Skip to content

[XPU][TritonGPUToLLVM] Use llvm.func attributes to express kernels ND-ranges #2770

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 8 commits into from
Nov 26, 2024
Merged

[XPU][TritonGPUToLLVM] Use llvm.func attributes to express kernels ND-ranges #2770

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
f800e9f
[XPU][TritonGPUToLLVM] Use `llvm.func` attributes to express kernels …
victor-eds Nov 20, 2024
27f8636
Merge branch 'main' into use-llvm-func-attrs
victor-eds Nov 20, 2024
824fbc1
Drop old tests
victor-eds Nov 20, 2024
7c192f6
Do not assert
victor-eds Nov 20, 2024
a8dcad0
Merge branch 'main' into use-llvm-func-attrs
victor-eds Nov 21, 2024
f0a2fcc
Add back attribute
victor-eds Nov 22, 2024
7907234
Drop comment
victor-eds Nov 22, 2024
bf8fc32
Fix attribute type
victor-eds Nov 22, 2024
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
4 changes: 2 additions & 2 deletions test/Conversion/intel/dpas_to_block_layout_convert.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@
module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 32 : i32, triton_gpu.shared = 67584 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @convert_dpas(
// CHECK-SAME: %[[VAL_0:.*]]: !llvm.ptr<1>,
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {noinline = false, triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [512 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {intel_reqd_sub_group_size = 16 : i32, noinline = false, reqd_work_group_size = array<i32: 512, 1, 1>} {
tt.func public @convert_dpas(%arg0: !tt.ptr<f16> {tt.divisibility = 16 : i32}) attributes {noinline = false} {
%cst = arith.constant dense<0.000000e+00> : tensor<128x256xf16, #mma>

Expand Down Expand Up @@ -69,7 +69,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 32
module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 32 : i32, triton_gpu.shared = 67584 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @convert_dpas(
// CHECK-SAME: %[[VAL_0:.*]]: !llvm.ptr<1>,
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {noinline = false, triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [512 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {intel_reqd_sub_group_size = 16 : i32, noinline = false, reqd_work_group_size = array<i32: 512, 1, 1>} {
tt.func public @convert_dpas(%arg0: !tt.ptr<f16> {tt.divisibility = 16 : i32}) attributes {noinline = false} {
%cst = arith.constant dense<0.000000e+00> : tensor<128x256xf16, #mma>

