Skip to content

[SYCL-MLIR]: Remove mlir inlining pass #7536

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Nov 25, 2022
Merged

[SYCL-MLIR]: Remove mlir inlining pass #7536

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
4cc7004
Remove inliner pass
etiotto Nov 24, 2022
4402e37
Update polygeist/tools/cgeist/Test/Verification/no_inline.c
etiotto Nov 24, 2022
5f92795
Update polygeist/tools/cgeist/Test/Verification/stream.cu
etiotto Nov 24, 2022
7a9f555
Update polygeist/tools/cgeist/Test/Verification/stream.cu
etiotto Nov 24, 2022
6f6066d
Address code review comments
etiotto Nov 24, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions polygeist/tools/cgeist/Test/Verification/base_with_virt.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@ void a() {
// CHECK: func @_ZN16mbasic_stringbufC1Ev(%arg0: !llvm.ptr<struct<(struct<packed (ptr<ptr<func<i32 (...)>>>, i32, array<4 x i8>)>, struct<(ptr<i8>)>)>>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
// CHECK-NEXT: %0 = llvm.getelementptr %arg0[0, 0] : (!llvm.ptr<struct<(struct<packed (ptr<ptr<func<i32 (...)>>>, i32, array<4 x i8>)>, struct<(ptr<i8>)>)>>) -> !llvm.ptr<struct<packed (ptr<ptr<func<i32 (...)>>>, i32, array<4 x i8>)>>
// CHECK-NEXT: call @_ZN1AC1Ev(%0) : (!llvm.ptr<struct<packed (ptr<ptr<func<i32 (...)>>>, i32, array<4 x i8>)>>) -> ()
// CHECK: call @_ZN12_Alloc_hiderC1Ev
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK: func @_ZN1AC1Ev(%arg0: !llvm.ptr<struct<packed (ptr<ptr<func<i32 (...)>>>, i32, array<4 x i8>)>>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
Expand Down
3 changes: 3 additions & 0 deletions polygeist/tools/cgeist/Test/Verification/base_with_virt2.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,9 +36,12 @@ void a() {

// clang-format off
// CHECK: func @_Z1av() attributes {llvm.linkage = #llvm.linkage<external>} {
// CHECK: call @_ZN16mbasic_stringbufC1Ev
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK: func @_ZN16mbasic_stringbufC1Ev(%arg0: !llvm.ptr<struct<(struct<(ptr<ptr<func<i32 (...)>>>)>, struct<(ptr<i8>)>)>>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
// CHECK: call @_ZN15basic_streambufC1Ev
// CHECK: call @_ZN12_Alloc_hiderC1Ev
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK: func @_ZN15basic_streambufC1Ev(%arg0: !llvm.ptr<struct<(ptr<ptr<func<i32 (...)>>>)>>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
Expand Down
13 changes: 3 additions & 10 deletions polygeist/tools/cgeist/Test/Verification/classrefmem.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: cgeist %s -O2 --function=* -S | FileCheck %s
// RUN: cgeist %s -O0 --function=* -S | FileCheck %s

