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

ggml-webgpu: Support non-contiguous src0 and overlapping src0/src1 in binary ops (#19850)

Browse Source 
* ggml-webgpu: Add binary op support for overlapping and non-contiguous.

* Add newline to binary.wgsl

* Append the test of binary op for src overlapping  to test_bin_bcast.

* Remove unnecessary newline.
This commit is contained in:
Masashi Yoshimura
2026-03-02 07:59:53 -08:00
committed by GitHub
parent feefb92836
commit 36a7a6589c
4 changed files with 131 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-webgpu/wgsl-shaders/binary.wgsl
+44 -10
View File
@@ -7,6 +7,13 @@ struct Params {
offset_src0: u32,
offset_src1: u32,
offset_dst: u32,
offset_merged_src0: u32,
offset_merged_src1: u32,
stride_src0_0: u32,
stride_src0_1: u32,
stride_src0_2: u32,
stride_src0_3: u32,
stride_src1_0: u32,
stride_src1_1: u32,
@@ -23,6 +30,21 @@ struct Params {
b_ne3: u32,
};
fn src0_index(_i: u32) -> u32 {
var i = _i;
let a_i3 = i / (params.a_ne2 * params.a_ne1 * params.a_ne0);
i = i % (params.a_ne2 * params.a_ne1 * params.a_ne0);
let a_i2 = i / (params.a_ne1 * params.a_ne0);
i = i % (params.a_ne1 * params.a_ne0);
let a_i1 = i / params.a_ne0;
let a_i0 = i % params.a_ne0;
return a_i0 * params.stride_src0_0 +
a_i1 * params.stride_src0_1 +
a_i2 * params.stride_src0_2 +
a_i3 * params.stride_src0_3;
}
fn src1_index(_i: u32) -> u32 {
var i = _i;
let a_i3 = i / (params.a_ne2 * params.a_ne1 * params.a_ne0);
@@ -53,17 +75,22 @@ fn src1_index(_i: u32) -> u32 {
#define DataType f16
#endif
#ifdef SRC_OVERLAP
@group(0) @binding(0)
var<storage, read_write> merged_src: array<DataType>;
@group(0) @binding(1)
var<storage, read_write> dst: array<DataType>;
@group(0) @binding(2)
var<uniform> params: Params;
#else
@group(0) @binding(0)
var<storage, read_write> src0: array<DataType>;
@group(0) @binding(1)
var<storage, read_write> src1 : array<DataType>;
#ifdef INPLACE
@group(0) @binding(2)
var<uniform> params: Params;
#elif defined(OVERLAP)
#if defined(INPLACE) || defined(OVERLAP)
@group(0) @binding(2)
var<uniform> params: Params;
@@ -74,6 +101,7 @@ var<storage, read_write> dst: array<DataType>;
@group(0) @binding(3)
var<uniform> params: Params;
#endif
#endif
fn op(a: DataType, b: DataType) -> DataType {
#ifdef OP_ADD
@@ -87,13 +115,17 @@ fn op(a: DataType, b: DataType) -> DataType {
#endif
}
fn update(dst_i: u32, src0_i: u32, src1_i: u32){
fn update(dst_i: u32, src0_i: u32, src1_i: u32) {
#ifdef SRC_OVERLAP
let result = op(merged_src[src0_i], merged_src[src1_i]);
#else
let result = op(src0[src0_i], src1[src1_i]);
#endif
#ifdef INPLACE
src0[dst_i] = result;
src0[src0_i] = result;
#elif defined(OVERLAP)
src1[dst_i] = result;
src1[src1_i] = result;
#else
dst[dst_i] = result;
#endif
@@ -102,6 +134,8 @@ fn update(dst_i: u32, src0_i: u32, src1_i: u32){
@compute @workgroup_size(WG_SIZE)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
if (gid.x < params.ne) {
update(params.offset_dst + gid.x, params.offset_src0 + gid.x, params.offset_src1 + src1_index(gid.x));
let src0_i = params.offset_src0 + params.offset_merged_src0 + src0_index(gid.x);
let src1_i = params.offset_src1 + params.offset_merged_src1 + src1_index(gid.x);
update(params.offset_dst + gid.x, src0_i, src1_i);
}
}