From 9bd125aeec7c1b72386534ac75ab8bc68801e437 Mon Sep 17 00:00:00 2001 From: Swung0x48 Date: Fri, 21 Aug 2026 21:21:54 -0400 Subject: [PATCH] [Fix, Test] (ShaderTranspiler): carry r11f_g11f_b10f storage images in rgba16f instead of losing the stage --- .../WidenImageFormatsTest.cpp | 108 +++++++++++++++--- .../SpirvPasses/WidenImageFormatsPass.cpp | 45 ++++++-- .../SpirvPasses/WidenImageFormatsPass.h | 23 +++- 3 files changed, 148 insertions(+), 28 deletions(-) diff --git a/MobileGL/MG_Test/ShaderTranspiler/WidenImageFormatsTest.cpp b/MobileGL/MG_Test/ShaderTranspiler/WidenImageFormatsTest.cpp index 8f94f164..e73c346d 100644 --- a/MobileGL/MG_Test/ShaderTranspiler/WidenImageFormatsTest.cpp +++ b/MobileGL/MG_Test/ShaderTranspiler/WidenImageFormatsTest.cpp @@ -14,10 +14,10 @@ // and every draw with the program silently renders nothing while GL_LINK_STATUS still says TRUE. // // What has to hold is the emulation's exactness, in three parts at once: the DECLARED format must -// become the core carrier of the same per-channel width, every imageStore through it must have its -// surplus components replaced by GL's own (0.., 1) so the carrier's extra channels never hold -// anything GL has not defined, and every imageLoad must come back masked the same way. A module -// that declares only core formats - or one of the nine formats with no exact carrier - must come +// become a core carrier that loses nothing, every imageStore through it must have its surplus +// components replaced by GL's own (0.., 1) so the carrier's extra channels never hold anything GL +// has not defined, and every imageLoad must come back masked the same way. A module that declares +// only core formats - or one of the eight formats with no lossless carrier at all - must come // out untouched, because widening those would be an approximation rather than an emulation. Real // GLSL through the same glslang path the backends use, for the same reason // ClampMultisampleFetchTest.cpp does it: what matters is what glslang actually emits. @@ -229,9 +229,23 @@ void main() { } )"; - // rg16 is one of the NINE with no core carrier of the same per-channel width. Widening it - // would change the quantisation an application sees, so it must be left alone and keep the - // honest "no GLSL ES spelling" diagnostic instead. + // r11f_g11f_b10f: THREE float channels in a packed 32-bit word, and the only format the four + // CTS allFormats/allTargets walkers still aborted on after the channel widening landed - it + // has no core carrier of the same per-channel width, so it took rgba16f, whose 5-bit exponent + // and longer mantissa represent every 11f and 10f value exactly. + const char* const kR11fG11fB10fLoadStore = R"(#version 430 core +layout(r11f_g11f_b10f, binding = 0) uniform image2D img; +out vec4 fragColor; +void main() { + vec4 texel = imageLoad(img, ivec2(gl_FragCoord.xy)); + imageStore(img, ivec2(gl_FragCoord.xy), vec4(1.0, 2.0, 3.0, 4.0)); + fragColor = texel; +} +)"; + + // rg16 is one of the EIGHT with no core carrier at all - core ESSL has no 16-bit normalized + // format, so every candidate loses range or changes the component type the texture presents. + // It must be left alone and keep the honest "no GLSL ES spelling" diagnostic instead. const char* const kRg16LoadStore = R"(#version 430 core layout(rg16, binding = 0) uniform image2D img; out vec4 fragColor; @@ -247,7 +261,7 @@ void main() { // the shader rewrite, the ES texture storage and the glBindImageTexture argument. If it drifts // the three stop agreeing, and a narrow texture read through a wide image goes out of bounds // silently on every driver tested. -TEST(WidenImageFormats, SeventeenNonCoreFormatsHaveAnExactSameWidthCarrier) { +TEST(WidenImageFormats, EighteenNonCoreFormatsHaveALosslessCoreCarrier) { struct Case { Uint requested; Uint carrier; @@ -272,6 +286,10 @@ TEST(WidenImageFormats, SeventeenNonCoreFormatsHaveAnExactSameWidthCarrier) { {0x8234, 0x8D76, 1, "GL_R16UI -> GL_RGBA16UI"}, {0x8238, 0x8D7C, 2, "GL_RG8UI -> GL_RGBA8UI"}, {0x8232, 0x8D7C, 1, "GL_R8UI -> GL_RGBA8UI"}, + // The one entry that is a re-encoding rather than a channel widening: 11f is e5m6 and 10f + // is e5m5 against a half's s1e5m10, so the carrier is still lossless - and three channels, + // so the mask has to pin only alpha. + {0x8C3A, 0x881A, 3, "GL_R11F_G11F_B10F -> GL_RGBA16F"}, }; for (const Case& testCase : cases) { EXPECT_EQ(ShaderCompiler::WidenedCoreEsslImageFormat(testCase.requested), testCase.carrier) @@ -287,7 +305,7 @@ TEST(WidenImageFormats, SeventeenNonCoreFormatsHaveAnExactSameWidthCarrier) { } } -TEST(WidenImageFormats, CoreFormatsAndTheNineWithoutAnExactCarrierAreRefused) { +TEST(WidenImageFormats, CoreFormatsAndTheEightWithoutALosslessCarrierAreRefused) { // The thirteen GLSL ES already has: nothing to carry. for (const Uint coreFormat : {0x8814u /*RGBA32F*/, 0x881Au /*RGBA16F*/, 0x822Eu /*R32F*/, 0x8058u /*RGBA8*/, 0x8F97u /*RGBA8_SNORM*/, 0x8D82u /*RGBA32I*/, @@ -297,10 +315,12 @@ TEST(WidenImageFormats, CoreFormatsAndTheNineWithoutAnExactCarrierAreRefused) { EXPECT_EQ(ShaderCompiler::WidenedCoreEsslImageFormat(coreFormat), 0u) << "core format 0x" << std::hex << coreFormat; } - // The nine with no core format of the same per-channel width. Carrying these would be an - // approximation - a different quantisation, or a different numeric domain for anything that - // samples the same texture - so they are deliberately left to the honest diagnostic. - for (const Uint hardFormat : {0x8C3Au /*R11F_G11F_B10F*/, 0x8059u /*RGB10_A2*/, + // The eight with no LOSSLESS core carrier: core ESSL has no 16-bit normalized format and no + // 10-bit one, so every candidate for these either loses range or changes the component type + // the texture presents to anything that samples it. Deliberately left to the honest + // diagnostic. r11f_g11f_b10f is NOT among them - rgba16f holds every value it can, so it is + // carried above. + for (const Uint hardFormat : {0x8059u /*RGB10_A2*/, 0x906Fu /*RGB10_A2UI*/, 0x805Bu /*RGBA16*/, 0x822Cu /*RG16*/, 0x822Au /*R16*/, 0x8F9Bu /*RGBA16_SNORM*/, 0x8F99u /*RG16_SNORM*/, 0x8F98u /*R16_SNORM*/}) { @@ -357,6 +377,68 @@ TEST(WidenImageFormats, TwoChannelFloatImageBecomesRgba32fWithBothAccessesMasked << "the mask must be a separate value, or it would feed itself"; } +// The three-channel case, which no format exercised before r11f_g11f_b10f was carried: only ALPHA +// is surplus, so the mask must take r, g and b from the texel and nothing but the fourth component +// from the (0, 0, 0, 1) constant. A mask that zeroed blue here - the shape a two-channel format +// wants - would silently drop the third channel of every store. +TEST(WidenImageFormats, ThreeChannelPackedFloatImageBecomesRgba16fWithOnlyAlphaPinned) { + const Vector spirv = CompileFragment(kR11fG11fB10fLoadStore); + ASSERT_FALSE(spirv.empty()); + ASSERT_TRUE(ShaderCompiler::DeclaresWidenableImageFormat(spirv)); + + const auto beforeTypes = CollectStorageImageTypes(spirv); + ASSERT_EQ(beforeTypes.size(), 1u); + EXPECT_EQ(beforeTypes.front().format, static_cast(spv::ImageFormat::R11fG11fB10f)); + + Vector widened; + ASSERT_TRUE(ShaderCompiler::WidenImageFormatsForEssl(spirv, widened, /*onlyFormatsSpirvCrossRefusesToPrint=*/false, + /*enableSpirvValidation=*/true)); + ASSERT_FALSE(widened.empty()); + EXPECT_TRUE(Validates(widened)); + EXPECT_FALSE(ShaderCompiler::DeclaresWidenableImageFormat(widened)); + + const auto afterTypes = CollectStorageImageTypes(widened); + ASSERT_EQ(afterTypes.size(), 1u); + EXPECT_EQ(afterTypes.front().format, static_cast(spv::ImageFormat::Rgba16f)); + + const auto shuffles = CollectVectorShuffles(widened); + + const auto texelIds = CollectImageWriteTexelIds(widened); + ASSERT_EQ(texelIds.size(), 1u); + const VectorShuffle* storeMask = FindShuffleWithResult(shuffles, texelIds.front()); + ASSERT_NE(storeMask, nullptr) << "the imageStore texel is not a masked value"; + EXPECT_TRUE(HasComponents(*storeMask, {0u, 1u, 2u, 7u})) + << "expected (r, g, b, 1) - components 0, 1 and 2 of the texel, then 3 of (0,0,0,1)"; + + const auto readIds = CollectImageReadResultIds(widened); + ASSERT_EQ(readIds.size(), 