Move some of the E2E tests for GEG & CEG into nightly runs

mlir/test/e2e/CMakeLists.txt

@@ -33,18 +33,10 @@ if (ROCMLIR_DRIVER_PR_E2E_TEST_ENABLED)
     PrAttentionBF16
     PrAttentionI8
     PrGemmSplitK
-    PrGemmElementwiseGemmF32
-    PrGemmElementwiseGemmF16
-    PrGemmElementwiseGemmBF16
-    PrConvElementwiseGemmF32
     PrConvElementwiseGemmF16
-    PrConvElementwiseGemmBF16
-    PrGemmElementwiseGemmF32SplitK
-    PrGemmElementwiseGemmF16SplitK
-    PrGemmElementwiseGemmBF16SplitK
-    PrConvElementwiseGemmF32SplitK
     PrConvElementwiseGemmF16SplitK
-    PrConvElementwiseGemmBF16SplitK
+    PrGemmElementwiseGemmF16
+    PrGemmElementwiseGemmF16SplitK
     PrGemmDirectToLDS
   )
   set(GEN_MODE "")
@@ -76,6 +68,19 @@ if (ROCK_E2E_TEST_ENABLED)
   gemm_split_k_bf16
   gemm_split_k_f32
   )
+  list(APPEND CONFIGS 
+  GemmElementwiseGemmF32
+  GemmElementwiseGemmBF16
+  ConvElementwiseGemmF32
+  ConvElementwiseGemmF16
+  ConvElementwiseGemmBF16
+  GemmElementwiseGemmF32SplitK
+  GemmElementwiseGemmF16SplitK
+  GemmElementwiseGemmBF16SplitK
+  ConvElementwiseGemmF32SplitK
+  ConvElementwiseGemmF16SplitK
+  ConvElementwiseGemmBF16SplitK
+  ) 
   list(APPEND CONFIGS
   GemmDirectToLDS
   )

mlir/test/e2e/PrConvElementwiseGemmBF16.toml → mlir/test/e2e/ConvElementwiseGemmBF16.toml

@@ -1,4 +1,4 @@
-directory = "PrConvElementwiseGemmBF16"
+directory = "ConvElementwiseGemmBF16"
 prefix = "rocmlir-gen"
 suffix = "--operation conv_gemm -t bf16 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext,%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
 
@@ -18,16 +18,16 @@ values = ["-fil_layout=gkcyx -in_layout=ngchw", "-fil_layout=gkyxc -in_layout=nh
 
 ## conv+gemm variant
 [[suite]]
-name = "pr_conv_gemm_bf16"
+name = "conv_gemm_bf16"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
 
 [[suite.test]]
 config = "-groupsize=1 -batchsize=2 -in_channels=256 -out_channels=128 -in_h=32 -in_w=32 -fil_h=1 -fil_w=1 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=0 -padding_h_r=0 -padding_w_l=0 -padding_w_r=0 -gemmO=128"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"

mlir/test/e2e/PrConvElementwiseGemmBF16SplitK.toml → mlir/test/e2e/ConvElementwiseGemmBF16SplitK.toml

@@ -1,4 +1,4 @@
-directory = "PrConvElementwiseGemmBF16SplitK"
+directory = "ConvElementwiseGemmBF16SplitK"
 prefix = "rocmlir-gen"
 suffix = "--operation conv_gemm -t bf16 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext,%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
 
@@ -25,10 +25,10 @@ prefix = "-perf_config="
 
 ## conv+gemm split-k variant
 [[suite]]
-name = "pr_conv_gemm_bf16_splitk"
+name = "conv_gemm_bf16_splitk"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"

mlir/test/e2e/ConvElementwiseGemmF16.toml

@@ -0,0 +1,33 @@
+directory = "ConvElementwiseGemmF16"
+prefix = "rocmlir-gen"
+suffix = "--operation conv_gemm -t f16 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext,%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
+
+[[axis]]
+name = "transC"
+values = ["true", "false"]
+prefix = "--transC="
+
+[[axis]]
+name = "transO"
+values = ["true", "false"]
+prefix = "--transO="
+
+[[axis]]
+name = "layout"
+values = ["-fil_layout=gkcyx -in_layout=ngchw", "-fil_layout=gkyxc -in_layout=nhwgc"]
+
+## conv+gemm variant
+[[suite]]
+name = "conv_gemm_f16"
+
+[[suite.test]]
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+
+[[suite.test]]
+config = "-groupsize=1 -batchsize=2 -in_channels=256 -out_channels=128 -in_h=32 -in_w=32 -fil_h=1 -fil_w=1 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=0 -padding_h_r=0 -padding_w_l=0 -padding_w_r=0 -gemmO=128"
+
+[[suite.test]]
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+
+[[suite.test]]
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"

