Skip to content
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

[TEST] Use LABEL scopes in lit tests #3293

Merged
merged 1 commit into from
Mar 6, 2024
Merged

[TEST] Use LABEL scopes in lit tests #3293

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
6 changes: 3 additions & 3 deletions test/Conversion/triton_ops.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
// RUN: triton-opt %s | FileCheck %s

// CHECK: #[[WMMA:.*]] = #triton_gpu.amd_wmma
// CHECK: #[[$WMMA:.*]] = #triton_gpu.amd_wmma

tt.func @cast_ops(%scalar_ptr: !tt.ptr<f32>, %scalar_f32: f32, %scalar_i64: i64) {
// scalar -> scalar
Expand Down Expand Up @@ -241,14 +241,14 @@ module attributes {"triton_gpu.compute-capability" = 0 : i32, "triton_gpu.num-ct
// CHECK-LABEL: wmma_layout
tt.func @wmma_layout(%0: tensor<16x16xf16, #blocked>) {
%1 = triton_gpu.convert_layout %0 : tensor<16x16xf16, #blocked> -> tensor<16x16xf16, #triton_gpu.amd_wmma<{warpsPerCTA = [1, 1]}>>
// CHECK: %{{.+}} = triton_gpu.convert_layout %{{.+}} : tensor<16x16xf16, #{{.+}}> -> tensor<16x16xf16, #[[WMMA]]>
// CHECK: %{{.+}} = triton_gpu.convert_layout %{{.+}} : tensor<16x16xf16, #{{.+}}> -> tensor<16x16xf16, #[[$WMMA]]>
tt.return
}

// CHECK-LABEL: wmma_dot_op_layout
tt.func @wmma_dot_op_layout(%0: tensor<16x16xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #blocked}>>) {
%1 = triton_gpu.convert_layout %0 : tensor<16x16xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #blocked}>> -> tensor<16x16xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #triton_gpu.amd_wmma<{warpsPerCTA = [1, 1]}>}>>
// CHECK: %{{.+}} = triton_gpu.convert_layout %{{.+}} : tensor<16x16xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #{{.+}}}>> -> tensor<16x16xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #[[WMMA]]}>>
// CHECK: %{{.+}} = triton_gpu.convert_layout %{{.+}} : tensor<16x16xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #{{.+}}}>> -> tensor<16x16xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #[[$WMMA]]}>>
tt.return
}
}
14 changes: 7 additions & 7 deletions test/TritonGPU/accelerate-matmul.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,7 +48,7 @@ module attributes {"triton_gpu.compute-capability" = 90 : i32, "triton_gpu.num-c

// -----

// CHECK-80: #[[MMA:.+]] = #triton_gpu.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [1, 8], instrShape = [16, 8]}>
// CHECK-80: #[[$MMA:.+]] = #triton_gpu.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [1, 8], instrShape = [16, 8]}>
#blocked = #triton_gpu.blocked<{sizePerThread = [4, 4], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}>
#blocked1 = #triton_gpu.blocked<{sizePerThread = [4, 4], threadsPerWarp = [1, 32], warpsPerCTA = [8, 1], order = [1, 0]}>
#blocked2 = #triton_gpu.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}>
Expand All @@ -60,12 +60,12 @@ module attributes {"triton_gpu.compute-capability" = 80 : i32, "triton_gpu.num-c
%arg2: tensor<64x128xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #blocked1}>>) -> tensor<64x128xf32, #blocked1> {
%cst_0 = arith.constant dense<0.000000e+00> : tensor<64x64xf32, #blocked>
%cst_1 = arith.constant dense<0.000000e+00> : tensor<64x128xf32, #blocked1>
// CHECK-80: tt.dot {{.*}} -> tensor<64x64xf32, #[[MMA]]>
// CHECK-80: tt.dot {{.*}} -> tensor<64x64xf32, #[[$MMA]]>
%d = tt.dot %arg0, %arg1, %cst_0 {allowTF32 = true, maxNumImpreciseAcc = 0 : i32} :
tensor<64x128xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #blocked}>> * tensor<128x64xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #blocked}>> -> tensor<64x64xf32, #blocked>
%t = arith.truncf %d : tensor<64x64xf32, #blocked> to tensor<64x64xf16, #blocked>
%c = triton_gpu.convert_layout %t : tensor<64x64xf16, #blocked> -> tensor<64x64xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #blocked1}>>
// CHECK-80: tt.dot {{.*}} -> tensor<64x128xf32, #[[MMA]]>
// CHECK-80: tt.dot {{.*}} -> tensor<64x128xf32, #[[$MMA]]>
%r = tt.dot %c, %arg2, %cst_1 {allowTF32 = true, maxNumImpreciseAcc = 0 : i32} :
tensor<64x64xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #blocked1}>> * tensor<64x128xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #blocked1}>> -> tensor<64x128xf32, #blocked1>
tt.return %r : tensor<64x128xf32, #blocked1>
Expand All @@ -74,7 +74,7 @@ module attributes {"triton_gpu.compute-capability" = 80 : i32, "triton_gpu.num-c