1u); + const VectorShuffle* loadMask = FindShuffleOver(shuffles, readIds.front()); + ASSERT_NE(loadMask, nullptr) << "the imageLoad result is consumed unmasked"; + EXPECT_TRUE(HasComponents(*loadMask, {0u, 1u, 2u, 7u})); +} + +// ...and the same module through the emitter, which is where the failure actually showed: ESSL has +// no `r11f_g11f_b10f` token, SPIRV-Cross throws for it, and the throw took every image uniform +// declared in the same stage with it. +TEST(WidenImageFormats, PackedFloatImageOnlyReachesEsslThroughTheCarrier) { + const Vector spirv = CompileFragment(kR11fG11fB10fLoadStore); + ASSERT_FALSE(spirv.empty()); + + const EsslAttempt before = EmitEssl(spirv); + EXPECT_FALSE(before.succeeded) + << "SPIRV-Cross printed r11f_g11f_b10f for an ES target; the widening's premise has " + "changed:\n" + << before.text; + + Vector widened; + ASSERT_TRUE(ShaderCompiler::WidenImageFormatsForEssl(spirv, widened, /*onlyFormatsSpirvCrossRefusesToPrint=*/false, + /*enableSpirvValidation=*/true)); + const EsslAttempt after = EmitEssl(widened); + ASSERT_TRUE(after.succeeded) << after.error; + EXPECT_NE(after.text.find("rgba16f"), String::npos) << after.text; + EXPECT_EQ(after.text.find("r11f_g11f_b10f"), String::npos) << after.text; +} + TEST(WidenImageFormats, SingleChannelUnsignedImageBecomesRgba8uiWithBothAccessesMasked) { const Vector spirv = CompileFragment(kR8uiLoadStore); ASSERT_FALSE(spirv.empty()); diff --git a/MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/WidenImageFormatsPass.cpp b/MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/WidenImageFormatsPass.cpp index 2b074fd5..174896a8 100644 --- a/MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/WidenImageFormatsPass.cpp +++ b/MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/WidenImageFormatsPass.cpp @@ -50,16 +50,40 @@ namespace MobileGL { constexpr uint32_t kImageAccessImageOperand = 0; constexpr uint32_t kImageWriteTexelOperand = 2; - // The exact carrier of a non-core image format: the core GLSL ES format with the - // SAME component type and the SAME per-channel width, differing only in channel - // count. `channels` is what the original format really has, which is what every - // access through the carrier is masked back to. + // The carrier of a non-core image format: a core GLSL ES format that represents + // every value the original can hold, WITHOUT LOSS. `channels` is what the original + // format really has, which is what every access through the carrier is masked back + // to. // - // Only formats that widen EXACTLY appear here. r11f_g11f_b10f, rgb10_a2, - // rgb10_a2ui, rgba16, rg16, r16, rgba16_snorm, rg16_snorm and r16_snorm have no - // same-width core carrier - every candidate is either lossy or changes the numeric - // domain a sampler would read - and are deliberately absent, so they keep the - // honest "no GLSL ES spelling" diagnostic rather than a silent approximation. + // Almost every entry is a pure CHANNEL widening - same component type, same + // per-channel width, more channels (rg32f -> rgba32f) - and for those the carrier + // is bit-exact: the storage holds the identical encoding, only wider. + // + // r11f_g11f_b10f is the one entry that is not. It has no same-width core carrier, + // so it takes rgba16f, and the two encodings differ. What matters is that the + // carrier is still LOSSLESS: an 11-bit float is e5m6 and a 10-bit float is e5m5, + // while a half is s1e5m10 - the SAME 5-bit exponent with a strictly longer + // mantissa - so every value the packed format can represent has an exact half. + // Nothing an application stores is rounded away. + // + // What DOES change is the reverse direction: the carrier can hold values the + // packed format could not - negatives (11f and 10f are unsigned), and mantissa + // bits finer than the 6 and 5 the format quantises to - so a value written through + // the image and then SAMPLED comes back on half's grid rather than the packed + // format's. That is a strictly finer grid, never a lossy one, and it is measured + // against