Skip to content

Commit

Permalink
restructure entire gemm benchmarking and remove all vmfbs from git repo
Browse files Browse the repository at this point in the history
  • Loading branch information
saienduri committed Oct 1, 2024
1 parent 7c14c18 commit e7231b3
Show file tree
Hide file tree
Showing 809 changed files with 1,354 additions and 1,259 deletions.
5 changes: 3 additions & 2 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,8 @@ wheelhouse
*.egg-info
*.whl

# Gemm Bench
# Bench Artifacts
gemm/vmfb/
attention/vmfb/
conv/vmfb/
results/

Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_12x384x64x64x384xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_16x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_16x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_16x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_16x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_1x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_1x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_1x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_1x4096x64x64x64xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_1x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_20x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_20x4096x64x64x64xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_2x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_2x1024x64x64x64xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_2x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_2x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_2x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_32x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_32x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_32x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_32x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_40x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_40x1024x64x64x64xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_48x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_48x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_48x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_48x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_4x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_4x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_4x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_4x4096x64x64x64xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_4x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_64x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_64x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_64x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_64x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_768x4096x64x64x64xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_8x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file removed attention/vmfb/attention_8x1024x64x64x64xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_8x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_8x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_8x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_96x1024x64x64x1024xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_96x2048x64x64x2048xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_96x4096x64x64x4096xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file removed attention/vmfb/attention_96x8192x64x64x8192xf16.vmfb
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_16_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x16xbf16>) -> tensor<10240x16xbf16> {
func.func @main(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x16xbf16>) -> tensor<10240x16xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<10240x16xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<10240x16xbf16>) -> tensor<10240x16xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xbf16>, tensor<8192x16xbf16>) outs(%1 : tensor<10240x16xbf16>) -> tensor<10240x16xbf16>
return %2 : tensor<10240x16xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_16_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x16xf16>) -> tensor<10240x16xf16> {
func.func @main(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x16xf16>) -> tensor<10240x16xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<10240x16xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<10240x16xf16>) -> tensor<10240x16xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xf16>, tensor<8192x16xf16>) outs(%1 : tensor<10240x16xf16>) -> tensor<10240x16xf16>
return %2 : tensor<10240x16xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_1_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x1xbf16>) -> tensor<10240x1xbf16> {
func.func @main(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x1xbf16>) -> tensor<10240x1xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<10240x1xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<10240x1xbf16>) -> tensor<10240x1xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xbf16>, tensor<8192x1xbf16>) outs(%1 : tensor<10240x1xbf16>) -> tensor<10240x1xbf16>
return %2 : tensor<10240x1xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_1_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x1xf16>) -> tensor<10240x1xf16> {
func.func @main(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x1xf16>) -> tensor<10240x1xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<10240x1xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<10240x1xf16>) -> tensor<10240x1xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xf16>, tensor<8192x1xf16>) outs(%1 : tensor<10240x1xf16>) -> tensor<10240x1xf16>
return %2 : tensor<10240x1xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_2_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x2xbf16>) -> tensor<10240x2xbf16> {
func.func @main(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x2xbf16>) -> tensor<10240x2xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<10240x2xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<10240x2xbf16>) -> tensor<10240x2xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xbf16>, tensor<8192x2xbf16>) outs(%1 : tensor<10240x2xbf16>) -> tensor<10240x2xbf16>
return %2 : tensor<10240x2xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_2_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x2xf16>) -> tensor<10240x2xf16> {
func.func @main(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x2xf16>) -> tensor<10240x2xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<10240x2xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<10240x2xf16>) -> tensor<10240x2xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xf16>, tensor<8192x2xf16>) outs(%1 : tensor<10240x2xf16>) -> tensor<10240x2xf16>
return %2 : tensor<10240x2xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_32_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x32xbf16>) -> tensor<10240x32xbf16> {
func.func @main(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x32xbf16>) -> tensor<10240x32xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<10240x32xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<10240x32xbf16>) -> tensor<10240x32xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xbf16>, tensor<8192x32xbf16>) outs(%1 : tensor<10240x32xbf16>) -> tensor<10240x32xbf16>
return %2 : tensor<10240x32xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_32_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x32xf16>) -> tensor<10240x32xf16> {
func.func @main(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x32xf16>) -> tensor<10240x32xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<10240x32xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<10240x32xf16>) -> tensor<10240x32xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xf16>, tensor<8192x32xf16>) outs(%1 : tensor<10240x32xf16>) -> tensor<10240x32xf16>
return %2 : tensor<10240x32xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_4_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x4xbf16>) -> tensor<10240x4xbf16> {