mlir/test/e2e/ConvElementwiseGemmF16SplitK.cfg

@@ -0,0 +1,2 @@
+if (not config.arch_support_mfma and not config.arch_support_wmma) or (not 'atomic_add_f16' in config.features):
+  config.unsupported = True

mlir/test/e2e/ConvElementwiseGemmF16SplitK.toml

@@ -0,0 +1,31 @@
+directory = "ConvElementwiseGemmF16SplitK"
+prefix = "rocmlir-gen"
+suffix = "--operation conv_gemm -t f16 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext,%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
+
+[[axis]]
+name = "transC"
+values = ["true", "false"]
+prefix = "--transC="
+
+[[axis]]
+name = "transO"
+values = ["true", "false"]
+prefix = "--transO="
+
+[[axis]]
+name = "layout"
+values = ["-fil_layout=gkcyx -in_layout=ngchw", "-fil_layout=gkyxc -in_layout=nhwgc"]
+
+[[axis]]
+name = "splitKFactor"
+values = ["attn:v2:64,128,32,16,32,16,4,3,1,2,1",
+          "attn:v2:64,128,32,16,32,16,4,4,1,2,1",
+          "attn:v2:32,128,32,32,32,16,8,4,1,2,1"]
+prefix = "-perf_config="
+
+## conv+gemm split-k variant
+[[suite]]
+name = "conv_gemm_f16_splitk"
+
+[[suite.test]]
+config = "-groupsize=1 -batchsize=1 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"

mlir/test/e2e/PrConvElementwiseGemmF32.toml → mlir/test/e2e/ConvElementwiseGemmF32.toml

@@ -1,4 +1,4 @@
-directory = "PrConvElementwiseGemmF32"
+directory = "ConvElementwiseGemmF32"
 prefix = "rocmlir-gen"
 suffix = "--operation conv_gemm -t f32 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext,%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
 
@@ -18,16 +18,16 @@ values = ["-fil_layout=gkcyx -in_layout=ngchw", "-fil_layout=gkyxc -in_layout=nh
 
 ## conv+gemm variant
 [[suite]]
-name = "pr_conv_gemm_f32"
+name = "conv_gemm_f32"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
 
 [[suite.test]]
 config = "-groupsize=1 -batchsize=2 -in_channels=256 -out_channels=128 -in_h=32 -in_w=32 -fil_h=1 -fil_w=1 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=0 -padding_h_r=0 -padding_w_l=0 -padding_w_r=0 -gemmO=128"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"

mlir/test/e2e/PrConvElementwiseGemmF32SplitK.toml → mlir/test/e2e/ConvElementwiseGemmF32SplitK.toml

@@ -1,4 +1,4 @@
-directory = "PrConvElementwiseGemmF32SplitK"
+directory = "ConvElementwiseGemmF32SplitK"
 prefix = "rocmlir-gen"
 suffix = "--operation conv_gemm -t f32 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext,%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
 
@@ -28,7 +28,7 @@ prefix = "-perf_config="
 name = "pr_conv_gemm_f32_splitk"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"

mlir/test/e2e/GemmElementwiseGemmBF16.cfg

@@ -0,0 +1,2 @@
+if (not config.arch_support_mfma) and (not config.arch_support_wmma):
+  config.unsupported = True

mlir/test/e2e/PrGemmElementwiseGemmBF16.toml → mlir/test/e2e/GemmElementwiseGemmBF16.toml

@@ -1,4 +1,4 @@
-directory = "PrGemmElementwiseGemmBF16"
+directory = "GemmElementwiseGemmBF16"
 prefix = "rocmlir-gen"
 suffix = "--operation gemm_gemm -t bf16 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
 
@@ -9,7 +9,7 @@ prefix = "--transB="
 
 ## gemm+gemm variant
 [[suite]]
-name = "pr_gemm_gemm_bf16"
+name = "gemm_gemm_bf16"
 
 [[suite.test]]
 config = "-m 384 -n 384 -k 64 -gemmO 64"

mlir/test/e2e/PrGemmElementwiseGemmBF16SplitK.toml → mlir/test/e2e/GemmElementwiseGemmBF16SplitK.toml

@@ -1,4 +1,4 @@
-directory = "PrGemmElementwiseGemmBF16SplitK"
+directory = "GemmElementwiseGemmBF16SplitK"
 prefix = "rocmlir-gen"
 suffix = "--operation gemm_gemm -t bf16 --arch %arch -RMS_threshold 0.01 -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
 
@@ -30,7 +30,7 @@ prefix = "-perf_config="
 