Expand Down
6 changes: 3 additions & 3 deletions test/Conversion/intel/shared_to_dot_layout_convert.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@
module attributes {"triton_gpu.num-warps" = 32 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @convert_dot(
// CHECK-SAME: %[[VAL_0:.*]]: !llvm.struct<(f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16)>,
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], {{.*}}} {
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {intel_reqd_sub_group_size = 16 : i32, {{.*}}} {
tt.func @convert_dot(%A: tensor<128x64xf16, #blocked0>) {
// CHECK-DAG: %[[CST_128:.*]] = llvm.mlir.constant(128 : i32) : i32
// CHECK-DAG: %[[CST_4:.*]] = llvm.mlir.constant(4 : i32) : i32
Expand Down Expand Up @@ -91,7 +91,7 @@ module attributes {"triton_gpu.num-warps" = 32 : i32, "triton_gpu.threads-per-wa
module attributes {"triton_gpu.num-warps" = 32 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @convert_dot(
// CHECK-SAME: %[[VAL_0:.*]]: !llvm.struct<(f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16)>,
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], {{.*}}} {
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {intel_reqd_sub_group_size = 16 : i32, {{.*}}} {
tt.func @convert_dot(%A: tensor<128x64xf16, #blocked0>) {
// CHECK-DAG: %[[CST_128:.*]] = llvm.mlir.constant(128 : i32) : i32
// CHECK-DAG: %[[CST_32:.*]] = llvm.mlir.constant(32 : i32) : i32
Expand Down Expand Up @@ -174,7 +174,7 @@ module attributes {"triton_gpu.num-warps" = 32 : i32, "triton_gpu.threads-per-wa
module attributes {"triton_gpu.num-warps" = 32 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @convert_dot(
// CHECK-SAME: %[[VAL_1:.*]]: !llvm.struct<(f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16)>,
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], {{.*}}} {
// CHECK-SAME: %[[SCRATCH_SLM:.*]]: !llvm.ptr<3>) attributes {intel_reqd_sub_group_size = 16 : i32, {{.*}}} {
tt.func @convert_dot(%B: tensor<64x256xf16, #blocked1>) {
// CHECK-DAG: %[[CST_128:.*]] = llvm.mlir.constant(128 : i32) : i32
// CHECK-DAG: %[[CST_256:.*]] = llvm.mlir.constant(256 : i32) : i32
Expand Down
2 changes: 1 addition & 1 deletion test/Conversion/intel/tritongpu_to_gen.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 : i32} {
// CHECK: llvm.func spir_kernelcc @test_empty_kernel(%arg0: i64, %arg1: !llvm.ptr<1>)
// Here the 128 comes from the 4 in module attribute multiples 32
// CHECK-SAME: attributes {triton_gen.intel_reqd_sub_group_size = [32 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: attributes {intel_reqd_sub_group_size = 32 : i32, reqd_work_group_size = array<i32: 128, 1, 1>} {
tt.func @test_empty_kernel(%lb : index, %A : !tt.ptr<f16>) {
// CHECK: llvm.return
tt.return
Expand Down
2 changes: 1 addition & 1 deletion test/Conversion/intel/tritongpu_to_gen_dot.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,7 +106,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 1 :
// CHECK: llvm.func spir_funccc @_Z38intel_sub_group_f16_f16_matrix_mad_k16Dv8_sDv8_iDv8_f(vector<8xi16>, vector<8xi32>, vector<8xf32>) -> vector<8xf32> attributes {convergent, memory_effects = #llvm.memory_effects<other = none, argMem = none, inaccessibleMem = none>, no_unwind, will_return}
// CHECK-LABEL: llvm.func spir_kernelcc @dot_rep_cluster_4_2(
// CHECK-SAME: %[[A:.*]]: !llvm.struct<(f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16)>, %[[B:.*]]: !llvm.struct<(f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16, f16)>,
// CHECK-SAME: %[[C:.*]]: !llvm.struct<(f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32)>) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [16 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[C:.*]]: !llvm.struct<(f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32, f32)>) attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 16, 1, 1>} {
tt.func @dot_rep_cluster_4_2(%a: tensor<32x32xf16, #dot_operand_a>, %b: tensor<32x32xf16, #dot_operand_b>, %c: tensor<32x32xf32, #dpas>) {
// CHECK: %[[VAL_3:.*]] = llvm.mlir.undef : vector<8xf32>
// CHECK: %[[CST_15:.*]] = llvm.mlir.constant(15 : i32) : i32
Expand Down
12 changes: 6 additions & 6 deletions test/Conversion/intel/tritongpu_to_llvm_intel_advanced_path.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,7 +114,7 @@ module attributes {"triton_intel_gpu.support_sg_2d_block", "triton_intel_gpu.sup

module attributes {"triton_intel_gpu.support_sg_2d_block", "triton_intel_gpu.support_dpas", "triton_gpu.num-warps" = 32 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @matmul_kernel_with_block_pointers_tf32(
// CHECK-SAME: [[VAL_0:%.*]]: !llvm.ptr<1>) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [512 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: [[VAL_0:%.*]]: !llvm.ptr<1>) attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 512, 1, 1>} {
tt.func public @matmul_kernel_with_block_pointers_tf32(%arg0: !tt.ptr<f32>) {
%c0_i64 = arith.constant 0 : i64
%c0_i32 = arith.constant 0 : i32
Expand All @@ -134,7 +134,7 @@ module attributes {"triton_intel_gpu.support_sg_2d_block", "triton_intel_gpu.sup

module attributes {"triton_intel_gpu.support_sg_2d_block", "triton_intel_gpu.support_dpas", "triton_gpu.num-warps" = 32 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @matmul_kernel_with_block_pointers_f16accu(
// CHECK-SAME: [[VAL_0:%.*]]: !llvm.ptr<1>) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [512 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: [[VAL_0:%.*]]: !llvm.ptr<1>) attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 512, 1, 1>} {
tt.func public @matmul_kernel_with_block_pointers_f16accu(%arg0: !tt.ptr<f16>) {
%c0_i64 = arith.constant 0 : i64
%c0_i32 = arith.constant 0 : i32
Expand All @@ -157,7 +157,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 :
// CHECK-DAG: llvm.func spir_funccc @_Z32sub_group_non_uniform_reduce_addf(f32) -> f32