extern int& moo;
void oadd(int& x) {
Expand All @@ -24,18 +24,11 @@ void Q(A& a) {
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK: func @_Z1QR1A(%arg0: memref<?x1xmemref<?xi32>>) attributes {llvm.linkage = #llvm.linkage<external>} {
// CHECK-NEXT: %c1_i32 = arith.constant 1 : i32
// CHECK-NEXT: %0 = affine.load %arg0[0, 0] : memref<?x1xmemref<?xi32>>
// CHECK-NEXT: %1 = affine.load %0[0] : memref<?xi32>
// CHECK-NEXT: %2 = arith.addi %1, %c1_i32 : i32
// CHECK-NEXT: affine.store %2, %0[0] : memref<?xi32>
// CHECK-NEXT: call @_ZN1A3addEv(%arg0) : (memref<?x1xmemref<?xi32>>) -> ()
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK: func @_ZN1A3addEv(%arg0: memref<?x1xmemref<?xi32>>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
// CHECK-NEXT: %c1_i32 = arith.constant 1 : i32
// CHECK-NEXT: %0 = affine.load %arg0[0, 0] : memref<?x1xmemref<?xi32>>
// CHECK-NEXT: %1 = affine.load %0[0] : memref<?xi32>
// CHECK-NEXT: %2 = arith.addi %1, %c1_i32 : i32
// CHECK-NEXT: affine.store %2, %0[0] : memref<?xi32>
// CHECK-NEXT: call @_Z4oaddRi(%0) : (memref<?xi32>) -> ()
// CHECK-NEXT: return
// CHECK-NEXT: }
19 changes: 7 additions & 12 deletions polygeist/tools/cgeist/Test/Verification/consabi.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,18 +22,13 @@ QStream ilaunch_kernel(QStream x) {
// CHECK-NEXT: return %1 : !llvm.struct<(struct<(f64, f64)>, i32)>
// CHECK-NEXT: }
// CHECK-NEXT: func @_ZN7QStreamC1EOS_(%arg0: !llvm.ptr<struct<(struct<(f64, f64)>, i32)>>, %arg1: !llvm.ptr<struct<(struct<(f64, f64)>, i32)>>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
// CHECK-NEXT: %0 = llvm.bitcast %arg1 : !llvm.ptr<struct<(struct<(f64, f64)>, i32)>> to !llvm.ptr<f64>
// CHECK-NEXT: %1 = llvm.load %0 : !llvm.ptr<f64>
// CHECK-NEXT: %2 = llvm.bitcast %arg0 : !llvm.ptr<struct<(struct<(f64, f64)>, i32)>> to !llvm.ptr<f64>
// CHECK-NEXT: llvm.store %1, %2 : !llvm.ptr<f64>
// CHECK-NEXT: %[[a4:.+]] = llvm.getelementptr %0[1] : (!llvm.ptr<f64>) -> !llvm.ptr<f64>
// CHECK-NEXT: %[[a5:.+]] = llvm.load %[[a4]] : !llvm.ptr<f64>
// CHECK-NEXT: %[[a7:.+]] = llvm.getelementptr %2[1] : (!llvm.ptr<f64>) -> !llvm.ptr<f64>
// CHECK-NEXT: llvm.store %[[a5]], %[[a7]] : !llvm.ptr<f64>
// CHECK-NEXT: %[[a8:.+]] = llvm.getelementptr %arg1[0, 1] : (!llvm.ptr<struct<(struct<(f64, f64)>, i32)>>) -> !llvm.ptr<i32>
// CHECK-NEXT: %[[a9:.+]] = llvm.load %[[a8]] : !llvm.ptr<i32>
// CHECK-NEXT: %[[a10:.+]] = llvm.getelementptr %arg0[0, 1] : (!llvm.ptr<struct<(struct<(f64, f64)>, i32)>>) -> !llvm.ptr<i32>
// CHECK-NEXT: llvm.store %[[a9]], %[[a10]] : !llvm.ptr<i32>
// CHECK-NEXT: %0 = "polygeist.pointer2memref"(%arg0) : (!llvm.ptr<struct<(struct<(f64, f64)>, i32)>>) -> memref<?x2xf64>
// CHECK-NEXT: %1 = "polygeist.pointer2memref"(%arg1) : (!llvm.ptr<struct<(struct<(f64, f64)>, i32)>>) -> memref<?x2xf64>
// CHECK-NEXT: call @_ZN1DC1EOS_(%0, %1) : (memref<?x2xf64>, memref<?x2xf64>) -> ()
// CHECK-NEXT: %2 = llvm.getelementptr %arg1[0, 1] : (!llvm.ptr<struct<(struct<(f64, f64)>, i32)>>) -> !llvm.ptr<i32>
// CHECK-NEXT: %3 = llvm.load %2 : !llvm.ptr<i32>
// CHECK-NEXT: %4 = llvm.getelementptr %arg0[0, 1] : (!llvm.ptr<struct<(struct<(f64, f64)>, i32)>>) -> !llvm.ptr<i32>
// CHECK-NEXT: llvm.store %3, %4 : !llvm.ptr<i32>
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK-NEXT: func @_ZN1DC1EOS_(%arg0: memref<?x2xf64>, %arg1: memref<?x2xf64>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
Expand Down
9 changes: 7 additions & 2 deletions polygeist/tools/cgeist/Test/Verification/cudaglobalcodegen.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,14 +14,19 @@ __global__ void bar(int * a)
void baz(int * a){
bar<<<dim3(1,1,1), dim3(1,1,1)>>>(a);
}

