ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-04 05:31:24 +00:00
Code Issues Packages Projects Releases Wiki Activity

Implement the fp16xint4 scale weight only kernel for Ali (#1786)

Browse Source 
* enable int4 scale (weight only) kernel

* format some files

* Add unit test for int4 weight only

* fixed and formatted code

* fixed

* formated

* formated

* fixed

* fixed a bug in the ckProfiler, and formatted the code

---------

Co-authored-by: mtgu0705 <mtgu@amd.com>
This commit is contained in:
Mingtao Gu
2025-01-03 18:35:21 +08:00
committed by GitHub
parent 4bc610416a
commit 4f62f6e9b7
21 changed files with 7562 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
include/ck/utility/amd_inline_asm.hpp
View File

@@ -4,8 +4,8 @@
#ifndef CK_AMD_INLINE_ASM_HPP
#define CK_AMD_INLINE_ASM_HPP
#include "data_type.hpp"
#include "c_style_pointer_cast.hpp"
#include "data_type.hpp"
// TODO: deprecate all amd_assembly_outer_product_xxx
@@ -21,14 +21,14 @@ inline __device__ int amd_assembly_and_or_b32(int a, int b, int d)
inline __device__ half2_t amd_assembly_pk_fma_f16(half2_t a, half2_t b, half2_t c)
{
half2_t d;
asm volatile("v_pk_fma_f16 %0, %1, %2, %3;\n" : "=v"(d) : "v"(a), "v"(b), "v"(c));
asm volatile("v_pk_fma_f16 %0, %1, %2, %3" : "=v"(d) : "v"(a), "v"(b), "v"(c));
return d;
}
inline __device__ half2_t amd_assembly_pk_add_f16(half2_t a, half2_t b)
{
half2_t c;
asm volatile("v_pk_add_f16 %0, %1, %2;\n" : "=v"(c) : "v"(a), "v"(b));
asm volatile("v_pk_add_f16 %0, %1, %2" : "=v"(c) : "v"(a), "v"(b));
return c;
}