diff options
| author | Yong He <yonghe@outlook.com> | 2025-09-29 17:45:08 -0700 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-09-30 00:45:08 +0000 |
| commit | a6deb5ed82cb8fc6b4f4c5c5fee264e09f97ff89 (patch) | |
| tree | 1c374bd52498cad2e142e3c7f5482fd42dca966f /source/slang/slang-ir-spirv-legalize.cpp | |
| parent | 2827c94de5901cac42a67f73a78ab2548771b28c (diff) | |
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<BigStruct> 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<BigStruct_std430>;
func computeMain:
%tmpVar : var<BigStruct>
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<BigStruct>
%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<BigStruct_std430>
%barr = fieldAddr(%v, "values")
%elementPtr : ptr<int> = 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>
Diffstat (limited to 'source/slang/slang-ir-spirv-legalize.cpp')
| -rw-r--r-- | source/slang/slang-ir-spirv-legalize.cpp | 46 |
1 files changed, 25 insertions, 21 deletions
diff --git a/source/slang/slang-ir-spirv-legalize.cpp b/source/slang/slang-ir-spirv-legalize.cpp index 2c4bd11cc..f795a6559 100644 --- a/source/slang/slang-ir-spirv-legalize.cpp +++ b/source/slang/slang-ir-spirv-legalize.cpp @@ -893,18 +893,22 @@ struct SPIRVLegalizationContext : public SourceEmitterBase IRBuilder builder(inst); builder.setInsertBefore(inst); auto funcType = as<IRFuncType>(funcValue->getFullType()); + bool argsChanged = false; for (UInt i = 0; i < inst->getArgCount(); i++) { auto arg = inst->getArg(i); auto paramType = funcType->getParamType(i); - if (as<IRPtrType>(paramType)) + if (auto ptrType = as<IRPtrType>(paramType)) { - // If the parameter has an explicit pointer type, - // then we know the user is using the variable pointer - // capability to pass a true pointer. - // In this case we should not rewrite the call. - newArgs.add(arg); - continue; + if (ptrType->getAddressSpace() == AddressSpace::UserPointer) + { + // If the parameter has an explicit pointer type, + // then we know the user is using the variable pointer + // capability to pass a true pointer. + // In this case we should not rewrite the call. + newArgs.add(arg); + continue; + } } auto ptrType = as<IRPtrTypeBase>(arg->getDataType()); if (!as<IRPtrTypeBase>(arg->getDataType())) @@ -953,13 +957,26 @@ struct SPIRVLegalizationContext : public SourceEmitterBase // If we reach here, we need to allocate a temp var. auto tempVar = builder.emitVar(ptrType->getValueType()); + builder.addDecoration(tempVar, kIROp_DisableCopyEliminationDecoration); auto load = builder.emitLoad(arg); builder.emitStore(tempVar, load); newArgs.add(tempVar); + argsChanged = true; + + // We may need to write the value back to the original pointer argument + // after the call. + // + // If callee doesn't modify the memory location, no need to write back. + if (funcType && funcType->getParamCount() > i && + as<IRConstRefType>(funcType->getParamType(i))) + continue; + // If the buffer location is immutable, don't write back. + if (isPointerToImmutableLocation(root)) + continue; writeBacks.add(WriteBackPair{arg, tempVar}); } SLANG_ASSERT((UInt)newArgs.getCount() == inst->getArgCount()); - if (writeBacks.getCount()) + if (argsChanged) { auto newCall = builder.emitCallInst(inst->getFullType(), inst->getCallee(), newArgs); for (auto wb : writeBacks) @@ -2297,19 +2314,6 @@ struct SPIRVLegalizationContext : public SourceEmitterBase // so we need to update the function types to match that. updateFunctionTypes(); - // Lower all loads/stores from buffer pointers to use correct storage types. - // We didn't do the lowering for buffer pointers because we don't know which pointer - // types are actual storage buffer pointers until we propagated the address space of - // pointers in this pass. In the future we should consider separate out IRAddress as - // the type for IRVar, and use IRPtrType to dedicate pointers in user code, so we can - // safely lower the pointer load stores early together with other buffer types. - BufferElementTypeLoweringOptions bufferElementTypeLoweringOptions; - bufferElementTypeLoweringOptions.lowerBufferPointer = true; - lowerBufferElementTypeToStorageType( - m_sharedContext->m_targetProgram, - m_module, - bufferElementTypeLoweringOptions); - // Look for structs that are both used as fields and marked with Block // decorations, and move the Block decoration to a wrapper struct. legalizeStructBlocks(); |