// CHECK: func private @_Z18__device_stub__barPi(%arg0: memref<?xi32>)
// CHECK-NEXT: call @_Z3barPi(%arg0) : (memref<?xi32>) -> ()
// CHECK-NEXT: return
// CHECK-NEXT: }

// CHECK: func.func private @_Z3barPi(%arg0: memref<?xi32>) attributes {llvm.linkage = #llvm.linkage<external>} {
// CHECK-NEXT: %c1_i32 = arith.constant 1 : i32
// CHECK-NEXT: affine.store %c1_i32, %arg0[0] : memref<?xi32>
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK: func @_Z3bazPi(%arg0: memref<?xi32>) attributes {llvm.linkage = #llvm.linkage<external>} {
// CHECK-NEXT: %c1 = arith.constant 1 : index
// CHECK-NEXT: gpu.launch blocks(%arg1, %arg2, %arg3) in (%arg7 = %c1, %arg8 = %c1, %arg9 = %c1) threads(%arg4, %arg5, %arg6) in (%arg10 = %c1, %arg11 = %c1, %arg12 = %c1) {
// CHECK: gpu.launch blocks(%arg1, %arg2, %arg3) in (%arg7 = %1, %arg8 = %3, %arg9 = %5) threads(%arg4, %arg5, %arg6) in (%arg10 = %7, %arg11 = %9, %arg12 = %11) {
// CHECK-NEXT: call @_Z18__device_stub__barPi(%arg0) : (memref<?xi32>) -> ()
// CHECK-NEXT: gpu.terminator
// CHECK-NEXT: }
Expand Down
16 changes: 12 additions & 4 deletions polygeist/tools/cgeist/Test/Verification/ident.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,9 +65,14 @@ void lt_kernel_cuda(MTensorIterator& iter) {
// CHECK-NEXT: return %2 : i8
// CHECK-NEXT: }
// CHECK: func private @_ZZ14lt_kernel_cudaENK3$_0clEv(%arg0: !llvm.ptr<!llvm.struct<(!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>)>>) attributes {llvm.linkage = #llvm.linkage<internal>} {
// CHECK-NEXT: %0 = llvm.getelementptr %arg0[0, 0] : (!llvm.ptr<!llvm.struct<(!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>)>>) -> !llvm.ptr<!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>>
// CHECK-NEXT: %1 = llvm.load %0 : !llvm.ptr<!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>>
// CHECK-NEXT: %2 = call @_ZNK15MTensorIterator6deviceEv(%1) : (!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>) -> i8
// CHECK-NEXT: %c1_i64 = arith.constant 1 : i64
// CHECK-NEXT: %0 = llvm.alloca %c1_i64 x !llvm.struct<(i8)> : (i64) -> !llvm.ptr<struct<(i8)>>
// CHECK-NEXT: %1 = llvm.alloca %c1_i64 x !llvm.struct<(i8)> : (i64) -> !llvm.ptr<struct<(i8)>>
