CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-13 10:57:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: fastdiv, launch bounds for mmvq + q8_1 quant (#15802)

Browse Source 
* CUDA: fastdiv, launch bounds for mmvq + q8_1 quant
This commit is contained in:
Johannes Gäßler
2025-09-05 16:07:02 +02:00
committed by GitHub
parent 3a550b5ca4
commit 5143fa895e
3 changed files with 67 additions and 77 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
ggml/src/ggml-cuda/quantize.cu
View File

@@ -1,26 +1,27 @@
#include "quantize.cuh"
#include <cstdint>
__launch_bounds__(CUDA_QUANTIZE_BLOCK_SIZE, 1)
static __global__ void quantize_q8_1(
const float * __restrict__ x, void * __restrict__ vy,
const int64_t ne00, const int64_t s01, const int64_t s02, const int64_t s03,
const int64_t ne0, const int ne1, const int ne2) {
const int64_t ne0, const uint32_t ne1, const uint3 ne2) {
const int64_t i0 = (int64_t)blockDim.x*blockIdx.x + threadIdx.x;
if (i0 >= ne0) {
return;
}
const int64_t i3 = fastdiv(blockIdx.z, ne2);
const int64_t i2 = blockIdx.z - i3*ne2.z;
const int64_t i1 = blockIdx.y;
const int64_t i2 = blockIdx.z % ne2;
const int64_t i3 = blockIdx.z / ne2;
const int64_t & i00 = i0;
const int64_t & i01 = i1;
const int64_t & i02 = i2;
const int64_t & i03 = i3;
const int64_t i_cont = ((i3*ne2 + i2) * ne1 + i1) * ne0 + i0;
const int64_t i_cont = ((i3*ne2.z + i2) * ne1 + i1) * ne0 + i0;
block_q8_1 * y = (block_q8_1 *) vy;
@@ -31,10 +32,10 @@ static __global__ void quantize_q8_1(
float amax = fabsf(xi);
float sum = xi;
amax = warp_reduce_max(amax);
sum = warp_reduce_sum(sum);
amax = warp_reduce_max<QK8_1>(amax);
sum = warp_reduce_sum<QK8_1>(sum);
const float d = amax / 127;
const float d = amax / 127.0f;
const int8_t q = amax == 0.0f ? 0 : roundf(xi / d);
y[ib].qs[iqs] = q;
@@ -43,8 +44,7 @@ static __global__ void quantize_q8_1(
return;
}
reinterpret_cast<half&>(y[ib].ds.x) = d;
reinterpret_cast<half&>(y[ib].ds.y) = sum;
y[ib].ds = make_half2(d, sum);
}
template <mmq_q8_1_ds_layout ds_layout>
@@ -152,10 +152,12 @@ void quantize_row_q8_1_cuda(
GGML_ASSERT(!ids);
GGML_ASSERT(ne0 % QK8_1 == 0);
const uint3 ne2_fastdiv = init_fastdiv_values(ne2);
const int64_t block_num_x = (ne0 + CUDA_QUANTIZE_BLOCK_SIZE - 1) / CUDA_QUANTIZE_BLOCK_SIZE;
const dim3 num_blocks(block_num_x, ne1, ne2*ne3);
const dim3 block_size(CUDA_QUANTIZE_BLOCK_SIZE, 1, 1);
quantize_q8_1<<<num_blocks, block_size, 0, stream>>>(x, vy, ne00, s01, s02, s03, ne0, ne1, ne2);
quantize_q8_1<<<num_blocks, block_size, 0, stream>>>(x, vy, ne00, s01, s02, s03, ne0, ne1, ne2_fastdiv);
GGML_UNUSED(type_src0);
}