (String(name) == wanted) return type;
}
return GL_NONE;
}
const char* kExternalVertexSource = "#version 300 es\n"
"out vec2 uv;\n"
"void main() { uv = vec2(0.5); gl_Position = vec4(0.0, 0.0, 0.0, 1.0); }\n";
} // namespace
TEST_F(ProgramTest, AnEsslExternalSamplerReportsItsOwnType) {
DrainProgramTestErrors();
const char* fs = "#version 300 es\n"
"#extension GL_OES_EGL_image_external_essl3 : require\n"
"precision mediump float;\n"
"uniform highp samplerExternalOES videoFrame;\n"
"uniform sampler2D overlay;\n"
"in vec2 uv;\n"
"out vec4 color;\n"
"void main() { color = texture(videoFrame, uv) + texture(overlay, uv); }\n";
String log;
const GLuint program = LinkExternalSamplerProgram(kExternalVertexSource, fs, log);
ASSERT_NE(program, 0u) << log;
EXPECT_EQ(UniformTypeByName(program, "videoFrame"), static_cast<GLenum>(GL_SAMPLER_EXTERNAL_OES));
EXPECT_EQ(UniformTypeByName(program, "overlay"), static_cast<GLenum>(GL_SAMPLER_2D));
DrainProgramTestErrors();
}
// A compositor's shader generator declares one name both ways, picking by #if: only the branch the
// preprocessor keeps says what the uniform is.
TEST_F(ProgramTest, AnExternalSamplerInAnInactiveBranchLeavesTheSampler2DAlone) {
DrainProgramTestErrors();
const char* fsTemplate = "#version 140\n"
"%s"
"#if TRAIT_MAP_TEXTURE\n"
"uniform sampler2D sampler;\n"
"#endif\n"
"#if TRAIT_MAP_EXTERNAL_TEXTURE\n"
"#extension GL_OES_EGL_image_external : require\n"
"uniform samplerExternalOES sampler;\n"
"#endif\n"
"in vec2 uv;\n"
"out vec4 color;\n"
"void main() { color = texture(sampler, uv); }\n";
const char* vs = "#version 140\n"
"out vec2 uv;\n"
"void main() { uv = vec2(0.5); gl_Position = vec4(0.0, 0.0, 0.0, 1.0); }\n";
for (const auto& [traits, expected] :
{std::pair{"#define TRAIT_MAP_TEXTURE 1\n", static_cast<GLenum>(GL_SAMPLER_2D)},
std::pair{"#define TRAIT_MAP_EXTERNAL_TEXTURE 1\n", static_cast<GLenum>(GL_SAMPLER_EXTERNAL_OES)}}) {
char fs[1024];
std::snprintf(fs, sizeof(fs), fsTemplate, traits);
String log;
const GLuint program = LinkExternalSamplerProgram(vs, fs, log);
ASSERT_NE(program, 0u) << traits << log;
EXPECT_EQ(UniformTypeByName(program, "sampler"), expected) << traits;
}
DrainProgramTestErrors();
}
@@ -123,11 +123,17 @@ TEST_F(ProgramUtilTest, UnformattedFloatStorageImagesKeepIntegerAtomicImagesType
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
layout(rgba16, binding = 0) uniform image2D floatImage;
layout(r32ui, binding = 1) uniform uimage2D atomicImage;
layout(rgba32f, binding = 2) uniform image2D otherFloatImages[2];
layout(std430, binding = 3) buffer Output0 { uint value0; };
layout(std430, binding = 4) buffer Output1 { uint value1; };
void main() {
ivec2 coordinate = ivec2(gl_GlobalInvocationID.xy);
imageStore(floatImage, coordinate, imageLoad(floatImage, coordinate));
imageStore(otherFloatImages[1], coordinate, imageLoad(otherFloatImages[0], coordinate));
imageAtomicAdd(atomicImage, coordinate, 1u);
value0 = 11u;
value1 = 22u;
}
)";
@@ -174,6 +180,12 @@ void main() {
<< outputText;
EXPECT_EQ(countOccurrences(outputText, "OpCapability StorageImageWriteWithoutFormat"), 1u)
<< outputText;
EXPECT_EQ(countOccurrences(outputText, "2D 0 0 0 2 Unknown"), 1u) << outputText;
EXPECT_NE(outputText.find("Binding 0"), String::npos) << outputText;
EXPECT_NE(outputText.find("Binding 1"), String::npos) << outputText;
EXPECT_NE(outputText.find("Binding 2"), String::npos) << outputText;
EXPECT_NE(outputText.find("\"Output0\""), String::npos) << outputText;
EXPECT_NE(outputText.find("\"Output1\""), String::npos) << outputText;
EXPECT_TRUE(tools.Validate(outputBinary));
Vector<Uint32> secondOutputBinary;
@@ -181,6 +193,67 @@ void main() {
EXPECT_EQ(secondOutputBinary, outputBinary);
}
TEST_F(ProgramUtilTest, LowerRectImagesMergesWithExisting2DImageTypes) {
using namespace MG_Util::ShaderTranspiler;
const String spirvText = R"(
OpCapability Shader
OpCapability ImageRect
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main"