func.func @main(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x4xbf16>) -> tensor<10240x4xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<10240x4xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<10240x4xbf16>) -> tensor<10240x4xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xbf16>, tensor<8192x4xbf16>) outs(%1 : tensor<10240x4xbf16>) -> tensor<10240x4xbf16>
return %2 : tensor<10240x4xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_4_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x4xf16>) -> tensor<10240x4xf16> {
func.func @main(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x4xf16>) -> tensor<10240x4xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<10240x4xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<10240x4xf16>) -> tensor<10240x4xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xf16>, tensor<8192x4xf16>) outs(%1 : tensor<10240x4xf16>) -> tensor<10240x4xf16>
return %2 : tensor<10240x4xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_8_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x8xbf16>) -> tensor<10240x8xbf16> {
func.func @main(%arg0: tensor<8192x10240xbf16>, %arg1: tensor<8192x8xbf16>) -> tensor<10240x8xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<10240x8xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<10240x8xbf16>) -> tensor<10240x8xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xbf16>, tensor<8192x8xbf16>) outs(%1 : tensor<10240x8xbf16>) -> tensor<10240x8xbf16>
return %2 : tensor<10240x8xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_10240_8_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x8xf16>) -> tensor<10240x8xf16> {
func.func @main(%arg0: tensor<8192x10240xf16>, %arg1: tensor<8192x8xf16>) -> tensor<10240x8xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<10240x8xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<10240x8xf16>) -> tensor<10240x8xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x10240xf16>, tensor<8192x8xf16>) outs(%1 : tensor<10240x8xf16>) -> tensor<10240x8xf16>
return %2 : tensor<10240x8xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_16_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x16xbf16>) -> tensor<1280x16xbf16> {
func.func @main(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x16xbf16>) -> tensor<1280x16xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<1280x16xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<1280x16xbf16>) -> tensor<1280x16xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xbf16>, tensor<8192x16xbf16>) outs(%1 : tensor<1280x16xbf16>) -> tensor<1280x16xbf16>
return %2 : tensor<1280x16xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_16_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x16xf16>) -> tensor<1280x16xf16> {
func.func @main(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x16xf16>) -> tensor<1280x16xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<1280x16xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<1280x16xf16>) -> tensor<1280x16xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xf16>, tensor<8192x16xf16>) outs(%1 : tensor<1280x16xf16>) -> tensor<1280x16xf16>
return %2 : tensor<1280x16xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_1_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x1xbf16>) -> tensor<1280x1xbf16> {
func.func @main(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x1xbf16>) -> tensor<1280x1xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<1280x1xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<1280x1xbf16>) -> tensor<1280x1xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xbf16>, tensor<8192x1xbf16>) outs(%1 : tensor<1280x1xbf16>) -> tensor<1280x1xbf16>
return %2 : tensor<1280x1xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_1_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x1xf16>) -> tensor<1280x1xf16> {
func.func @main(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x1xf16>) -> tensor<1280x1xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<1280x1xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<1280x1xf16>) -> tensor<1280x1xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xf16>, tensor<8192x1xf16>) outs(%1 : tensor<1280x1xf16>) -> tensor<1280x1xf16>
return %2 : tensor<1280x1xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_2_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x2xbf16>) -> tensor<1280x2xbf16> {
func.func @main(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x2xbf16>) -> tensor<1280x2xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<1280x2xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<1280x2xbf16>) -> tensor<1280x2xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xbf16>, tensor<8192x2xbf16>) outs(%1 : tensor<1280x2xbf16>) -> tensor<1280x2xbf16>
return %2 : tensor<1280x2xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_2_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x2xf16>) -> tensor<1280x2xf16> {
func.func @main(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x2xf16>) -> tensor<1280x2xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<1280x2xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<1280x2xf16>) -> tensor<1280x2xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xf16>, tensor<8192x2xf16>) outs(%1 : tensor<1280x2xf16>) -> tensor<1280x2xf16>
return %2 : tensor<1280x2xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_32_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x32xbf16>) -> tensor<1280x32xbf16> {
func.func @main(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x32xbf16>) -> tensor<1280x32xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<1280x32xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<1280x32xbf16>) -> tensor<1280x32xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xbf16>, tensor<8192x32xbf16>) outs(%1 : tensor<1280x32xbf16>) -> tensor<1280x32xbf16>
return %2 : tensor<1280x32xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_32_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x32xf16>) -> tensor<1280x32xf16> {
func.func @main(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x32xf16>) -> tensor<1280x32xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<1280x32xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<1280x32xf16>) -> tensor<1280x32xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xf16>, tensor<8192x32xf16>) outs(%1 : tensor<1280x32xf16>) -> tensor<1280x32xf16>
return %2 : tensor<1280x32xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_4_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x4xbf16>) -> tensor<1280x4xbf16> {
func.func @main(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x4xbf16>) -> tensor<1280x4xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<1280x4xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<1280x4xbf16>) -> tensor<1280x4xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xbf16>, tensor<8192x4xbf16>) outs(%1 : tensor<1280x4xbf16>) -> tensor<1280x4xbf16>
return %2 : tensor<1280x4xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_4_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x4xf16>) -> tensor<1280x4xf16> {
func.func @main(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x4xf16>) -> tensor<1280x4xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<1280x4xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<1280x4xf16>) -> tensor<1280x4xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xf16>, tensor<8192x4xf16>) outs(%1 : tensor<1280x4xf16>) -> tensor<1280x4xf16>
return %2 : tensor<1280x4xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_8_8192_bf16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x8xbf16>) -> tensor<1280x8xbf16> {
func.func @main(%arg0: tensor<8192x1280xbf16>, %arg1: tensor<8192x8xbf16>) -> tensor<1280x8xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<1280x8xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<1280x8xbf16>) -> tensor<1280x8xbf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xbf16>, tensor<8192x8xbf16>) outs(%1 : tensor<1280x8xbf16>) -> tensor<1280x8xbf16>
return %2 : tensor<1280x8xbf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_1280_8_8192_f16_tA.mlir
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@

