summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--lib/SPIRV/SPRVWriter.cpp6
-rw-r--r--lib/SPIRV/libSPIRV/SPRVModule.cpp16
-rw-r--r--test/image-unoptimized.ll76
-rw-r--r--test/store.ll4
-rw-r--r--tools/llvm-spirv/CMakeLists.txt1
5 files changed, 93 insertions, 10 deletions
diff --git a/lib/SPIRV/SPRVWriter.cpp b/lib/SPIRV/SPRVWriter.cpp
index ca0e1b6..44ed346 100644
--- a/lib/SPIRV/SPRVWriter.cpp
+++ b/lib/SPIRV/SPRVWriter.cpp
@@ -86,6 +86,10 @@ using namespace llvm;
using namespace SPRV;
using namespace OCLUtil;
+namespace llvm {
+ FunctionPass *createPromoteMemoryToRegisterPass();
+}
+
namespace SPRV{
bool SPRVDbgSaveRegularizedModule = false;
@@ -2140,10 +2144,10 @@ LLVMToSPRV::dumpUsers(Value* V) {
SPRVDBG(dbgs() << " " << **UI << '\n');
}
-
void
LLVMToSPRV::oclRegularize() {
PassManager PassMgr;
+ PassMgr.add(createPromoteMemoryToRegisterPass());
PassMgr.add(createSPRVRegularizeOCL20());
PassMgr.add(createSPRVLowerOCLBlocks());
PassMgr.add(createSPRVLowerBool());
diff --git a/lib/SPIRV/libSPIRV/SPRVModule.cpp b/lib/SPIRV/libSPIRV/SPRVModule.cpp
index 87880a2..ec86627 100644
--- a/lib/SPIRV/libSPIRV/SPRVModule.cpp
+++ b/lib/SPIRV/libSPIRV/SPRVModule.cpp
@@ -376,14 +376,14 @@ SPRVModuleImpl::optimizeDecorates() {
continue;
}
SPRVDBG(bildbgs() << " add deco group. erase equal range\n");
- auto G = new SPRVDecorationGroup(this, getId());
- std::vector<SPRVId> Targets;
- Targets.push_back(D->getTargetId());
- const_cast<SPRVDecorateGeneric*>(D)->setTargetId(G->getId());
- G->getDecorations().insert(D);
- for (I = ER.first; I != ER.second; ++I) {
- auto E = *I;
- if (*E == *D)
+ auto G = new SPRVDecorationGroup(this, getId());
+ std::vector<SPRVId> Targets;
+ Targets.push_back(D->getTargetId());
+ const_cast<SPRVDecorateGeneric*>(D)->setTargetId(G->getId());
+ G->getDecorations().insert(D);
+ for (I = ER.first; I != ER.second; ++I) {
+ auto E = *I;
+ if (*E == *D)
continue;
Targets.push_back(E->getTargetId());
}
diff --git a/test/image-unoptimized.ll b/test/image-unoptimized.ll
new file mode 100644
index 0000000..f3c1be4
--- /dev/null
+++ b/test/image-unoptimized.ll
@@ -0,0 +1,76 @@
+; RUN: llvm-as < %s | llvm-spirv -spirv-text -o %t
+; RUN: FileCheck < %t %s
+target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir-unknown-unknown"
+
+%opencl.image2d_t = type opaque
+
+; Function Attrs: nounwind
+; CHECK: {{[0-9]*}} Store
+; CHECK-NEXT: 1 Return
+define spir_kernel void @test_fn(%opencl.image2d_t addrspace(1)* %srcimg, i32 %sampler, <4 x float> addrspace(1)* %results) #0 {
+entry:
+ %srcimg.addr = alloca %opencl.image2d_t addrspace(1)*, align 4
+ %sampler.addr = alloca i32, align 4
+ %results.addr = alloca <4 x float> addrspace(1)*, align 4
+ %tid_x = alloca i32, align 4
+ %tid_y = alloca i32, align 4
+ %.compoundliteral = alloca <2 x i32>, align 8
+ store %opencl.image2d_t addrspace(1)* %srcimg, %opencl.image2d_t addrspace(1)** %srcimg.addr, align 4
+ store i32 %sampler, i32* %sampler.addr, align 4
+ store <4 x float> addrspace(1)* %results, <4 x float> addrspace(1)** %results.addr, align 4
+ %call = call spir_func i32 @_Z13get_global_idj(i32 0) #2
+ store i32 %call, i32* %tid_x, align 4
+ %call1 = call spir_func i32 @_Z13get_global_idj(i32 1) #2
+ store i32 %call1, i32* %tid_y, align 4
+ %0 = load %opencl.image2d_t addrspace(1)** %srcimg.addr, align 4