// CHECK-NEXT: %2 = llvm.getelementptr %arg0[0, 0] : (!llvm.ptr<!llvm.struct<(!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>)>>) -> !llvm.ptr<!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>>
// CHECK-NEXT: %3 = llvm.load %2 : !llvm.ptr<!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>>
// CHECK-NEXT: %4 = llvm.load %1 : !llvm.ptr<struct<(i8)>>
// CHECK-NEXT: llvm.store %4, %0 : !llvm.ptr<struct<(i8)>>
// CHECK-NEXT: call @_Z11igpu_kernelIZZ14lt_kernel_cudaENK3$_0clEvEUlvE_EvR15MTensorIteratorRKT_(%3, %0) : (!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>, !llvm.ptr<struct<(i8)>>) -> ()
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK: func @_ZNK12MSmallVectorI12MOperandInfoEixEi(%arg0: memref<?x1xmemref<?x2xi8>>, %arg1: i32) -> memref<?x2xi8> attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
Expand All @@ -76,11 +81,14 @@ void lt_kernel_cuda(MTensorIterator& iter) {
// CHECK-NEXT: %2 = "polygeist.subindex"(%0, %1) : (memref<?x2xi8>, index) -> memref<?x2xi8>
// CHECK-NEXT: return %2 : memref<?x2xi8>
// CHECK-NEXT: }
// CHECK: func.func private @_Z11igpu_kernelIZZ14lt_kernel_cudaENK3$_0clEvEUlvE_EvR15MTensorIteratorRKT_(%arg0: !llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>, %arg1: !llvm.ptr<struct<(i8)>>) attributes {llvm.linkage = #llvm.linkage<internal>} {
// CHECK-NEXT: %0 = call @_ZNK15MTensorIterator6deviceEv(%arg0) : (!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>) -> i8
// CHECK-NEXT: return
// CHECK-NEXT: }
// CHECK-NEXT: func @_ZNK15MTensorIterator6deviceEv(%arg0: !llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>) -> i8 attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
// CHECK-NEXT: %c0_i32 = arith.constant 0 : i32
// CHECK-NEXT: %0 = "polygeist.pointer2memref"(%arg0) : (!llvm.ptr<!llvm.struct<(!llvm.struct<(memref<?x2xi8>)>)>>) -> memref<?x1xmemref<?x2xi8>>
// CHECK-NEXT: %1 = call @_ZNK12MSmallVectorI12MOperandInfoEixEi(%0, %c0_i32) : (memref<?x1xmemref<?x2xi8>>, i32) -> memref<?x2xi8>
// CHECK-NEXT: %2 = affine.load %1[0, 0] : memref<?x2xi8>
// CHECK-NEXT: return %2 : i8
// CHECK-NEXT: }