// CHECK-LABEL: llvm.func spir_kernelcc @reduce_sum(
// CHECK-SAME: [[VAL_0:%.*]]: vector<8xf32>) -> f32 attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]}
// CHECK-SAME: [[VAL_0:%.*]]: vector<8xf32>) -> f32 attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 128, 1, 1>}
tt.func public @reduce_sum(%arg0: tensor<8x16xf32>) -> f32 {
// CHECK: [[VAL_1:%.*]] = llvm.mlir.constant(0 : i32) : i32
// CHECK: [[VAL_2:%.*]] = llvm.extractelement [[VAL_0]][[[VAL_1]] : i32] : vector<8xf32>
Expand All @@ -172,7 +172,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 :
}

// CHECK-LABEL: llvm.func spir_kernelcc @reduce_max(
// CHECK-SAME: [[VAL_0:%.*]]: vector<8xf32>) -> f32 attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]}
// CHECK-SAME: [[VAL_0:%.*]]: vector<8xf32>) -> f32 attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 128, 1, 1>}
tt.func public @reduce_max(%arg0: tensor<8x16xf32>) -> f32 {
// CHECK: [[VAL_1:%.*]] = llvm.mlir.constant(0 : i32) : i32
// CHECK: [[VAL_2:%.*]] = llvm.extractelement [[VAL_0]][[[VAL_1]] : i32] : vector<8xf32>
Expand Down Expand Up @@ -229,7 +229,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 :
}

// CHECK-LABEL: llvm.func spir_kernelcc @addptr(
// CHECK-SAME: [[VAL_0:%.*]]: !llvm.ptr<1>) -> !llvm.ptr<1> attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]}
// CHECK-SAME: [[VAL_0:%.*]]: !llvm.ptr<1>) -> !llvm.ptr<1> attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 128, 1, 1>}
tt.func public @addptr(%arg0: !tt.ptr<f16>) -> !tt.ptr<f16> {
// CHECK: [[VAL_1:%.*]] = llvm.mlir.constant(0 : i32) : i32
// CHECK: [[VAL_2:%.*]] = llvm.call spir_funccc @_Z12get_group_idj([[VAL_1]]) {{.*}} : (i32) -> i64
Expand Down Expand Up @@ -368,7 +368,7 @@ module attributes {"triton_gpu.num-warps" = 4 : i32, "triton_gpu.threads-per-war
#warp = #triton_intel_gpu.warp<{sizePerThread = [16, 64], threadsPerWarp = [1, 1], order = [1, 0]}>

// CHECK-LABEL: llvm.func spir_kernelcc @test(
// CHECK-SAME: %[[VAL_0:.*]]: f32) -> vector<16xf32> attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [64 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[VAL_0:.*]]: f32) -> vector<16xf32> attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 64, 1, 1>} {
// CHECK: %[[VAL_2:.*]] = llvm.mlir.poison : vector<16xf32>
// CHECK: %[[VAL_3:.*]] = llvm.mlir.constant(0 : i32) : i32
// CHECK: %[[VAL_4:.*]] = llvm.call spir_funccc @_Z17sub_group_shufflefj(%[[VAL_0]], %[[VAL_3]])
Expand Down
4 changes: 2 additions & 2 deletions test/Conversion/intel/tritongpu_transposed_reduction.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 :
// CHECK: }

