From a6deb5ed82cb8fc6b4f4c5c5fee264e09f97ff89 Mon Sep 17 00:00:00 2001 From: Yong He Date: Mon, 29 Sep 2025 17:45:08 -0700 Subject: Rewriting the lower-buffer-element-type pass to avoid unnecessary packing/unpacking. (#8526) Part of the effort to improve the performance of generated SPIRV code. The existing lower-buffer-element-type pass works by loading the entire buffer element content from memory, and translate it to logical type stored in a local variable at the earliest reference of a buffer handle. This means that is can generate inefficient code that reads more than necessary. Consider this example: ``` struct BigStruct { bool values[1024]; } ConstantBuffer cb; void test(BigStruct v) { if (v.values[0]) { printf("ok"); } } [numthreads(1,1,1)] void computeMain() { test(cb); } ``` In IR, the `computeMain` function before lower-buffer-element-type pass is something like following: ``` func test: %v = param : BigStruct %barr = fieldExtract(%v, "values") %element = elementExtract(%barr, 0) ... // uses %element func computeMain: %v = load(cb) call %test %v ``` The existing lower-buffer-element-type pass will rewrite the bool array in `BigStruct` into `int` array so it is legal in SPIRV. However, it does so by inserting the translation on the first `load` of the constant buffer: ``` struct BigStruct_std430 { int values[1024]; } var cb : ConstantBuffer; func computeMain: %tmpVar : var call %unpackStorage(%tmpVar, cb) %v : BigStruct = load %tmpVar call %test %v ``` This means that the entire array will be loaded and translated to int, before calling `test`, which only uses one element. It turns out that the downstream compiler isn't always able to optimize out this inefficient translation/copy. This PR completely rewrites the way buffer-element-type lowering is handled to avoid producing this inefficient code. It works in two parts: first we turn on the `transformParamsToConstRef` pass for SPIRV target as well, so we will translate the `test` function to take the `v` parameter as `constref`. The second part is a redesigned buffer-element-type pass that defers the storage-type to logical-type translation until a value is actually used by a `load` instruction. In this example, after `transformParamsToConstRef`, the IR is: ``` func test: %v = param : ConstRef %barr = fieldAddr(%v, "values") %elementPtr = elementAddr(%barr, 0) %element = load(%elementPtr) ... // uses %element func computeMain: call %test %cb ``` The new `buffer-element-type-lowering` pass will take this IR, and insert translation at latest possible time across the entire call graph, and translate the IR into: ``` func test: %v = param : ConstRef %barr = fieldAddr(%v, "values") %elementPtr : ptr = elementAddr(%barr, 0) %element_int = load(%elementPtr) %element = cast(%element_int) : %bool ... // uses %element func computeMain: call %test %cb ``` In this new IR, there is no longer a load and conversion of the entire array. See new comment in `slang-ir-lower-buffer-element-type.cpp` for more details of how the pass works. This PR also address many other issues surfaced by turning on `transformParamsToConstRef` pass on SPIRV backend. --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com> --- source/slang/slang-ir-defer-buffer-load.cpp | 38 +++-------------------------- 1 file changed, 3 insertions(+), 35 deletions(-) (limited to 'source/slang/slang-ir-defer-buffer-load.cpp') diff --git a/source/slang/slang-ir-defer-buffer-load.cpp b/source/slang/slang-ir-defer-buffer-load.cpp index e71892fe5..51c6a161b 100644 --- a/source/slang/slang-ir-defer-buffer-load.cpp +++ b/source/slang/slang-ir-defer-buffer-load.cpp @@ -67,38 +67,6 @@ struct DeferBufferLoadContext return result; } - static bool isImmutableLocation(IRInst* loc) - { - switch (loc->getOp()) - { - case kIROp_GetStructuredBufferPtr: - case kIROp_ImageSubscript: - return isImmutableLocation(loc->getOperand(0)); - default: - break; - } - - auto type = loc->getDataType(); - if (!type) - return false; - - switch (type->getOp()) - { - case kIROp_HLSLStructuredBufferType: - case kIROp_HLSLByteAddressBufferType: - case kIROp_ConstantBufferType: - case kIROp_ParameterBlockType: - return true; - default: - break; - } - - if (auto textureType = as(type)) - return textureType->getAccess() == SLANG_RESOURCE_ACCESS_READ; - - return false; - } - static bool isImmutableBufferLoad(IRInst* inst) { // Note: we cannot defer loads from RWStructuredBuffer because there can be other @@ -111,7 +79,7 @@ struct DeferBufferLoadContext case kIROp_Load: { auto rootAddr = getRootAddr(inst->getOperand(0)); - return isImmutableLocation(rootAddr); + return isPointerToImmutableLocation(rootAddr); } default: return false; @@ -132,14 +100,14 @@ struct DeferBufferLoadContext align = load->findAttr(); if (!as(ptr->getParent())) { - builder.setInsertAfter(ptr); + setInsertAfterOrdinaryInst(&builder, ptr); IRType* valueType = tryGetPointedToType(&builder, ptr->getFullType()); result = builder.emitLoad(valueType, ptr, align); mapPtrToValue[ptr] = result; } else { - builder.setInsertBefore(loadInst); + setInsertBeforeOrdinaryInst(&builder, loadInst); IRType* valueType = tryGetPointedToType(&builder, ptr->getFullType()); result = builder.emitLoad(valueType, ptr, align); // Since we are inserting the load in a local scope, we can't register -- cgit v1.2.3