OpExecutionMode %main LocalSize 1 1 1
OpName %Output0 "Output0"
OpName %Output1 "Output1"
OpDecorate %Output0 Block
OpDecorate %Output1 Block
OpMemberDecorate %Output0 0 Offset 0
OpMemberDecorate %Output1 0 Offset 0
OpDecorate %buffer0 DescriptorSet 0
OpDecorate %buffer0 Binding 2
OpDecorate %buffer1 DescriptorSet 0
OpDecorate %buffer1 Binding 3
OpDecorate %rect DescriptorSet 0
OpDecorate %rect Binding 0
OpDecorate %flat DescriptorSet 0
OpDecorate %flat Binding 1
%void = OpTypeVoid
%uint = OpTypeInt 32 0
%Output0 = OpTypeStruct %uint
%Output1 = OpTypeStruct %uint
%outputPtr0 = OpTypePointer StorageBuffer %Output0
%outputPtr1 = OpTypePointer StorageBuffer %Output1
%buffer0 = OpVariable %outputPtr0 StorageBuffer
%buffer1 = OpVariable %outputPtr1 StorageBuffer
%rectType = OpTypeImage %uint Rect 0 0 0 2 R32ui
%flatType = OpTypeImage %uint 2D 0 0 0 2 R32ui
%rectPtr = OpTypePointer UniformConstant %rectType
%flatPtr = OpTypePointer UniformConstant %flatType
%rect = OpVariable %rectPtr UniformConstant
%flat = OpVariable %flatPtr UniformConstant
%mainType = OpTypeFunction %void
%main = OpFunction %void None %mainType
%entry = OpLabel
%r = OpLoad %rectType %rect
%f = OpLoad %flatType %flat
OpReturn
OpFunctionEnd
)";
spvtools::SpirvTools tools(SPV_ENV_VULKAN_1_1);
Vector<Uint32> inputBinary;
ASSERT_TRUE(tools.Assemble(spirvText, &inputBinary));
Vector<Uint32> outputBinary;
ASSERT_TRUE(ShaderCompiler::LowerRectImages(inputBinary, outputBinary));
EXPECT_TRUE(tools.Validate(outputBinary));
String outputText;
ASSERT_TRUE(tools.Disassemble(outputBinary, &outputText));
EXPECT_NE(outputText.find("Binding 0"), String::npos);
EXPECT_NE(outputText.find("Binding 1"), String::npos);
const auto firstImage = outputText.find("OpTypeImage");
ASSERT_NE(firstImage, String::npos);
EXPECT_EQ(outputText.find("OpTypeImage", firstImage + 1), String::npos);
EXPECT_NE(outputText.find("\"Output0\""), String::npos) << outputText;
EXPECT_NE(outputText.find("\"Output1\""), String::npos) << outputText;
}
TEST_F(ProgramUtilTest, UnformattedFloatStorageImagesKeepFloatAtomicImageTypesTyped) {
using namespace MG_Util::ShaderTranspiler;
@@ -60,6 +60,7 @@ using Uint32 = std::uint32_t;
#include <MG_Backend/DirectVulkan/Renderer/WireColorBlitSpirv.h>
#include <MG_Backend/DirectVulkan/Renderer/WireDepthMipmapSpirv.h>
#include <MG_Backend/DirectVulkan/Renderer/WireMultisampleResolveSpirv.h>
#include <MG_Backend/DirectVulkan/Renderer/WireYuvConvertSpirv.h>
struct BakedShader {
const char* name; // ctest entry name, and the bake file's stem
@@ -92,6 +93,18 @@ const BakedShader kBakedShaders[] = {
"kWireColorBlitFragmentSpirv", EShLangFragment,
glslang::EShTargetVulkan_1_0, glslang::EShTargetSpv_1_0,
kWireColorBlitFragmentSpirv),
MGL_BAKED_ROW("WireYuvConvertVertex",
"MobileGL/MG_Backend/DirectVulkan/Renderer/WireYuvConvert.vert",
"MobileGL/MG_Backend/DirectVulkan/Renderer/WireYuvConvertSpirv.h",
"kWireYuvConvertVertexSpirv", EShLangVertex,
glslang::EShTargetVulkan_1_0, glslang::EShTargetSpv_1_0,
kWireYuvConvertVertexSpirv),
MGL_BAKED_ROW("WireYuvConvertFragment",
"MobileGL/MG_Backend/DirectVulkan/Renderer/WireYuvConvert.frag",
"MobileGL/MG_Backend/DirectVulkan/Renderer/WireYuvConvertSpirv.h",
"kWireYuvConvertFragmentSpirv", EShLangFragment,
glslang::EShTargetVulkan_1_0, glslang::EShTargetSpv_1_0,
kWireYuvConvertFragmentSpirv),
MGL_BAKED_ROW("WireDepthMipmapVertex",
"MobileGL/MG_Backend/DirectVulkan/Renderer/WireDepthMipmap.vert",
"MobileGL/MG_Backend/DirectVulkan/Renderer/WireDepthMipmapSpirv.h",
+56
View File
@@ -6789,3 +6789,59 @@ TEST_F(TextureTest, GenerateMipmapKeepsA2DArrayLayerCountAtEveryLevel) {
MG_Impl::GLImpl::BindTexture(GL_TEXTURE_2D_ARRAY, 0);
DrainPendingGlErrors();
}
// GL_OES_EGL_image_external: the external target is a binding point of its own on every unit,
// with the extension's restrictions - no data upload, no mipmaps, clamp-to-edge and non-mipmapped
// filtering, LINEAR from birth - and one image unit per texture.