// -----

// CHECK-80: #[[MMA:.+]] = #triton_gpu.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [4, 2], instrShape = [16, 8]}>
// CHECK-80: #[[$MMA:.+]] = #triton_gpu.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [4, 2], instrShape = [16, 8]}>
#blocked = #triton_gpu.blocked<{sizePerThread = [4, 4], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}>
#blocked1 = #triton_gpu.blocked<{sizePerThread = [4, 4], threadsPerWarp = [1, 32], warpsPerCTA = [8, 1], order = [1, 0]}>
#blocked2 = #triton_gpu.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}>
Expand All @@ -85,9 +85,9 @@ module attributes {"triton_gpu.compute-capability" = 80 : i32, "triton_gpu.num-c
%arg1: tensor<128x64xf8E5M2, #triton_gpu.dot_op<{opIdx = 1, parent = #blocked}>>,
%arg2: tensor<64x128xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #blocked1}>>) -> tensor<64x64xf32, #blocked> {
%cst_0 = arith.constant dense<0.000000e+00> : tensor<64x64xf32, #blocked>
// CHECK-80: tt.fp_to_fp {{.*}} : tensor<64x128xf8E4M3B11FNUZ, #triton_gpu.dot_op<{opIdx = 0, parent = #[[MMA]], kWidth = 4}>> -> tensor<64x128xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #[[MMA]], kWidth = 4}>>
// CHECK-80: tt.fp_to_fp {{.*}} : tensor<128x64xf8E5M2, #triton_gpu.dot_op<{opIdx = 1, parent = #[[MMA]], kWidth = 4}>> -> tensor<128x64xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #[[MMA]], kWidth = 4}>>
// CHECK-80: tt.dot {{.*}} -> tensor<64x64xf32, #[[MMA]]>
// CHECK-80: tt.fp_to_fp {{.*}} : tensor<64x128xf8E4M3B11FNUZ, #triton_gpu.dot_op<{opIdx = 0, parent = #[[$MMA]], kWidth = 4}>> -> tensor<64x128xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #[[$MMA]], kWidth = 4}>>
// CHECK-80: tt.fp_to_fp {{.*}} : tensor<128x64xf8E5M2, #triton_gpu.dot_op<{opIdx = 1, parent = #[[$MMA]], kWidth = 4}>> -> tensor<128x64xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #[[$MMA]], kWidth = 4}>>
// CHECK-80: tt.dot {{.*}} -> tensor<64x64xf32, #[[$MMA]]>
%d = tt.dot %arg0, %arg1, %cst_0 {allowTF32 = true, maxNumImpreciseAcc = 0 : i32} :
tensor<64x128xf8E4M3B11FNUZ, #triton_gpu.dot_op<{opIdx = 0, parent = #blocked}>> * tensor<128x64xf8E5M2, #triton_gpu.dot_op<{opIdx = 1, parent = #blocked}>> -> tensor<64x64xf32, #blocked>
tt.return %d : tensor<64x64xf32, #blocked>
Expand Down
4 changes: 2 additions & 2 deletions test/TritonGPU/loop-pipeline.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -877,12 +877,12 @@ module attributes {"triton_gpu.compute-capability" = 80 : i32, "triton_gpu.num-c
} // end module

// -----
// CHECK: #[[SHARED_LAYOUT:shared.*]] = #triton_gpu.shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [0], hasLeadingOffset = false}>
// CHECK: #[[$SHARED_LAYOUT:shared.*]] = #triton_gpu.shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [0], hasLeadingOffset = false}>
// CHECK-LABEL: tt.func @indirect_load_shared_layout
// CHECK: scf.for
// CHECK: %[[NEXT_BUFFER_1:.*]] = tt.addptr %{{.*}}, {{.*}}
// CHECK: triton_gpu.async_copy_global_to_local %[[NEXT_BUFFER_1]]
// CHECK: %[[IND_BUFFER_0:.*]] = triton_gpu.memdesc_subview {{.*}} : !tt.memdesc<1x16xi64, #[[SHARED_LAYOUT]], mutable> -> !tt.memdesc<16xi64, #[[SHARED_LAYOUT]], mutable>
// CHECK: %[[IND_BUFFER_0:.*]] = triton_gpu.memdesc_subview {{.*}} : !tt.memdesc<1x16xi64, #[[$SHARED_LAYOUT]], mutable> -> !tt.memdesc<16xi64, #[[$SHARED_LAYOUT]], mutable>
// CHECK: %[[IND_BUFFER_1:.*]] = triton_gpu.local_load %[[IND_BUFFER_0]]
// CHECK: %[[IND_BUFFER_2:.*]] = tt.expand_dims %[[IND_BUFFER_1]] {axis = 1 : i32}
// CHECK: %[[IND_BUFFER_3:.*]] = tt.broadcast %[[IND_BUFFER_2]]
Expand Down
6 changes: 6 additions & 0 deletions test/lit.cfg.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,12 @@
config.test_exec_root = os.path.join(config.triton_obj_root, 'test')
config.triton_tools_dir = os.path.join(config.triton_obj_root, 'bin')
config.filecheck_dir = os.path.join(config.triton_obj_root, 'bin', 'FileCheck')

# FileCheck -enable-var-scope is enabled by default in MLIR test
# This option avoids to accidentally reuse variable across -LABEL match,
# it can be explicitly opted-in by prefixing the variable name with $
config.environment["FILECHECK_OPTS"] = "--enable-var-scope"

tool_dirs = [config.triton_tools_dir, config.llvm_tools_dir, config.filecheck_dir]

# Tweak the PATH to include the tools dir.
Expand Down
Loading