ggml-webgpu: port all AOT operators to JIT (#20728)

* port cpy pipeline to shader lib with JIT compilation
 * port glu pipeline to shader lib with JIT compilation
 * port rope pipeline to shader lib with JIT compilation
 * port soft_max pipeline to shader lib with JIT compilation
 * removed unused functions from embed_wgsl.py which were used for
old AOT template expansion
This commit is contained in:
Abhijit Ramesh
2026-03-31 15:38:16 -07:00
committed by GitHub
parent 0fcb3760b2
commit 825eb91a66
8 changed files with 764 additions and 1001 deletions
@@ -0,0 +1,81 @@
enable f16;
#ifdef SRC_F32
#define SRC_TYPE f32
#elif defined(SRC_F16)
#define SRC_TYPE f16
#endif
#ifdef DST_F32
#define DST_TYPE f32
#elif defined(DST_F16)
#define DST_TYPE f16
#elif defined(DST_I32)
#define DST_TYPE i32
#endif
@group(0) @binding(0)
var<storage, read_write> src: array<SRC_TYPE>;
@group(0) @binding(1)
var<storage, read_write> dst: array<DST_TYPE>;
struct Params{
ne: u32,
offset_src: u32,
offset_dst: u32,
stride_src0: u32,
stride_src1: u32,
stride_src2: u32,
stride_src3: u32,
stride_dst0: u32,
stride_dst1: u32,
stride_dst2: u32,
stride_dst3: u32,
src_ne0: u32,
src_ne1: u32,
src_ne2: u32,
dst_ne0: u32,
dst_ne1: u32,
dst_ne2: u32
};
@group(0) @binding(2)
var<uniform> params: Params;
@compute @workgroup_size(WG_SIZE)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
if (gid.x >= params.ne) {
return;
}
var i = gid.x;
let i3 = i / (params.src_ne2 * params.src_ne1 * params.src_ne0);
i = i % (params.src_ne2 * params.src_ne1 * params.src_ne0);
let i2 = i / (params.src_ne1 * params.src_ne0);
i = i % (params.src_ne1 * params.src_ne0);
let i1 = i / params.src_ne0;
let i0 = i % params.src_ne0;
var j = gid.x;
let j3 = j / (params.dst_ne2 * params.dst_ne1 * params.dst_ne0);
j = j % (params.dst_ne2 * params.dst_ne1 * params.dst_ne0);
let j2 = j / (params.dst_ne1 * params.dst_ne0);
j = j % (params.dst_ne1 * params.dst_ne0);
let j1 = j / params.dst_ne0;
let j0 = j % params.dst_ne0;
let src_idx = i0 * params.stride_src0 + i1 * params.stride_src1 +
i2 * params.stride_src2 + i3 * params.stride_src3;
let dst_idx = j0 * params.stride_dst0 + j1 * params.stride_dst1 +
j2 * params.stride_dst2 + j3 * params.stride_dst3;
dst[params.offset_dst + dst_idx] = DST_TYPE((src[params.offset_src + src_idx]));
}