TEST_F(TextureTest, TheExternalTargetIsABindingPointOfItsOwn) {
using namespace MG_Impl::GLImpl;
GLuint names[2] = {};
GenTextures(2, names);
ActiveTexture(GL_TEXTURE3);
BindTexture(GL_TEXTURE_2D, names[0]);
BindTexture(GL_TEXTURE_EXTERNAL_OES, names[1]);
EXPECT_EQ(GetError(), GL_NO_ERROR);
GLint bound2D = 0, boundExternal = 0;
GetIntegerv(GL_TEXTURE_BINDING_2D, &bound2D);
GetIntegerv(GL_TEXTURE_BINDING_EXTERNAL_OES, &boundExternal);
EXPECT_EQ(static_cast<GLuint>(bound2D), names[0]);
EXPECT_EQ(static_cast<GLuint>(boundExternal), names[1]);
EXPECT_EQ(MG_State::pGLContext->GetTextureObject(names[1])->GetTarget(), TextureTarget::External);
// A texture first bound as external stays external.
BindTexture(GL_TEXTURE_2D, names[1]);
ExpectSingleGlError(GL_INVALID_OPERATION);
GLint value = 0;
GetTexParameteriv(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MIN_FILTER, &value);
EXPECT_EQ(value, GL_LINEAR);
GetTexParameteriv(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_S, &value);
EXPECT_EQ(value, GL_CLAMP_TO_EDGE);
GetTexParameteriv(GL_TEXTURE_EXTERNAL_OES, GL_REQUIRED_TEXTURE_IMAGE_UNITS_OES, &value);
EXPECT_EQ(value, 1);
EXPECT_EQ(GetError(), GL_NO_ERROR);
GetTexParameteriv(GL_TEXTURE_2D, GL_REQUIRED_TEXTURE_IMAGE_UNITS_OES, &value);
ExpectSingleGlError(GL_INVALID_ENUM);
TexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
TexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
EXPECT_EQ(GetError(), GL_NO_ERROR);
TexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_S, GL_REPEAT);
ExpectSingleGlError(GL_INVALID_ENUM);
TexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
ExpectSingleGlError(GL_INVALID_ENUM);
TexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_BASE_LEVEL, 1);
ExpectSingleGlError(GL_INVALID_OPERATION);
// Its only storage is an EGLImage.
TexImage2D(GL_TEXTURE_EXTERNAL_OES, 0, GL_RGBA8, 4, 4, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr);
ExpectSingleGlError(GL_INVALID_ENUM);
GenerateMipmap(GL_TEXTURE_EXTERNAL_OES);
ExpectSingleGlError(GL_INVALID_ENUM);
BindTexture(GL_TEXTURE_EXTERNAL_OES, 0);
BindTexture(GL_TEXTURE_2D, 0);
ActiveTexture(GL_TEXTURE0);
DeleteTextures(2, names);
EXPECT_EQ(GetError(), GL_NO_ERROR);
}
+26 -7
View File
@@ -141,7 +141,7 @@ namespace {
// F2 (P9 W3). The flood: kFloodRecords records published while the peer reads nothing, each of
// which makes the server's apply thread log kFloodWarnLinesPerRecord WARN lines of ~650 bytes and
// one ERROR line (ServerLoop.cpp, MOBILEGL_TEST_APPLY_LOG_FLOOD) - about 16 MiB, against a
// server send buffer this host caps at 4 MiB and a forward queue of 1 MiB for <= WARN.
// deliberately small server send buffer and a forward queue of 1 MiB for <= WARN.
constexpr std::uint32_t kFloodWarnLinesPerRecord = 64;
constexpr std::uint32_t kFloodRecords = 400;
// THE PEER'S RECEIVE BUFFER WHILE IT IS NOT READING. Without this the kernel absorbs the flood
@@ -225,7 +225,10 @@ namespace {
env.emplace_back("MOBILEGL_IPC_ROLE=server");
env.emplace_back("MOBILEGL_IPC_DIAL=no");
env.emplace_back(forwardLog ? "MOBILEGL_IPC_LOG_FORWARD=1" : "MOBILEGL_IPC_LOG_FORWARD=0");
if (floodLines != 0) env.emplace_back("MOBILEGL_TEST_APPLY_LOG_FLOOD=" + std::to_string(floodLines));
if (floodLines != 0) {
env.emplace_back("MOBILEGL_TEST_APPLY_LOG_FLOOD=" + std::to_string(floodLines));
env.emplace_back("MOBILEGL_TEST_LOG_SMALL_SNDBUF=1");
}
env.emplace_back(WaitKnobEntry(waitMs));
env.emplace_back("MOBILEGL_LOG_FILE_PATH=" + logBase);
const std::string image = ServerImage();
@@ -358,6 +361,7 @@ namespace {
// F2 (P9 W3).
kNoControlReader = 22,
kFloodPublish = 23,
kControlBuffer = 24,
};
// A pbuffer context on the server's apply thread, through the same forwarders
@@ -381,11 +385,19 @@ namespace {
// kFloodRecords set_patch_state records are published without waiting. set_patch_state because it
// is pure applier state: no GL call, no reply, no event, so the only thing each one produces on
// the server is the flood's log lines.
int StopReadingControlAndFlood(Client::ClientSession& client) {
int StopReadingControlAndFlood(Client::ClientSession& client, int& originalReceiveBuffer) {
auto* control = client.ControlSocketForTest();
if (control == nullptr) return kNoControlReader;
socklen_t size = sizeof(originalReceiveBuffer);
if (::getsockopt(control->StreamFd(), SOL_SOCKET, SO_RCVBUF, &originalReceiveBuffer, &size) != 0)
return kControlBuffer;
#if defined(__linux__)
// Linux reports twice the SO_RCVBUF request, including bookkeeping space.