// CHECK: llvm.func spir_kernelcc @reduce_sum(
// CHECK-SAME: %[[VAL_0:.*]]: vector<16xf32>) -> f32 attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[VAL_0:.*]]: vector<16xf32>) -> f32 attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 128, 1, 1>} {
// CHECK: %[[VAL_2:.*]] = llvm.mlir.constant(0 : i32) : i32
// CHECK: %[[VAL_3:.*]] = llvm.extractelement %[[VAL_0]]{{\[}}%[[VAL_2]] : i32] : vector<16xf32>
// CHECK: %[[VAL_4:.*]] = llvm.mlir.constant(1 : i32) : i32
Expand Down Expand Up @@ -78,7 +78,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 :
}

// CHECK: llvm.func spir_kernelcc @reduce_max(
// CHECK-SAME: %[[VAL_0:.*]]: vector<16xf32>) -> f32 attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[VAL_0:.*]]: vector<16xf32>) -> f32 attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 128, 1, 1>} {
// CHECK: %[[VAL_2:.*]] = llvm.mlir.constant(0 : i32) : i32
// CHECK: %[[VAL_3:.*]] = llvm.extractelement %[[VAL_0]]{{\[}}%[[VAL_2]] : i32] : vector<16xf32>
// CHECK: %[[VAL_4:.*]] = llvm.mlir.constant(1 : i32) : i32
Expand Down
4 changes: 2 additions & 2 deletions test/TritonIntelGPU/blockptr_load.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,7 +57,7 @@ module attributes {"triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-war
module attributes {"triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @dot_op_a_2d_load(
// CHECK-SAME: %[[VAL_0:.*]]: !llvm.ptr<1>,
// CHECK-SAME: %[[VAL_1:.*]]: i64, %[[VAL_2:.*]]: i64, %[[VAL_3:.*]]: i64, %[[VAL_4:.*]]: i64) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[VAL_1:.*]]: i64, %[[VAL_2:.*]]: i64, %[[VAL_3:.*]]: i64, %[[VAL_4:.*]]: i64) attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 128, 1, 1>} {
tt.func public @dot_op_a_2d_load(%arg0: !tt.ptr<f16>, %arg2: i64, %arg4: i64, %arg5: i64, %arg7: i64) {
%c0_i32 = arith.constant 0 : i32
%c1_i64 = arith.constant 1 : i64
Expand Down Expand Up @@ -129,7 +129,7 @@ module attributes {"triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-war
module attributes {"triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @dot_op_b_2d_load(
// CHECK-SAME: %[[VAL_0:.*]]: !llvm.ptr<1>,
// CHECK-SAME: %[[VAL_1:.*]]: i64, %[[VAL_2:.*]]: i64, %[[VAL_3:.*]]: i64) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[VAL_1:.*]]: i64, %[[VAL_2:.*]]: i64, %[[VAL_3:.*]]: i64) attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 128, 1, 1>} {
tt.func public @dot_op_b_2d_load(%arg1: !tt.ptr<f16>, %arg3: i64, %arg4: i64, %arg7: i64) {
%c0_i32 = arith.constant 0 : i32
%c1_i64 = arith.constant 1 : i64
Expand Down
2 changes: 1 addition & 1 deletion test/TritonIntelGPU/blockptr_store.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,7 +61,7 @@ module attributes {"triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-war
module attributes {"triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
// CHECK-LABEL: llvm.func spir_kernelcc @dpas_layout_2d_store_rep_cluster_4_2(
// CHECK-SAME: %[[base:.*]]: !llvm.ptr<1>,
// CHECK-SAME: %[[width:.*]]: i64, %[[height:.*]]: i64, %[[rowStride:.*]]: i64) attributes {triton_gen.intel_reqd_sub_group_size = [16 : i32], triton_gen.max_work_group_size = [128 : i32, 1 : i32, 1 : i32]} {
// CHECK-SAME: %[[width:.*]]: i64, %[[height:.*]]: i64, %[[rowStride:.*]]: i64) attributes {intel_reqd_sub_group_size = 16 : i32, reqd_work_group_size = array<i32: 128, 1, 1>} {
tt.func public @dpas_layout_2d_store_rep_cluster_4_2(%base: !tt.ptr<f16>, %width: i64, %height: i64, %rowStride: i64) {
%cst = arith.constant dense<0.000000e+00> : tensor<32x32xf16, #dpas>
%c0_i32 = arith.constant 0 : i32
Expand Down
Loading
Loading