[Fix, Test] (ShaderTranspiler): let the image-array select ladder carry an imageSize query, not just a read

This commit is contained in:
2026-08-21 11:49:30 -04:00
parent 77c05b151a
commit 86fdc68efa
3 changed files with 88 additions and 32 deletions
@@ -342,6 +342,21 @@ void main() {
// An imageAtomic* reaches the array through OpImageTexelPointer, and running one per element // An imageAtomic* reaches the array through OpImageTexelPointer, and running one per element
// would perform every other element's atomic as well. The pass has to decline rather than // would perform every other element's atomic as well. The pass has to decline rather than
// lower this. // lower this.
// imageSize() on a dynamically indexed image array. The query carries the image in the same
// leading operand position as an imageLoad and answers with an int vector, so the select
// ladder spells it exactly - and unlike a read it touches no memory at all, so evaluating it
// for every element cannot even return undefined data.
constexpr const char* kUniformIndexedImageSizeQuery = R"(#version 450 core
layout(local_size_x = 1) in;
layout(rgba32f, binding = 0) uniform image2D g_image[4];
layout(rgba32f, binding = 4) uniform image2DArray g_layered[2];
layout(std430, binding = 8) buffer Out { ivec2 size; int layers; } g_out;
uniform int g_index;
void main() {
g_out.size = imageSize(g_image[g_index]);
g_out.layers = imageSize(g_layered[g_index]).z;
}
)";
constexpr const char* kUniformIndexedImageAtomic = R"(#version 450 core constexpr const char* kUniformIndexedImageAtomic = R"(#version 450 core
layout(local_size_x = 1) in; layout(local_size_x = 1) in;
layout(r32ui, binding = 0) uniform uimage2D g_image[4]; layout(r32ui, binding = 0) uniform uimage2D g_image[4];
@@ -514,6 +529,36 @@ TEST(LegalizeResourceArrayIndexPass, LeavesADynamicallyIndexedSamplerArrayByteId
EXPECT_EQ(output, input); EXPECT_EQ(output, input);
} }
// imageSize() on a dynamically indexed image array used to lose the whole stage: the consumer
// whitelist accepted only OpImageRead/OpImageWrite, so the chain was declined and the illegal
// subscript reached the ES compiler intact. It is the same select ladder as a read - the query
// takes the image in in-operand 0 and produces an int vector - and it reads no memory, so the
// elements the shader did not ask for cost nothing but the instruction.
TEST(LegalizeResourceArrayIndexPass, LowersAUniformIndexedImageSizeQueryToSelects) {
const Vector<Uint32> input = CompileCompute(kUniformIndexedImageSizeQuery);
ASSERT_FALSE(input.empty());
EXPECT_TRUE(HasDynamicImageArrayIndex(input));
EXPECT_EQ(CountOpcode(input, spv::Op::OpImageQuerySize), 2u);
EXPECT_EQ(CountOpcode(input, spv::Op::OpSelect), 0u);
Vector<Uint32> output;
ASSERT_TRUE(ShaderCompiler::LegalizeResourceArrayIndexingForEssl(input, output, true));
ASSERT_FALSE(output.empty());
EXPECT_FALSE(HasDynamicImageArrayIndex(output));
// Four elements for g_image and two for g_layered, one query apiece, and one select per
// element past the first of each ladder.
EXPECT_EQ(CountOpcode(output, spv::Op::OpImageQuerySize), 6u);
EXPECT_EQ(CountOpcode(output, spv::Op::OpSelect), 4u);
EXPECT_EQ(CountOpcode(output, spv::Op::OpSwitch), 0u) << "a query produces a value, so no control flow";
EXPECT_TRUE(Validates(output));
const EsslAttempt after = EmitEssl(output);
ASSERT_TRUE(after.succeeded) << after.error;
for (int element = 0; element < 4; ++element) {
EXPECT_NE(after.text.find("g_image[" + std::to_string(element) + "]"), String::npos) << after.text;
}
}
// An imageAtomic* is the shape the lowering must refuse: its per-element rebuild would run every // An imageAtomic* is the shape the lowering must refuse: its per-element rebuild would run every
// other element's read-modify-write. Declining leaves the illegal subscript in place - which is // other element's read-modify-write. Declining leaves the illegal subscript in place - which is
// what the latched warning in LegalizeResourceArrayIndexingForEssl is for - but a half-transform // what the latched warning in LegalizeResourceArrayIndexingForEssl is for - but a half-transform
@@ -737,12 +737,18 @@ namespace MobileGL {
return; return;
case spv::Op::OpImageWrite: case spv::Op::OpImageWrite:
case spv::Op::OpImageRead: case spv::Op::OpImageRead:
// imageSize()/imageSamples() carry the image in the same leading operand
// position as an OpImageRead and produce an int or int vector, so the same
// select ladder rebuilds them exactly - and a size query touches no memory
// at all, which makes evaluating it for every element strictly safer than
// the read the ladder was written for.