originalReceiveBuffer /= 2;
#endif
const int receiveBuffer = kStalledReceiveBufferBytes;
(void)::setsockopt(control->StreamFd(), SOL_SOCKET, SO_RCVBUF, &receiveBuffer, sizeof(receiveBuffer));
if (::setsockopt(control->StreamFd(), SOL_SOCKET, SO_RCVBUF, &receiveBuffer, sizeof(receiveBuffer)) != 0)
return kControlBuffer;
if (!client.PauseControlReaderForTest(true)) return kNoControlReader;
for (std::uint32_t i = 0; i < kFloodRecords; ++i) {
MGP::MGPPatchState patch{};
@@ -900,13 +912,16 @@ namespace {
ConfigureClient(control);
auto& client = Client::ClientSessionInstance();
if (client.StartSpawned() != MOBILEGL_OK) ::_exit(kNoWelcome);
int originalReceiveBuffer = 0;
if (alsoStopDraining) {
const int filled = FillTheEventRingWithoutDraining(client, [&client] { return StopReadingControlAndFlood(client); });
const int filled = FillTheEventRingWithoutDraining(client, [&] {
return StopReadingControlAndFlood(client, originalReceiveBuffer);
});
if (filled != kChildOk) ::_exit(filled);
} else {
const int up = BringUpPbufferContext();
if (up != kChildOk) ::_exit(up);
const int flooded = StopReadingControlAndFlood(client);
const int flooded = StopReadingControlAndFlood(client, originalReceiveBuffer);
if (flooded != kChildOk) ::_exit(flooded);
}
const char one = 1;
@@ -914,7 +929,11 @@ namespace {
char go = 0;
(void)::read(release, &go, 1);
if (alsoStopDraining) ::_exit(kChildOk);
client.PauseControlReaderForTest(false);
// End the receive-window fault before waiting for the ack behind the queued logs.
if (::setsockopt(client.ControlSocketForTest()->StreamFd(), SOL_SOCKET, SO_RCVBUF,
&originalReceiveBuffer, sizeof(originalReceiveBuffer)) != 0)
::_exit(kControlBuffer);
if (!client.PauseControlReaderForTest(false)) ::_exit(kNoControlReader);
client.SyncPeerLog();
(void)::write(synced, &one, 1);
(void)::read(release, &go, 1);
+6 -5
View File
@@ -322,12 +322,13 @@ namespace {
EXPECT_EQ(r.glxLoaded, 1);
EXPECT_EQ(r.clientQueries, 1);
// A process that asked for software rendering is left to the system's other GL, server or not.
// Mesa's software-renderer selection must not disable MobileGL's EGL/GLX entry points.
const EntryReport software = RunEntryPoints(server.endpoint, logBase, true);
EXPECT_EQ(software.vendorDisplay, 0);
EXPECT_EQ(software.getDisplay, 0);
EXPECT_EQ(software.vendorDevices, 0);
EXPECT_EQ(software.glxLoaded, 0);
EXPECT_EQ(software.vendorDisplay, 1);
EXPECT_EQ(software.getDisplay, 1);
EXPECT_EQ(software.vendorDevices, 1);
EXPECT_EQ(software.glxLoaded, 1);
EXPECT_EQ(software.clientQueries, 1);
std::error_code ec;
std::filesystem::remove(Debug::RoleLogPath(logBase.c_str(), Debug::LogRole::Client), ec);
}
+17
View File
@@ -470,6 +470,23 @@ TEST(SessionTest, AppliedSeqAdvancesExactlyOncePerRecordAndIsNeverBatched) {
EXPECT_EQ(applied, kRecords);
}
TEST(SessionTest, TerminalApplyRejectionDoesNotAcknowledgeTheRecord) {
SessionFixture session;
ASSERT_TRUE(session.Build(TestSizes()));
ASSERT_NE(session.cmdProducer.Reserve(1, kRecNone, 8), nullptr);
ASSERT_NE(session.cmdProducer.Reserve(1, kRecNone, 8), nullptr);
session.producer.PublishAndNotify(2);
ASSERT_TRUE(session.consumer.ApplyOne([](const RingRecordView&) { return true; }));
bool checked = false;
EXPECT_FALSE(session.consumer.ApplyOne([&](const RingRecordView&) {
checked = true;
return false;
}));
EXPECT_TRUE(checked);
EXPECT_EQ(session.consumer.AppliedSeq(), 1u);
EXPECT_EQ(session.Control().Progress.appliedSeq.load(), 1u);
}
// retiredSeq: advanced once the SEG_STAGE bytes a record referenced are finished with, and the
// staging allocator reclaims behind it. Late is merely slow; EARLY hands live bytes back to the
// producer, so the advance clamps to appliedSeq rather than believing its caller.
+222 -1
View File
@@ -19,7 +19,10 @@
#include <MG_Remote/Transport/SocketTransport.h>
#include <gtest/gtest.h>
#include <array>
#include <cstring>
#include <memory>
#include <vector>
#include <string>
#include <utility>
@@ -91,7 +94,7 @@ TEST(SharedImage, AnImageDiesWithItsLastReferenceAndIsNoLongerIdentified) {
TEST(SharedImage, UnsupportedFormatsAndSizesAreRefused) {
std::string why;
EXPECT_EQ(SI::Allocate(16, 16, SI::FourCC('N', 'V', '1', '2'), why), nullptr);
EXPECT_EQ(SI::Allocate(16, 16, SI::FourCC('R', 'G', '1', '6'), why), nullptr);
// The ARGB orders are allocatable: the layout is the server's, and X ignores alpha.