+ %1 = load i32* %sampler.addr, align 4
+ %2 = load i32* %tid_x, align 4
+ %vecinit = insertelement <2 x i32> undef, i32 %2, i32 0
+ %3 = load i32* %tid_y, align 4
+ %vecinit2 = insertelement <2 x i32> %vecinit, i32 %3, i32 1
+ store <2 x i32> %vecinit2, <2 x i32>* %.compoundliteral
+ %4 = load <2 x i32>* %.compoundliteral
+ %call3 = call spir_func <4 x float> @_Z11read_imagef11ocl_image2d11ocl_samplerDv2_i(%opencl.image2d_t addrspace(1)* %0, i32 %1, <2 x i32> %4) #2
+ %5 = load i32* %tid_y, align 4
+ %6 = load %opencl.image2d_t addrspace(1)** %srcimg.addr, align 4
+ %call4 = call spir_func i32 @_Z15get_image_width11ocl_image2d(%opencl.image2d_t addrspace(1)* %6) #2
+ %mul = mul nsw i32 %5, %call4
+ %7 = load i32* %tid_x, align 4
+ %add = add nsw i32 %mul, %7
+ %8 = load <4 x float> addrspace(1)** %results.addr, align 4
+ %arrayidx = getelementptr inbounds <4 x float> addrspace(1)* %8, i32 %add
+ store <4 x float> %call3, <4 x float> addrspace(1)* %arrayidx, align 16
+ ret void
+}
+
+; Function Attrs: nounwind readnone
+declare spir_func i32 @_Z13get_global_idj(i32) #1
+
+; Function Attrs: nounwind readnone
+declare spir_func <4 x float> @_Z11read_imagef11ocl_image2d11ocl_samplerDv2_i(%opencl.image2d_t addrspace(1)*, i32, <2 x i32>) #1
+
+; Function Attrs: nounwind readnone
+declare spir_func i32 @_Z15get_image_width11ocl_image2d(%opencl.image2d_t addrspace(1)*) #1
+
+attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-realign-stack" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { nounwind readnone "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-realign-stack" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #2 = { nounwind readnone }
+
+!opencl.kernels = !{!0}
+!opencl.enable.FP_CONTRACT = !{}
+!opencl.spir.version = !{!6}
+!opencl.ocl.version = !{!6}
+!opencl.used.extensions = !{!7}
+!opencl.used.optional.core.features = !{!8}
+!opencl.compiler.options = !{!7}
+
+!0 = !{void (%opencl.image2d_t addrspace(1)*, i32, <4 x float> addrspace(1)*)* @test_fn, !1, !2, !3, !4, !5}
+!1 = !{!"kernel_arg_addr_space", i32 1, i32 0, i32 1}
+!2 = !{!"kernel_arg_access_qual", !"read_only", !"none", !"none"}
+!3 = !{!"kernel_arg_type", !"image2d_t", !"sampler_t", !"float4*"}
+!4 = !{!"kernel_arg_base_type", !"image2d_t", !"sampler_t", !"float4*"}
+!5 = !{!"kernel_arg_type_qual", !"", !"", !""}
+!6 = !{i32 1, i32 2}
+!7 = !{}
+!8 = !{!"cl_images"}
diff --git a/test/store.ll b/test/store.ll
index 3c4b328..7da3cc4 100644
--- a/test/store.ll
+++ b/test/store.ll
@@ -8,8 +8,10 @@ target triple = "spir-unknown-unknown"
define spir_kernel void @foo(i32 addrspace(1)* %a) #0 {
entry:
%a.addr = alloca i32 addrspace(1)*, align 4
-; CHECK: 5 Store {{[0-9]+}} {{[0-9]+}} 2 4
store i32 addrspace(1)* %a, i32 addrspace(1)** %a.addr, align 4
+ %0 = load i32 addrspace(1)** %a.addr, align 4
+; CHECK: 5 Store {{[0-9]+}} {{[0-9]+}} 2 4
+ store i32 0, i32 addrspace(1)* %0, align 4
ret void
}
diff --git a/tools/llvm-spirv/CMakeLists.txt b/tools/llvm-spirv/CMakeLists.txt
index eec7607..d868689 100644
--- a/tools/llvm-spirv/CMakeLists.txt
+++ b/tools/llvm-spirv/CMakeLists.txt
@@ -6,6 +6,7 @@ set(LLVM_LINK_COMPONENTS
SPIRVLib
Core
Support
+ TransformUtils
)
add_llvm_tool(llvm-spirv
generated by cgit v1.2.3 (git 2.39.1) at 2025-05-21 19:52:07 +0000