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

ggml-webgpu: address quantization precision and backend lifecycle managment (#21521)

Browse Source 
* ggml(webgpu): fix the busy-polls in Emscripten  in the waitAny after #20618, and remove the busy webgpu log

* Merge with upstream

* Fix GET_ROWS packed integer NaN when using f16 as memory buffer in shader quants

* Update Unary wgsl EXP and EXPM1 for f16 stability

* Fix GET_ROWS IQ4_XS strcut for NaN f16 canonicalization

* Fix numerical percision for unary sqrt when working with f16

* Fix NaN canonicalization for packed integers using f16

* Update err threshold for binary div ops when using f16

* backend: Keep one Dawn/WebGPU instance alive for the lifetime of the static backend

* clean: uncomment existing code logs

* clean: clean the unncessary debug info

* Refactor and generalize dequant helpers

* Remove deprecated quant structs

* Refactor shader defines to reduce repetition

* Remove error override for F16 type

* fix: fix the accidential removal of the proper initialization of ctx

* clean: clean legacy and format code

* fix: did not modify tests ops

---------

Co-authored-by: Jeremy J. Hartmann <jeremy@mtion.tv>
This commit is contained in:
Chen Yuan
2026-04-10 10:52:01 -07:00
committed by GitHub
co-authored by Jeremy J. Hartmann
parent 5dd102539b
commit e4fed9d08d
8 changed files with 383 additions and 330 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp
+52 -3
View File
@@ -1115,6 +1115,32 @@ class ggml_webgpu_shader_lib {
std::string type_upper = type_str;
std::transform(type_upper.begin(), type_upper.end(), type_upper.begin(), ::toupper);
switch (key.src_type)
{
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q5_0:
case GGML_TYPE_Q8_0:
case GGML_TYPE_Q3_K:
case GGML_TYPE_Q6_K:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_IQ3_XXS:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ1_S:
case GGML_TYPE_IQ4_NL:
{
// Quantized types using u32 buffers for portability.
defines.push_back("SRC_TYPE=u32");
defines.push_back("U32_DEQUANT_HELPERS");
break;
}
default:
{
defines.push_back(std::string("SRC_TYPE=") + type_str);
}
}
defines.push_back("BYTE_HELPERS");
defines.push_back(type_upper + "_T");
defines.push_back(type_upper);
@@ -1125,7 +1151,6 @@ class ggml_webgpu_shader_lib {
variant += "_";
variant += type_str;
defines.push_back(std::string("SRC_TYPE=") + type_str);
defines.push_back("DST_TYPE=f32");
if ((key.src_type >= GGML_TYPE_Q4_0 && key.src_type <= GGML_TYPE_Q8_1) ||
@@ -1593,11 +1618,35 @@ class ggml_webgpu_shader_lib {
break;
default:
{
// quantized types
std::string type_upper = src0_name;
std::transform(type_upper.begin(), type_upper.end(), type_upper.begin(), ::toupper);
defines.push_back(std::string("SRC0_TYPE=") + src0_name);
switch (context.src0->type)
{
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q5_0:
case GGML_TYPE_Q8_0:
case GGML_TYPE_Q3_K:
case GGML_TYPE_Q6_K:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_IQ3_XXS:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ1_S:
case GGML_TYPE_IQ4_NL:
{
// Quantized types using u32 buffers for portability.
defines.push_back("SRC0_TYPE=u32");
defines.push_back("U32_DEQUANT_HELPERS");
break;
}
default:
{
defines.push_back(std::string("SRC0_TYPE=") + src0_name);
}
}
defines.push_back("BYTE_HELPERS");
defines.push_back(type_upper + "_T");
defines.push_back(type_upper);