SI::ImageRef argb = SI::Allocate(16, 16, SI::kFourccArgb8888, why);
ASSERT_NE(argb, nullptr) << why;
@@ -295,3 +298,221 @@ TEST(SharedImageSync, AFlushPublishesEveryUsedImageAsAWriteAndARead) {
EXPECT_EQ(SI::DupWriteFence(*windowPixmap, &generation), -1);
EXPECT_EQ(generation, 2u);
}
// ---- YUV images (SharedImageYuv.h) ---------------------------------------------------------------
namespace {
// A memfd holding a W x H NV12 frame with padded pitches and a gap before the chroma plane, the
// way a decoder lays its buffers out; sample values are a function of position.
struct ForeignFrame {
int Fd = -1;
SI::PlaneLayout Layout;
SizeT Size = 0;
};
Uint8 LumaAt(Uint32 x, Uint32 y) { return static_cast<Uint8>(x * 3 + y * 7); }
Uint8 CbAt(Uint32 x, Uint32 y) { return static_cast<Uint8>(100 + x + y * 5); }
Uint8 CrAt(Uint32 x, Uint32 y) { return static_cast<Uint8>(200 - x - y * 3); }
ForeignFrame MakeForeignNv12(Uint32 width, Uint32 height, Uint8 bias) {
ForeignFrame frame;
frame.Layout.Count = 2;
frame.Layout.Pitch[0] = width + 32;
frame.Layout.Pitch[1] = width + 64;
frame.Layout.Offset[0] = 0;
frame.Layout.Offset[1] = frame.Layout.Pitch[0] * (height + 8);
frame.Size = frame.Layout.Offset[1] + static_cast<SizeT>(frame.Layout.Pitch[1]) * ((height + 1) / 2);
frame.Fd = ::memfd_create("foreign-nv12", MFD_CLOEXEC);
EXPECT_GE(frame.Fd, 0);
EXPECT_EQ(::ftruncate(frame.Fd, static_cast<off_t>(frame.Size)), 0);
auto* map = static_cast<Uint8*>(::mmap(nullptr, frame.Size, PROT_READ | PROT_WRITE, MAP_SHARED, frame.Fd, 0));
EXPECT_NE(map, MAP_FAILED);
for (Uint32 y = 0; y < height; ++y)
for (Uint32 x = 0; x < width; ++x)
map[y * frame.Layout.Pitch[0] + x] = static_cast<Uint8>(LumaAt(x, y) + bias);
for (Uint32 y = 0; y < (height + 1) / 2; ++y) {
for (Uint32 x = 0; x < (width + 1) / 2; ++x) {
map[frame.Layout.Offset[1] + y * frame.Layout.Pitch[1] + 2 * x] = static_cast<Uint8>(CbAt(x, y) + bias);
map[frame.Layout.Offset[1] + y * frame.Layout.Pitch[1] + 2 * x + 1] =
static_cast<Uint8>(CrAt(x, y) + bias);
}
}
::munmap(map, frame.Size);
return frame;
}
// The image's own samples (a host image is a tight semi-planar memfd) equal the frame's.
void ExpectImageHolds(const SI::Image& image, Uint8 bias) {
const SizeT size = image.Plane1Offset + static_cast<SizeT>(image.Plane1Stride) * ((image.Height + 1) / 2);
auto* map = static_cast<Uint8*>(::mmap(nullptr, size, PROT_READ, MAP_SHARED, image.Fd, 0));
ASSERT_NE(map, MAP_FAILED);
Uint32 wrong = 0;
for (Uint32 y = 0; y < image.Height; ++y)
for (Uint32 x = 0; x < image.Width; ++x)
wrong += map[y * image.Stride + x] != static_cast<Uint8>(LumaAt(x, y) + bias);
for (Uint32 y = 0; y < (image.Height + 1) / 2; ++y) {
for (Uint32 x = 0; x < (image.Width + 1) / 2; ++x) {
wrong += map[image.Plane1Offset + y * image.Plane1Stride + 2 * x] != static_cast<Uint8>(CbAt(x, y) + bias);
wrong += map[image.Plane1Offset + y * image.Plane1Stride + 2 * x + 1] !=
static_cast<Uint8>(CrAt(x, y) + bias);
}
}
EXPECT_EQ(wrong, 0u);
::munmap(map, size);
}
} // namespace
TEST(SharedImageYuv, AnNv12ImageHasTwoPlanesInOneDescriptor) {
std::string why;
ASSERT_TRUE(SI::FourccSupported(SI::kFourccNv12));
SI::ImageRef image = SI::Allocate(64, 32, SI::kFourccNv12, why);
ASSERT_NE(image, nullptr) << why;
EXPECT_TRUE(SI::FourccIsYuv(image->Fourcc));
EXPECT_FALSE(image->IsForeign());
EXPECT_GE(image->Stride, 64u);
EXPECT_GE(image->Plane1Stride, 64u);
EXPECT_GE(image->Plane1Offset, image->Stride * 32u);
// Its own descriptor names it, like any image this server exported.
const int peer = ThroughSocket(image->Fd);
EXPECT_EQ(SI::Identify(peer, why), image);
::close(peer);
}
// THE CPU-COPY FALLBACK: a buffer this server never allocated becomes an image of its own, filled
// from the buffer's planes at each refresh; the same buffer imported again is the same image.
TEST(SharedImageYuv, AForeignBufferIsCopiedIntoAnImageOfItsOwn) {
std::string why;
ForeignFrame frame = MakeForeignNv12(48, 30, 0);
ASSERT_GE(frame.Fd, 0);
// Not one of the server's: Identify refuses it, the foreign import takes it.
EXPECT_EQ(SI::Identify(frame.Fd, why), nullptr);
const int peer = ThroughSocket(frame.Fd);
SI::ImageRef image = SI::ImportForeignYuv(peer, 48, 30, SI::kFourccNv12, frame.Layout, why);
ASSERT_NE(image, nullptr) << why;
EXPECT_TRUE(image->IsForeign());
ASSERT_TRUE(SI::RefreshForeign(*image, why)) << why;
ExpectImageHolds(*image, 0);
// New content in the same buffer shows at the next refresh.