module {
func.func @main_0(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x8xf16>) -> tensor<1280x8xf16> {
func.func @main(%arg0: tensor<8192x1280xf16>, %arg1: tensor<8192x8xf16>) -> tensor<1280x8xf16> {
%cst = arith.constant 0.000000e+00 : f16
%0 = tensor.empty() : tensor<1280x8xf16>
%1 = linalg.fill ins(%cst : f16) outs(%0 : tensor<1280x8xf16>) -> tensor<1280x8xf16>
%2 = linalg.matmul_transpose_a ins(%arg0, %arg1 : tensor<8192x1280xf16>, tensor<8192x8xf16>) outs(%1 : tensor<1280x8xf16>) -> tensor<1280x8xf16>
return %2 : tensor<1280x8xf16>
}
}
}
4 changes: 2 additions & 2 deletions gemm/mlir/gemm_128_1280_2048_bf16.mlir
Original file line number Diff line number Diff line change
@@ -1,9 +1,9 @@
module {
func.func @main_0(%arg0: tensor<128x2048xbf16>, %arg1: tensor<2048x1280xbf16>) -> tensor<128x1280xbf16> {
func.func @main(%arg0: tensor<128x2048xbf16>, %arg1: tensor<2048x1280xbf16>) -> tensor<128x1280xbf16> {
%cst = arith.constant 0.000000e+00 : bf16
%0 = tensor.empty() : tensor<128x1280xbf16>
%1 = linalg.fill ins(%cst : bf16) outs(%0 : tensor<128x1280xbf16>) -> tensor<128x1280xbf16>
%2 = linalg.matmul ins(%arg0, %arg1 : tensor<128x2048xbf16>, tensor<2048x1280xbf16>) outs(%1 : tensor<128x1280xbf16>) -> tensor<128x1280xbf16>
return %2 : tensor<128x1280xbf16>
}
}
}
Loading

0 comments on commit e7231b3

Please sign in to comment.