 ## gemm+gemm split-k variant
 [[suite]]
-name = "pr_gemm_gemm_bf16_splitk"
+name = "gemm_gemm_bf16_splitk"
 
 [[suite.test]]
 config = "-m 384 -n 384 -k 128 -gemmO 128"

mlir/test/e2e/GemmElementwiseGemmF16SplitK.cfg

@@ -0,0 +1,2 @@
+if (not config.arch_support_mfma and not config.arch_support_wmma) or (not 'atomic_add_f16' in config.features):
+  config.unsupported = True

mlir/test/e2e/GemmElementwiseGemmF16SplitK.toml

@@ -0,0 +1,37 @@
+directory = "GemmElementwiseGemmF16SplitK"
+prefix = "rocmlir-gen"
+suffix = "--operation gemm_gemm -t f16 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
+
+[[axis]]
+name = "transA"
+values = ["true", "false"]
+prefix = "--transA="
+
+[[axis]]
+name = "transB"
+values = ["true", "false"]
+prefix = "--transB="
+
+[[axis]]
+name = "transC"
+values = ["true", "false"]
+prefix = "--transC="
+
+[[axis]]
+name = "transO"
+values = ["true", "false"]
+prefix = "--transO="
+
+[[axis]]
+name = "splitKFactor"
+values = ["attn:v2:64,128,32,16,32,16,4,3,1,2,1",
+          "attn:v2:64,128,32,16,32,16,4,8,1,2,1",
+          "attn:v2:32,128,32,32,32,16,8,4,1,2,1"]
+prefix = "-perf_config="
+
+## gemm+gemm split-k variant
+[[suite]]
+name = "gemm_gemm_f16_splitk"
+
+[[suite.test]]
+config = "-m 384 -n 384 -k 128 -gemmO 128"

mlir/test/e2e/PrGemmElementwiseGemmF32.toml → mlir/test/e2e/GemmElementwiseGemmF32.toml

@@ -1,4 +1,4 @@
-directory = "PrGemmElementwiseGemmF32"
+directory = "GemmElementwiseGemmF32"
 prefix = "rocmlir-gen"
 suffix = "--operation gemm_gemm -t f32 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
 
@@ -24,7 +24,7 @@ prefix = "--transO="
 
 ## gemm+gemm variant
 [[suite]]
-name = "pr_gemm_gemm_f32"
+name = "gemm_gemm_f32"
 
 [[suite.test]]
 config = "-m 384 -n 384 -k 64 -gemmO 64"

mlir/test/e2e/PrGemmElementwiseGemmF32SplitK.toml → mlir/test/e2e/GemmElementwiseGemmF32SplitK.toml

@@ -1,4 +1,4 @@
-directory = "PrGemmElementwiseGemmF32SplitK"
+directory = "GemmElementwiseGemmF32SplitK"
 prefix = "rocmlir-gen"
 suffix = "--operation gemm_gemm -t f32 --arch %arch -pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
 
@@ -31,7 +31,7 @@ prefix = "-perf_config="
 
 ## gemm+gemm split-k variant
 [[suite]]
-name = "pr_gemm_gemm_f32_splitk"
+name = "gemm_gemm_f32_splitk"
 
 [[suite.test]]
 config = "-m 384 -n 384 -k 128 -gemmO 128"

mlir/test/e2e/PrConvElementwiseGemmF16.toml

@@ -21,13 +21,4 @@ values = ["-fil_layout=gkcyx -in_layout=ngchw", "-fil_layout=gkyxc -in_layout=nh
 name = "pr_conv_gemm_f16"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
-
-[[suite.test]]
-config = "-groupsize=1 -batchsize=2 -in_channels=256 -out_channels=128 -in_h=32 -in_w=32 -fil_h=1 -fil_w=1 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=0 -padding_h_r=0 -padding_w_l=0 -padding_w_r=0 -gemmO=128"
-
-[[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
-
-[[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=2 -in_channels=32 -out_channels=32 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"

mlir/test/e2e/PrConvElementwiseGemmF16SplitK.toml

@@ -28,7 +28,4 @@ prefix = "-perf_config="
 name = "pr_conv_gemm_f16_splitk"
 
 [[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
-
-[[suite.test]]
-config = "-groupsize=1 -batchsize=64 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=2 -dilation_w=2 -conv_stride_h=1 -conv_stride_w=1 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"
+config = "-groupsize=1 -batchsize=1 -in_channels=128 -out_channels=128 -in_h=28 -in_w=28 -fil_h=3 -fil_w=3 -dilation_h=1 -dilation_w=1 -conv_stride_h=2 -conv_stride_w=2 -padding_h_l=1 -padding_h_r=1 -padding_w_l=1 -padding_w_r=1 -gemmO=64"

mlir/test/e2e/PrGemmElementwiseGemmF16SplitK.toml

@@ -33,8 +33,5 @@ prefix = "-perf_config="
 [[suite]]
 name = "pr_gemm_gemm_f16_splitk"
 
-[[suite.test]]
-config = "-m 384 -n 384 -k 128 -gemmO 128"
-
 [[suite.test]]
 config = "-m 64 -n 64 -k 128 -gemmO 128"