{
ForeignFrame next = MakeForeignNv12(48, 30, 9);
auto* source = static_cast<Uint8*>(::mmap(nullptr, next.Size, PROT_READ, MAP_SHARED, next.Fd, 0));
auto* target = static_cast<Uint8*>(::mmap(nullptr, frame.Size, PROT_READ | PROT_WRITE, MAP_SHARED, frame.Fd, 0));
std::memcpy(target, source, frame.Size);
::munmap(source, next.Size);
::munmap(target, frame.Size);
::close(next.Fd);
}
ASSERT_TRUE(SI::RefreshForeign(*image, why)) << why;
ExpectImageHolds(*image, 9);
// The same buffer with the same planes is the same image; other planes make another one.
EXPECT_EQ(SI::ImportForeignYuv(frame.Fd, 48, 30, SI::kFourccNv12, frame.Layout, why), image);
SI::PlaneLayout shifted = frame.Layout;
shifted.Pitch[1] -= 2;
SI::ImageRef other = SI::ImportForeignYuv(frame.Fd, 48, 30, SI::kFourccNv12, shifted, why);
ASSERT_NE(other, nullptr) << why;
EXPECT_NE(other, image);
::close(peer);
::close(frame.Fd);
}
TEST(SharedImageYuv, AForeignBufferTooSmallForItsPlanesIsRefused) {
std::string why;
ForeignFrame frame = MakeForeignNv12(48, 30, 0);
SI::PlaneLayout beyond = frame.Layout;
beyond.Offset[1] = static_cast<Uint32>(frame.Size);
EXPECT_EQ(SI::ImportForeignYuv(frame.Fd, 48, 30, SI::kFourccNv12, beyond, why), nullptr);
EXPECT_FALSE(why.empty());
SI::PlaneLayout narrow = frame.Layout;
narrow.Pitch[0] = 40; // shorter than a row
EXPECT_EQ(SI::ImportForeignYuv(frame.Fd, 48, 30, SI::kFourccNv12, narrow, why), nullptr);
SI::PlaneLayout onePlane = frame.Layout;
onePlane.Count = 1;
EXPECT_EQ(SI::ImportForeignYuv(frame.Fd, 48, 30, SI::kFourccNv12, onePlane, why), nullptr);
EXPECT_EQ(SI::ImportForeignYuv(frame.Fd, 48, 30, SI::kFourccAbgr8888, frame.Layout, why), nullptr);
::close(frame.Fd);
}
// The plane copy into the layouts a platform's lock can hand out: semi-planar Cb first (copied by
// rows), Cr first and fully planar (sample by sample), and 16-bit samples.
TEST(SharedImageYuv, ThePlaneCopyFillsEveryDestinationLayout) {
constexpr Uint32 W = 6, H = 4;
Uint8 y[H][W], cbcr[H / 2][W];
for (Uint32 r = 0; r < H; ++r)
for (Uint32 c = 0; c < W; ++c) y[r][c] = static_cast<Uint8>(r * 10 + c);
for (Uint32 r = 0; r < H / 2; ++r)
for (Uint32 c = 0; c < W / 2; ++c) {
cbcr[r][2 * c] = static_cast<Uint8>(100 + r * 10 + c);
cbcr[r][2 * c + 1] = static_cast<Uint8>(200 + r * 10 + c);
}
const auto check = [&](const SI::Yuv::PlanarLayout& dst) {
for (Uint32 r = 0; r < H; ++r)
for (Uint32 c = 0; c < W; ++c) EXPECT_EQ(dst.Y.Data[r * dst.Y.RowStride + c * dst.Y.PixelStride], y[r][c]);
for (Uint32 r = 0; r < H / 2; ++r)
for (Uint32 c = 0; c < W / 2; ++c) {
EXPECT_EQ(dst.Cb.Data[r * dst.Cb.RowStride + c * dst.Cb.PixelStride], cbcr[r][2 * c]);
EXPECT_EQ(dst.Cr.Data[r * dst.Cr.RowStride + c * dst.Cr.PixelStride], cbcr[r][2 * c + 1]);
}
};
std::vector<Uint8> luma(16 * H), chroma(16 * H);
SI::Yuv::PlanarLayout nv12{{luma.data(), 16, 1}, {chroma.data(), 16, 2}, {chroma.data() + 1, 16, 2}};
ASSERT_TRUE(SI::Yuv::CopySemiPlanar(&y[0][0], W, &cbcr[0][0], W, W, H, 1, nv12));
check(nv12);
SI::Yuv::PlanarLayout nv21{{luma.data(), 16, 1}, {chroma.data() + 1, 16, 2}, {chroma.data(), 16, 2}};
ASSERT_TRUE(SI::Yuv::CopySemiPlanar(&y[0][0], W, &cbcr[0][0], W, W, H, 1, nv21));
check(nv21);
std::vector<Uint8> cb(8 * H), cr(8 * H);
SI::Yuv::PlanarLayout planar{{luma.data(), 16, 1}, {cb.data(), 8, 1}, {cr.data(), 8, 1}};
ASSERT_TRUE(SI::Yuv::CopySemiPlanar(&y[0][0], W, &cbcr[0][0], W, W, H, 1, planar));
check(planar);
SI::Yuv::PlanarLayout missing{{luma.data(), 16, 1}, {nullptr, 8, 1}, {cr.data(), 8, 1}};
EXPECT_FALSE(SI::Yuv::CopySemiPlanar(&y[0][0], W, &cbcr[0][0], W, W, H, 1, missing));
// 16-bit samples (P010): two bytes each, Cb Cr pairs of four.
