Temp edits

This commit is contained in:
Rostyslav Geyyer
2025-04-30 16:36:53 +00:00
parent b068584afb
commit 44f47ac7c5

View File

@@ -520,8 +520,8 @@ struct intrin_mfma_f32_32x32x64f8f6f4<32, 32>
using arg_type = int32x8_t;
printf("!!!!!!! %d %d %d %d \n", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
printf("??????? %d %d %d %d \n", arg_b[0], arg_b[1], arg_b[2], arg_b[3]);
// printf("!!!!!!! %d %d %d %d \n", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
// printf("??????? %d %d %d %d \n", 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(
@@ -591,8 +591,8 @@ struct intrin_mfma_scale_f32_32x32x64f8f6f4<32, 32>
using arg_type = int32x8_t;
printf("!!!!!!! %d %d %d %d \n", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
printf("??????? %d %d %d %d \n", arg_b[0], arg_b[1], arg_b[2], arg_b[3]);
// printf("!!!!!!! %d %d %d %d \n", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
// printf("??????? %d %d %d %d \n", 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(
@@ -663,8 +663,8 @@ struct intrin_mfma_scale_f32_16x16x128f8f6f4<16, 16>
using arg_type = int32x8_t;
printf("!!!!!!! %d %d %d %d \n", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
printf("??????? %d %d %d %d \n", arg_b[0], arg_b[1], arg_b[2], arg_b[3]);
// printf("!!!!!!! %d %d %d %d \n", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
// printf("??????? %d %d %d %d \n", 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(
@@ -731,8 +731,8 @@ struct intrin_mfma_f32_16x16x128f8f6f4<16, 16>
using arg_type = int32x8_t;
printf("!!!!!!! %d %d %d %d \n", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
printf("??????? %d %d %d %d \n", arg_b[0], arg_b[1], arg_b[2], arg_b[3]);
// printf("!!!!!!! %d %d %d %d \n", arg_a[0], arg_a[1], arg_a[2], arg_a[3]);
// printf("??????? %d %d %d %d \n", 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(