leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 26 Packages Projects Releases Wiki Activity

releaase 2.11 (#703)

Browse Source
This commit is contained in:
Aditya Atluri
2022-11-19 09:02:15 -05:00
committed by GitHub
parent 3c90f6aea6
commit c975e2ccbb
329 changed files with 47332 additions and 10607 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/unit/gemm/device/CMakeLists.txt
+58 -3
View File
@@ -68,7 +68,12 @@ cutlass_test_unit_add_executable(
simt_sgemm_nt_sm80.cu
simt_sgemm_tn_sm80.cu
simt_cgemm_nt_sm80.cu
simt_cgemm_tn_sm80.cu
simt_f8gemm_tn_sm50.cu
simt_cgemm_nn_sm50.cu
simt_cgemm_nt_sm50.cu
simt_cgemm_tn_sm50.cu
@@ -239,6 +244,13 @@ cutlass_test_unit_add_executable(
gemm_cf64n_cf64t_cf64t_tensor_op_f64_gaussian_sm80.cu
gemm_cf64t_cf64n_cf64t_tensor_op_f64_gaussian_sm80.cu
# SM90 device level tests
gemm_f64n_f64t_f64t_tensor_op_f64_sm90.cu
gemm_f64t_f64n_f64t_tensor_op_f64_sm90.cu
gemm_cf64n_cf64t_cf64t_tensor_op_f64_sm90.cu
gemm_cf64t_cf64n_cf64t_tensor_op_f64_sm90.cu
gemm_cf64n_cf64t_cf64t_tensor_op_f64_gaussian_sm90.cu
gemm_cf64t_cf64n_cf64t_tensor_op_f64_gaussian_sm90.cu
)
cutlass_test_unit_add_executable(
@@ -430,7 +442,7 @@ cutlass_test_unit_add_executable(
BATCH_SOURCES ON
BATCH_SIZE 4
## SYRK
## SYRK
# Syrk SM80 f64 tests
syrk_f64n_f64t_tensor_op_f64_sm80.cu
syrk_f64t_f64n_tensor_op_f64_sm80.cu
@@ -452,6 +464,12 @@ cutlass_test_unit_add_executable(
syrk_cf32n_cf32t_tensor_op_fast_f32_sm80.cu
syrk_cf32n_cf32n_tensor_op_fast_f32_sm80.cu
# Syrk SM90 f64 tests
syrk_f64_f64_tensor_op_f64_sm90.cu
# Syrk SM90 complex f64 tests
syrk_cf64_cf64_tensor_op_f64_sm90.cu
## HERK
# Herk SM80 complex f64 tests
herk_cf64h_cf64n_tensor_op_f64_sm80.cu
@@ -460,6 +478,9 @@ cutlass_test_unit_add_executable(
herk_cf32h_cf32n_tensor_op_f32_sm80.cu
herk_cf32h_cf32n_tensor_op_fast_f32_sm80.cu
# Herk SM90 complex f64 tests
herk_cf64_cf64_tensor_op_f64_sm90.cu
## TRMM
# Trmm SM80 f64 tests
trmm_f64n_f64n_f64t_tensor_op_f64_ls_sm80.cu
@@ -486,6 +507,12 @@ cutlass_test_unit_add_executable(
trmm_cf32n_cf32n_cf32t_tensor_op_f32_sm80.cu
trmm_cf32n_cf32n_cf32t_tensor_op_fast_f32_sm80.cu
# Trmm SM90 f64 tests
trmm_f64_f64_f64_tensor_op_f64_sm90.cu
# Trmm SM90 complex f64 tests
trmm_cf64_cf64_cf64_tensor_op_f64_sm90.cu
## SYR2K
# Syr2k SM80 f64 tests
syr2k_f64n_f64t_tensor_op_f64_sm80.cu
@@ -508,6 +535,12 @@ cutlass_test_unit_add_executable(
syr2k_cf32n_cf32n_tensor_op_fast_f32_sm80.cu
syr2k_cf32n_cf32t_tensor_op_fast_f32_sm80.cu
# Syr2k SM90 f64 tests
syr2k_f64_f64_tensor_op_f64_sm90.cu
# Syr2k SM90 complex f64 tests
syr2k_cf64_cf64_tensor_op_f64_sm90.cu
## HER2K
# Her2k SM80 complex f64 tests
her2k_cf64n_cf64n_tensor_op_f64_sm80.cu
@@ -516,6 +549,9 @@ cutlass_test_unit_add_executable(
her2k_cf32h_cf32n_tensor_op_f32_sm80.cu
her2k_cf32h_cf32n_tensor_op_fast_f32_sm80.cu
# Her2k SM90 complex f64 tests
her2k_cf64_cf64_tensor_op_f64_sm90.cu
## SYMM
# Symm SM80 f64 tests
symm_f64n_f64n_tensor_op_f64_ls_sm80.cu
@@ -546,6 +582,12 @@ cutlass_test_unit_add_executable(
symm_cf32n_cf32n_tensor_op_fast_f32_ls_sm80.cu
symm_cf32n_cf32n_tensor_op_fast_f32_rs_sm80.cu
# Symm SM90 f64 tests
symm_f64_f64_tensor_op_f64_sm90.cu
# Symm SM90 complex f64 tests
symm_cf64_cf64_cf64_tensor_op_f64_sm90.cu
# Hemm SM80 complex f64 tests
hemm_cf64h_cf64n_cf64n_tensor_op_ls_f64_sm80.cu
hemm_cf64h_cf64n_cf64n_tensor_op_rs_f64_sm80.cu
@@ -556,7 +598,10 @@ cutlass_test_unit_add_executable(
hemm_cf32h_cf32n_tensor_op_f32_rs_sm80.cu
hemm_cf32h_cf32n_tensor_op_fast_f32_ls_sm80.cu
hemm_cf32h_cf32n_tensor_op_fast_f32_rs_sm80.cu
)
# Hemm SM90 complex f64 tests
hemm_cf64_cf64_cf64_tensor_op_f64_sm90.cu
)
cutlass_test_unit_add_executable(
cutlass_test_unit_gemm_device_grouped_blas3
@@ -582,3 +627,13 @@ cutlass_test_unit_add_executable(
)
endif()
if (NOT CUDA_COMPILER MATCHES "[Cc]lang")
cutlass_test_unit_add_executable(
cutlass_test_unit_gemm_device_broadcast
gemm_f16t_f16n_f16t_tensor_op_f16_broadcast_sm80.cu
)
endif()