Uint16 y16[H][W], c16[H / 2][W];
for (Uint32 r = 0; r < H; ++r)
for (Uint32 c = 0; c < W; ++c) y16[r][c] = static_cast<Uint16>((r * 10 + c) << 6);
for (Uint32 r = 0; r < H / 2; ++r)
for (Uint32 c = 0; c < W; ++c) c16[r][c] = static_cast<Uint16>((500 + r * 10 + c) << 6);
std::vector<Uint16> luma16(16 * H), chroma16(16 * H);
SI::Yuv::PlanarLayout p010{{reinterpret_cast<Uint8*>(luma16.data()), 32, 2},
{reinterpret_cast<Uint8*>(chroma16.data()), 32, 4},
{reinterpret_cast<Uint8*>(chroma16.data()) + 2, 32, 4}};
ASSERT_TRUE(SI::Yuv::CopySemiPlanar(reinterpret_cast<const Uint8*>(&y16[0][0]), W * 2,
reinterpret_cast<const Uint8*>(&c16[0][0]), W * 2, W, H, 2, p010));
for (Uint32 r = 0; r < H; ++r)
for (Uint32 c = 0; c < W; ++c) EXPECT_EQ(luma16[r * 16 + c], y16[r][c]);
for (Uint32 r = 0; r < H / 2; ++r)
for (Uint32 c = 0; c < W; ++c) EXPECT_EQ(chroma16[r * 16 + c], c16[r][c]);
}
// The conversion an import's hints select: BT.709 narrow-range codes of saturated colours decode
// to those colours, and the defaults are BT.601 narrow.
TEST(SharedImageYuv, TheConversionDecodesTheHintedColourSpace) {
const auto decode = [](const SI::Yuv::Conversion& c, double y, double cb, double cr, double max) {
const double in[3] = {y / max - c.Offset[0], cb / max - c.Offset[1], cr / max - c.Offset[2]};
std::array<double, 3> rgb{};
for (int row = 0; row < 3; ++row)
for (int col = 0; col < 3; ++col) rgb[row] += c.Matrix[col * 3 + row] * in[col];
return rgb;
};
SI::Yuv::Hints bt709;
bt709.ColorSpace = SI::Yuv::kEglItuRec709;
bt709.Range = SI::Yuv::kEglYuvNarrowRange;
const auto c709 = SI::Yuv::ConversionFor(bt709, 255.0);
// Red, green, blue and white in BT.709 narrow range (8-bit codes).
const double codes[4][3] = {{63, 102, 240}, {173, 42, 26}, {32, 240, 118}, {235, 128, 128}};
const double want[4][3] = {{1, 0, 0}, {0, 1, 0}, {0, 0, 1}, {1, 1, 1}};
for (int i = 0; i < 4; ++i) {
const auto rgb = decode(c709, codes[i][0], codes[i][1], codes[i][2], 255.0);
for (int k = 0; k < 3; ++k) EXPECT_NEAR(rgb[k], want[i][k], 0.02) << "colour " << i << " channel " << k;
}
// Full range: Y passes through.
SI::Yuv::Hints full;
full.Range = SI::Yuv::kEglYuvFullRange;
const auto cFull = SI::Yuv::ConversionFor(full, 255.0);
const auto grey = decode(cFull, 128, 128, 128, 255.0);
for (int k = 0; k < 3; ++k) EXPECT_NEAR(grey[k], 128.0 / 255.0, 0.005);
EXPECT_EQ(SI::Yuv::Hints{}.EffectiveColorSpace(), SI::Yuv::kEglItuRec601);
EXPECT_FALSE(SI::Yuv::Hints{}.FullRange());
}
@@ -21,6 +21,17 @@
#endif
namespace MobileGL::MG_Util::BackendLoader {
Bool HasExtensionWord(const char* list, const char* word) {
if (list == nullptr || word == nullptr || *word == '\0') return false;
const SizeT n = std::strlen(word);
for (const char* at = std::strstr(list, word); at != nullptr; at = std::strstr(at + 1, word)) {
const Bool startOk = at == list || at[-1] == ' ';
const Bool endOk = at[n] == '\0' || at[n] == ' ';
if (startOk && endOk) return true;
}
return false;
}
static Bool UseAngle() {
return MG_Config::Features.EsprytUseAngle;
}
@@ -170,12 +181,30 @@ namespace MobileGL::MG_Util::BackendLoader {
}
MG_External::EGL::eglSwapBuffersWithDamageEXT_PTR
ResolveSwapBuffersWithDamage(MG_External::EGL::eglSwapBuffersWithDamageEXT_PTR ext,
MG_External::EGL::eglSwapBuffersWithDamageKHR_PTR khr) {
if (ext != nullptr) {
return ext;
ResolveSwapBuffersWithDamage(const MG_External::EGLFunctionsTable& funcs) {
if (!funcs.eglGetCurrentDisplay || !funcs.eglQueryString || !funcs.eglGetProcAddress) {
return nullptr;
}
return (MG_External::EGL::eglSwapBuffersWithDamageEXT_PTR)khr;
const EGLDisplay display = funcs.eglGetCurrentDisplay();
if (display == EGL_NO_DISPLAY) {
return nullptr;
}
const char* extensions = funcs.eglQueryString(display, EGL_EXTENSIONS);
if (HasExtensionWord(extensions, "EGL_EXT_swap_buffers_with_damage")) {
auto ext = reinterpret_cast<MG_External::EGL::eglSwapBuffersWithDamageEXT_PTR>(
funcs.eglGetProcAddress("eglSwapBuffersWithDamageEXT"));
if (ext != nullptr) {
return ext;
}
}
if (HasExtensionWord(extensions, "EGL_KHR_swap_buffers_with_damage")) {
auto khr = reinterpret_cast<MG_External::EGL::eglSwapBuffersWithDamageKHR_PTR>(
funcs.eglGetProcAddress("eglSwapBuffersWithDamageKHR"));
if (khr != nullptr) {
return khr;
}
}
return nullptr;
}
void AcquireGLESFunctions(MG_External::GLESFunctionsTable& funcs,
@@ -806,25 +835,8 @@ namespace MobileGL::MG_Util::BackendLoader {
INIT_EGL_FUNC(eglWaitSync)
}
{
// One slot, two names. EGL_EXT_swap_buffers_with_damage and
// EGL_KHR_swap_buffers_with_damage define the same entry point with the same
// signature, and a driver may expose either, both or neither - Adreno exposes only
// the KHR spelling - so the winner of the EXT -> KHR chain lands in the EXT slot.