case spv::Op::OpImageQuerySize:
case spv::Op::OpImageQuerySizeLod:
if (consumer == nullptr) consumer = user; if (consumer == nullptr) consumer = user;
return; return;
default: default:
// A sampled-image construction, a query, a copy, an argument to a // A sampled-image construction, a copy, an argument to a function:
// function: shapes whose per-element rebuild this pass cannot spell // shapes whose per-element rebuild this pass cannot spell exactly.
// exactly.
unsupportedConsumer = true; unsupportedConsumer = true;
return; return;
} }
@@ -762,7 +768,7 @@ namespace MobileGL {
return consumer->opcode() == spv::Op::OpImageWrite return consumer->opcode() == spv::Op::OpImageWrite
? LowerImageWrite(accessChain, arrayLength, load, consumer) ? LowerImageWrite(accessChain, arrayLength, load, consumer)
: LowerImageRead(accessChain, arrayLength, load, consumer); : LowerImageReadOrQuery(accessChain, arrayLength, load, consumer);
} }
// Drops |load| and |accessChain| once the rewrite above has taken their last user, // Drops |load| and |accessChain| once the rewrite above has taken their last user,
@@ -865,38 +871,42 @@ namespace MobileGL {
return LoweringOutcome::Changed; return LoweringOutcome::Changed;
} }
// A read needs no control flow: read every element through a constant index and pick // A value-producing consumer - an OpImageRead, or an imageSize()/imageSamples() query -
// needs no control flow: run it against every element through a constant index and pick
// with OpSelect. The selection happens on the RESULT, not on the image object - an // with OpSelect. The selection happens on the RESULT, not on the image object - an
// opaque type may not be selected at all (pre-1.4 OpSelect takes pointers, scalars // opaque type may not be selected at all (pre-1.4 OpSelect takes pointers, scalars
// and vectors only, and ESSL has no ternary on an image), so what is duplicated is // and vectors only, and ESSL has no ternary on an image), so what is duplicated is
// the OpImageRead. // the consuming instruction itself. Every such consumer carries the image in in-operand
// 0 and nothing else that is per-element, so one rebuild spells all of them.
// //
// Reading the elements the shader did not ask for is safe: every one of them is an // Running the elements the shader did not ask for is safe: every one of them is an
// image this stage already declares, and GL 4.6 7.11.2 makes a load through an // image this stage already declares, and GL 4.6 7.11.2 makes a load through an
// image unit whose binding is missing or incompatible return undefined DATA - never // image unit whose binding is missing or incompatible return undefined DATA - never
// an error, and never a fault - which the select then discards. Contrast an // an error, and never a fault - which the select then discards. A size query does not
// imageAtomic*, which LowerImageChain refuses for exactly the opposite reason. // even touch memory. Contrast an imageAtomic*, which LowerImageChain refuses for
// exactly the opposite reason.
LegalizeResourceArrayIndexPass::LoweringOutcome LegalizeResourceArrayIndexPass::LoweringOutcome
LegalizeResourceArrayIndexPass::LowerImageRead(Instruction* accessChain, uint32_t arrayLength, LegalizeResourceArrayIndexPass::LowerImageReadOrQuery(Instruction* accessChain,
Instruction* load, Instruction* imageRead) { uint32_t arrayLength, Instruction* load,
Instruction* consumer) {
auto* irContext = context(); auto* irContext = context();
uint32_t conditionTypeId = 0; uint32_t conditionTypeId = 0;
uint32_t dimension = 0; uint32_t dimension = 0;
if (!TryGetSelectConditionType(irContext, imageRead->type_id(), &conditionTypeId, if (!TryGetSelectConditionType(irContext, consumer->type_id(), &conditionTypeId,
&dimension)) { &dimension)) {
MGLOG_D("[spirv] image array index: read type is not selectable, declining"); MGLOG_D("[spirv] image array index: result type is not selectable, declining");
return LoweringOutcome::Declined; return LoweringOutcome::Declined;
} }
const uint32_t boolTypeId = irContext->get_type_mgr()->GetBoolTypeId(); const uint32_t boolTypeId = irContext->get_type_mgr()->GetBoolTypeId();
