Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

HIP: Cleanup hipification header (#15285)

Browse Source 
add expicit conversion operator to support older versions of rocm
Switch over to hip_bf16 from legacy hip_bfloat16
Simplify RDNA3 define
Reduce swap over of new hipblas api to rocm 6.5 as this version is used for rocm 7.0 previews

---------

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
This commit is contained in:
uvos
2025-08-14 16:23:56 +02:00
committed by GitHub
co-authored by Johannes Gäßler
parent b204a5a234
commit 5ba36f6103
7 changed files with 32 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/convert.cu
+3 -3
View File
@@ -31,8 +31,8 @@ static __global__ void dequantize_block(const void * __restrict__ vx, dst_t * __
dequantize_kernel(vx, ib, iqs, v);
const int64_t iy0 = ((i03*ne02 + i02)*ne01 + i01)*ne00 + iybs + iqs;
y[iy0 + 0] = float(v.x);
y[iy0 + y_offset] = float(v.y);
y[iy0 + 0] = ggml_cuda_cast<dst_t>(v.x);
y[iy0 + y_offset] = ggml_cuda_cast<dst_t>(v.y);
}
template <bool need_check>
@@ -630,7 +630,7 @@ static __global__ void convert_unary(
const int64_t ix = i03*s03 + i02*s02 + i01*s01 + i00;
const int64_t iy = ((i03*ne02 + i02)*ne01 + i01)*ne00 + i00;
y[iy] = float(x[ix]);
y[iy] = ggml_cuda_cast<dst_t>(x[ix]);
}
template <typename src_t, typename dst_t>