// Null when the driver offers neither, which is the normal "no damage-based present
// here" answer: the caller keeps plain eglSwapBuffers (and note that a call with
// n_rects == 0 is defined to be equivalent to eglSwapBuffers anyway).
funcs.eglSwapBuffersWithDamageEXT = ResolveSwapBuffersWithDamage(
reinterpret_cast<MG_External::EGL::eglSwapBuffersWithDamageEXT_PTR>(
resolveEGLProc("eglSwapBuffersWithDamageEXT")),
reinterpret_cast<MG_External::EGL::eglSwapBuffersWithDamageKHR_PTR>(
resolveEGLProc("eglSwapBuffersWithDamageKHR")));
if (funcs.eglSwapBuffersWithDamageEXT == nullptr) {
MGLOG_I_ONCE("This driver has no damage-based present entry point (neither "
"eglSwapBuffersWithDamageEXT nor eglSwapBuffersWithDamageKHR); "
"present calls fall back to eglSwapBuffers");
}
}
// Display extensions can only be checked after the display is initialized.
funcs.eglSwapBuffersWithDamageEXT = nullptr;
}
// Detects whether indirect draws leak the command's baseInstance word ("reserved, must
@@ -124,7 +124,7 @@ namespace MobileGL {
EGL_FUNC_DECL(eglSurfaceAttrib)
EGL_FUNC_DECL(eglSwapBuffers)
// Holds whichever spelling of the damage-based present the driver provided, EXT
// preferred; null when it provides neither (see AcquireEGLFunctions).
// preferred; resolved after making the display current.
EGL_FUNC_DECL(eglSwapBuffersWithDamageEXT)
EGL_FUNC_DECL(eglSwapInterval)
EGL_FUNC_DECL(eglTerminate)
@@ -1364,17 +1364,19 @@ namespace MobileGL {
} // namespace MG_External
namespace MG_Util::BackendLoader {
// Match a complete name in a space-separated GLES/EGL extension string.
Bool HasExtensionWord(const char* list, const char* word);
void AcquireGLESFunctions(MG_External::GLESFunctionsTable& funcs,
MG_External::EGL::eglGetProcAddress_PTR procAddress);
void AcquireEGLFunctions(MG_External::EGLFunctionsTable& funcs);
// The EXT -> KHR -> nothing chain for the damage-based present entry point. A null
// Query the current display's extensions before resolving EXT -> KHR -> nothing.
// Requires a current, initialized EGL display. A null
// result is not a failure: it means this driver offers neither spelling, and the
// caller presents with plain eglSwapBuffers, which a zero-rectangle call to either
// spelling is defined to be equivalent to anyway. Exposed so MG_Test can pin the
// preference order without a driver.
MG_External::EGL::eglSwapBuffersWithDamageEXT_PTR
ResolveSwapBuffersWithDamage(MG_External::EGL::eglSwapBuffersWithDamageEXT_PTR ext,
MG_External::EGL::eglSwapBuffersWithDamageKHR_PTR khr);
ResolveSwapBuffersWithDamage(const MG_External::EGLFunctionsTable& funcs);
Bool FillInGLESCapabilities(MG_External::GLESCapabilities& caps,
const MG_External::GLESFunctionsTable& glesFuncs);
// Detects whether indirect draws leak the command's baseInstance word into
@@ -51,6 +51,8 @@ namespace MobileGL {
case GL_TEXTURE_2D_MULTISAMPLE_ARRAY:
case GL_PROXY_TEXTURE_2D_MULTISAMPLE_ARRAY:
return TextureTarget::Texture2DMultisampleArray;
case GL_TEXTURE_EXTERNAL_OES:
return TextureTarget::External;
default:
return TextureTarget::Unknown;
}
@@ -36,6 +36,8 @@ namespace MobileGL {
return GL_TEXTURE_RECTANGLE;
case TextureTarget::Texture2DMultisampleArray:
return GL_TEXTURE_2D_MULTISAMPLE_ARRAY;
case TextureTarget::External:
return GL_TEXTURE_EXTERNAL_OES;
default:
return GL_UNKNOWN_MGL;
}
@@ -36,6 +36,8 @@ namespace MobileGL {
return "Texture1DArray";
case TextureTarget::Texture2DMultisampleArray:
return "Texture2DMultisampleArray";
case TextureTarget::External:
return "External";
default:
return "Unknown";
}

Some files were not shown because too many files have changed in this diff Show More