const uint32_t indexId = accessChain->GetSingleWordInOperand(1); const uint32_t indexId = accessChain->GetSingleWordInOperand(1);
const uint32_t imageTypeId = load->type_id(); const uint32_t imageTypeId = load->type_id();
std::vector<Operand> tailOperands; std::vector<Operand> tailOperands;
for (uint32_t i = 1; i < imageRead->NumInOperands(); ++i) { for (uint32_t i = 1; i < consumer->NumInOperands(); ++i) {
tailOperands.push_back(imageRead->GetInOperand(i)); tailOperands.push_back(consumer->GetInOperand(i));
} }
InstructionBuilder builder( InstructionBuilder builder(
irContext, imageRead, irContext, consumer,
IRContext::kAnalysisDefUse | IRContext::kAnalysisInstrToBlockMapping); IRContext::kAnalysisDefUse | IRContext::kAnalysisInstrToBlockMapping);
uint32_t selectedId = 0; uint32_t selectedId = 0;
@@ -906,18 +916,18 @@ namespace MobileGL {
CloneChainWithConstantIndex(builder, irContext, accessChain, constantId); CloneChainWithConstantIndex(builder, irContext, accessChain, constantId);
Instruction* elementImage = builder.AddLoad(imageTypeId, elementChain->result_id()); Instruction* elementImage = builder.AddLoad(imageTypeId, elementChain->result_id());
std::vector<Operand> readOperands; std::vector<Operand> elementOperands;
readOperands.push_back({SPV_OPERAND_TYPE_ID, {elementImage->result_id()}}); elementOperands.push_back({SPV_OPERAND_TYPE_ID, {elementImage->result_id()}});
for (const Operand& tailOperand : tailOperands) { for (const Operand& tailOperand : tailOperands) {
readOperands.push_back(tailOperand); elementOperands.push_back(tailOperand);
} }
Instruction* elementRead = builder.AddInstruction( Instruction* elementResult = builder.AddInstruction(
MakeUnique<Instruction>(irContext, spv::Op::OpImageRead, imageRead->type_id(), MakeUnique<Instruction>(irContext, consumer->opcode(), consumer->type_id(),
irContext->TakeNextId(), readOperands)); irContext->TakeNextId(), elementOperands));
if (element == 0) { if (element == 0) {
// Element 0 is the else-arm of the whole ladder, so an out-of-range // Element 0 is the else-arm of the whole ladder, so an out-of-range
// index reads it - an undefined element for an undefined index. // index reads it - an undefined element for an undefined index.
selectedId = elementRead->result_id(); selectedId = elementResult->result_id();
continue; continue;
} }
@@ -929,17 +939,17 @@ namespace MobileGL {
conditionId = builder.AddCompositeConstruct(conditionTypeId, components)->result_id(); conditionId = builder.AddCompositeConstruct(conditionTypeId, components)->result_id();
} }
selectedId = builder selectedId = builder
.AddSelect(imageRead->type_id(), conditionId, .AddSelect(consumer->type_id(), conditionId,
elementRead->result_id(), selectedId) elementResult->result_id(), selectedId)
->result_id(); ->result_id();
} }
irContext->ReplaceAllUsesWith(imageRead->result_id(), selectedId); irContext->ReplaceAllUsesWith(consumer->result_id(), selectedId);
irContext->KillInst(imageRead); irContext->KillInst(consumer);
KillImageChainIfDead(accessChain, load); KillImageChainIfDead(accessChain, load);
irContext->InvalidateAnalysesExceptFor(IRContext::kAnalysisNone); irContext->InvalidateAnalysesExceptFor(IRContext::kAnalysisNone);
MGLOG_D("[spirv] image array index: lowered a dynamic imageLoad to %u constant-indexed " MGLOG_D("[spirv] image array index: lowered a dynamic image read/query to %u "
"reads", "constant-indexed operations",
arrayLength); arrayLength);
return LoweringOutcome::Changed; return LoweringOutcome::Changed;
} }
@@ -146,9 +146,10 @@ namespace MobileGL {
LoweringOutcome LowerImageWrite(spvtools::opt::Instruction* accessChain, uint32_t arrayLength, LoweringOutcome LowerImageWrite(spvtools::opt::Instruction* accessChain, uint32_t arrayLength,
spvtools::opt::Instruction* load, spvtools::opt::Instruction* load,
spvtools::opt::Instruction* imageWrite); spvtools::opt::Instruction* imageWrite);
LoweringOutcome LowerImageRead(spvtools::opt::Instruction* accessChain, uint32_t arrayLength, LoweringOutcome LowerImageReadOrQuery(spvtools::opt::Instruction* accessChain,
spvtools::opt::Instruction* load, uint32_t arrayLength,
spvtools::opt::Instruction* imageRead); spvtools::opt::Instruction* load,
spvtools::opt::Instruction* consumer);
Mode m_mode; Mode m_mode;
}; };