the alternative, which is not a more faithful quantisation but no + // program at all: `layout(r11f_g11f_b10f)` has no ESSL spelling, SPIRV-Cross + // throws for it, and the stage - with every other image uniform declared beside it + // - is lost (KHR-GL43.shader_image_load_store.basic-allFormats-*, which fail on + // this format alone, and multiple-uniforms, where one such declaration killed a + // program holding eight images). + // + // The remaining eight - rgb10_a2, rgb10_a2ui, rgba16, rg16, r16, rgba16_snorm, + // rg16_snorm and r16_snorm - stay absent, and for a stronger reason than + // quantisation: core ESSL has no 16-bit normalized format at all and no 10-bit + // one, so every candidate carrier for them either loses range or changes the + // component TYPE the texture presents. They keep the honest "no GLSL ES spelling" + // diagnostic rather than a silent approximation. struct ImageFormatWidening { spv::ImageFormat Carrier = spv::ImageFormat::Unknown; uint32_t Channels = 0; @@ -73,6 +97,9 @@ namespace MobileGL { case spv::ImageFormat::Rg32f: return {spv::ImageFormat::Rgba32f, 2}; case spv::ImageFormat::Rg16f: return {spv::ImageFormat::Rgba16f, 2}; case spv::ImageFormat::R16f: return {spv::ImageFormat::Rgba16f, 1}; + // Not a channel widening but a lossless re-encoding - see above. Three + // channels, so the fourth reads as the 1 GL defines for a format without one. + case spv::ImageFormat::R11fG11fB10f: return {spv::ImageFormat::Rgba16f, 3}; // Unsigned normalized. case spv::ImageFormat::Rg8: return {spv::ImageFormat::Rgba8, 2}; case spv::ImageFormat::R8: return {spv::ImageFormat::Rgba8, 1}; diff --git a/MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/WidenImageFormatsPass.h b/MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/WidenImageFormatsPass.h index f5d7be0a..e684dc0f 100644 --- a/MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/WidenImageFormatsPass.h +++ b/MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/WidenImageFormatsPass.h @@ -54,12 +54,23 @@ namespace MobileGL { // alone survives storage this shader never wrote (glTexStorage with no upload, whose // surplus channels are undefined). // - // The other NINE (r11f_g11f_b10f, rgb10_a2, rgb10_a2ui, rgba16, rg16, r16, - // rgba16_snorm, rg16_snorm, r16_snorm) have NO same-width core carrier and are - // deliberately NOT widened here: every carrier for them is either lossy or changes the - // numeric domain of the texture a `sampler2D` would read from it. They keep the honest - // "no GLSL ES spelling" diagnostic instead of silently changing an application's - // quantisation behaviour. + // r11f_g11f_b10f has no same-width core carrier either, and takes rgba16f anyway, + // because that carrier is still LOSSLESS: 11f is e5m6 and 10f is e5m5 against a half's + // s1e5m10 - the SAME 5-bit exponent with a strictly longer mantissa - so every value + // the packed format can hold has an exact half. Only the reverse direction differs + // (the carrier also holds negatives, which 11f and 10f cannot sign, and mantissa bits + // finer than the 6 and 5 they quantise to, so a value written through the image and + // then SAMPLED lands on half's grid rather than the packed format's). That is measured + // against the alternative, which is not a truer quantisation but no program at all: + // the SPIRV-Cross throw takes the whole stage, every image uniform declared beside it + // included. + // + // The other EIGHT (rgb10_a2, rgb10_a2ui, rgba16, rg16, r16, rgba16_snorm, rg16_snorm, + // r16_snorm) are deliberately NOT widened here: core ESSL has no 16-bit normalized + // format at all and no 10-bit one, so every carrier for them either loses range or + // changes the component TYPE the texture a `sampler2D` would read presents. They keep + // the honest "no GLSL ES spelling" diagnostic instead of silently changing an + // application's numeric domain. // // MUST MOVE WITH THE OTHER TWO LAYERS. The widening is not a shader-local rewrite: the // ES texture behind the image has to be allocated in the carrier format too, and