2 changes: 1 addition & 1 deletion polygeist/tools/cgeist/Test/Verification/no_inline.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@ void foo(int A[10]) {
// CHECK-LABEL: func @main()
// CHECK: call @foo
// OPT1-LABEL: func @main()
// OPT1-NOT: call @foo
// OPT1: call @foo
int main() {
int A[10];
foo(A);
Expand Down
11 changes: 7 additions & 4 deletions polygeist/tools/cgeist/Test/Verification/refptrabi.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,10 +18,13 @@ float ll(void* data) {

// CHECK: func @ll(%arg0: !llvm.ptr<i8>) -> f32 attributes {llvm.linkage = #llvm.linkage<external>} {
// CHECK-NEXT: %alloca = memref.alloca() : memref<1x1xi16>
// CHECK-NEXT: %0 = llvm.bitcast %arg0 : !llvm.ptr<i8> to !llvm.ptr<i16>
// CHECK-NEXT: %1 = llvm.load %0 : !llvm.ptr<i16>
// CHECK-NEXT: %alloca_0 = memref.alloca() : memref<1x1xi16>
// CHECK-NEXT: %cast = memref.cast %alloca_0 : memref<1x1xi16> to memref<?x1xi16>
// CHECK-NEXT: %0 = "polygeist.pointer2memref"(%arg0) : (!llvm.ptr<i8>) -> memref<?x1xi16>
// CHECK-NEXT: call @_ZN4HalfC1ERKS_(%cast, %0) : (memref<?x1xi16>, memref<?x1xi16>) -> ()
// CHECK-NEXT: %1 = affine.load %alloca_0[0, 0] : memref<1x1xi16>
// CHECK-NEXT: affine.store %1, %alloca[0, 0] : memref<1x1xi16>
// CHECK-NEXT: %cast = memref.cast %alloca : memref<1x1xi16> to memref<?x1xi16>
// CHECK-NEXT: %2 = call @thing(%cast) : (memref<?x1xi16>) -> f32
// CHECK-NEXT: %cast_1 = memref.cast %alloca : memref<1x1xi16> to memref<?x1xi16>
// CHECK-NEXT: %2 = call @thing(%cast_1) : (memref<?x1xi16>) -> f32
// CHECK-NEXT: return %2 : f32
// CHECK-NEXT: }
7 changes: 2 additions & 5 deletions polygeist/tools/cgeist/Test/Verification/stream.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,11 +14,8 @@ void run(cudaStream_t stream1, int *array, int n) {
}

// CHECK: func.func @_Z3runP10cudaStreamPii(%arg0: !llvm.ptr<struct<()>>, %arg1: memref<?xi32>, %arg2: i32) attributes {llvm.linkage = #llvm.linkage<external>} {
// CHECK-NEXT: %c10 = arith.constant 10 : index
// CHECK-NEXT: %c1 = arith.constant 1 : index
// CHECK-NEXT: %c20 = arith.constant 20 : index
// CHECK-NEXT: %0 = "polygeist.stream2token"(%arg0) : (!llvm.ptr<struct<()>>) -> !gpu.async.token
// CHECK-NEXT: %1 = gpu.launch async [%0] blocks(%arg3, %arg4, %arg5) in (%arg9 = %c10, %arg10 = %c1, %arg11 = %c1) threads(%arg6, %arg7, %arg8) in (%arg12 = %c20, %arg13 = %c1, %arg14 = %c1) {
// CHECK: %12 = "polygeist.stream2token"(%arg0) : (!llvm.ptr<struct<()>>) -> !gpu.async.token
// CHECK-NEXT: %13 = gpu.launch async [%12] blocks(%arg3, %arg4, %arg5) in (%arg9 = %1, %arg10 = %3, %arg11 = %5) threads(%arg6, %arg7, %arg8) in (%arg12 = %7, %arg13 = %9, %arg14 = %11) {
// CHECK-NEXT: func.call @_Z21__device_stub__squarePii(%arg1, %arg2) : (memref<?xi32>, i32) -> ()
// CHECK-NEXT: gpu.terminator
// CHECK-NEXT: }
Expand Down
4 changes: 2 additions & 2 deletions polygeist/tools/cgeist/driver.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -346,7 +346,7 @@ static int optimize(mlir::MLIRContext &context,
if (RaiseToAffine)
optPM.addPass(mlir::createLowerAffinePass());
optPM.addPass(mlir::createCanonicalizerPass(canonicalizerConfig, {}, {}));
pm.addPass(mlir::createInlinerPass());

mlir::OpPassManager &optPM2 = pm.nestAny();
optPM2.addPass(mlir::createCanonicalizerPass(canonicalizerConfig, {}, {}));
optPM2.addPass(mlir::createCSEPass());
Expand Down Expand Up @@ -407,7 +407,7 @@ static int optimizeCUDA(mlir::MLIRContext &context,
noptPM.addPass(mlir::createCanonicalizerPass(canonicalizerConfig, {}, {}));
noptPM.addPass(polygeist::createMem2RegPass());
noptPM.addPass(mlir::createCanonicalizerPass(canonicalizerConfig, {}, {}));
pm.addPass(mlir::createInlinerPass());

mlir::OpPassManager &noptPM2 = pm.nestAny();
noptPM.addPass(mlir::createCanonicalizerPass(canonicalizerConfig, {}, {}));
noptPM2.addPass(polygeist::createMem2RegPass());
Expand Down