From b333b06772c89d96aacb5490d6a219fba7c09cc6 Mon Sep 17 00:00:00 2001 From: Mitja Felicijan Date: Thu, 12 Feb 2026 20:57:17 +0100 Subject: Engage! --- .../src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl | 90 ++++++++++++++++++++++ 1 file changed, 90 insertions(+) create mode 100644 llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl (limited to 'llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl') diff --git a/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl b/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl new file mode 100644 index 0000000..040e80d --- /dev/null +++ b/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl @@ -0,0 +1,90 @@ +#define(VARIANTS) + +[ + { + "SHADER_NAME": "scale_f32", + "DECLS": ["NOT_INPLACE"] + }, + { + "SHADER_NAME": "scale_f32_inplace", + "DECLS": ["INPLACE"] + } +] + +#end(VARIANTS) + +#define(DECLS) + +#decl(NOT_INPLACE) +@group(0) @binding(1) +var dst: array; + +@group(0) @binding(2) +var params: Params; + +fn store_scale(val: f32, offset: u32) { + dst[offset] = val; +} +#enddecl(NOT_INPLACE) + +#decl(INPLACE) +@group(0) @binding(1) +var params: Params; + +fn store_scale(val: f32, offset: u32) { + src[offset] = val; +} +#enddecl(INPLACE) + +#end(DECLS) + +#define(SHADER) + +struct Params { + offset_src: u32, + offset_dst: u32, + + // Strides (in elements) + stride_src1: u32, + stride_src2: u32, + stride_src3: u32, + + stride_dst1: u32, + stride_dst2: u32, + stride_dst3: u32, + + ne: u32, + ne0: u32, + ne1: u32, + ne2: u32, + + scale: f32, + bias: f32 +}; + +@group(0) @binding(0) +var src: array; + +DECLS + +override wg_size: u32; +@compute @workgroup_size(wg_size) +fn main(@builtin(global_invocation_id) gid: vec3) { + if (gid.x >= params.ne) { + return; + } + + var i = gid.x; + let i3 = i / (params.ne2 * params.ne1 * params.ne0); + i = i % (params.ne2 * params.ne1 * params.ne0); + let i2 = i / (params.ne1 * params.ne0); + i = i % (params.ne1 * params.ne0); + let i1 = i / params.ne0; + let i0 = i % params.ne0; + + let i_src = params.offset_src + i3 * params.stride_src3 + i2 * params.stride_src2 + i1 * params.stride_src1 + i0; + let i_dst = params.offset_dst + i3 * params.stride_dst3 + i2 * params.stride_dst2 + i1 * params.stride_dst1 + i0; + + store_scale(src[i_src] * params.scale + params.bias, i_dst); +} +#end(SHADER) -- cgit v1.2.3