ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-07 00:04:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add more debug info

Browse Source
This commit is contained in:
Rostyslav Geyyer
2025-03-19 20:23:05 +00:00
parent 02e3796ee1
commit 221862c912
1 changed files with 9 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
include/ck/utility/amd_xdlops.hpp
View File

@@ -520,6 +520,9 @@ struct intrin_mfma_f32_32x32x64f8f6f4<32, 32>
using arg_type = int32x8_t;
printf("!!!!!!! %d %d %d %d ", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
printf("??????? %d %d %d %d ", arg_b[0], arg_b[1], arg_b[2], arg_b[3]);
reg_c.template AsType<float16_t>()(Number<0>{}) =
__builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4(
arg_type{arg_a[0], arg_a[1], arg_a[2], arg_a[3], 0, 0, 0, 0},
@@ -588,6 +591,9 @@ struct intrin_mfma_scale_f32_32x32x64f8f6f4<32, 32>
using arg_type = int32x8_t;
printf("!!!!!!! %d %d %d %d ", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
printf("??????? %d %d %d %d ", arg_b[0], arg_b[1], arg_b[2], arg_b[3]);
reg_c.template AsType<float16_t>()(Number<0>{}) =
__builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4(
arg_type{arg_a[0], arg_a[1], arg_a[2], arg_a[3], 0, 0, 0, 0},
@@ -657,6 +663,9 @@ struct intrin_mfma_scale_f32_16x16x128f8f6f4<16, 16>
using arg_type = int32x8_t;
printf("!!!!!!! %d %d %d %d ", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
printf("??????? %d %d %d %d ", arg_b[0], arg_b[1], arg_b[2], arg_b[3]);
reg_c.template AsType<float4_t>()(Number<0>{}) =
__builtin_amdgcn_mfma_scale_f32_16x16x128_f8f6f4(
arg_type{arg_a[0], arg_a[1], arg_a[2], arg_a[3], 0, 0, 0, 0},
@@ -722,8 +731,6 @@ struct intrin_mfma_f32_16x16x128f8f6f4<16, 16>
using arg_type = int32x8_t;
static_assert(0 == 0);
printf("!!!!!!! %d %d %d %d ", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
printf("??????? %d %d %d %d ", arg_b[0], arg_b[1], arg_b[2], arg_b[3]);