CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-11 10:36:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: Add bfloat16 support (#12554)

Browse Source 
* vulkan: Add bfloat16 support

This adds bfloat16 matrix multiply support based on VK_KHR_shader_bfloat16.
The extension is required for coopmat multiply support, but matrix-vector
multiply trivially promotes bf16 to fp32 and doesn't require the extension.
The copy/get_rows shaders also don't require the extension.

It's probably possible to fall back to non-coopmat and promote to fp32 when
the extension isn't supported, but this change doesn't do that.

The coopmat support also requires a glslc that supports the extension, which
currently requires a custom build.

* vulkan: Support bf16 tensors without the bf16 extension or coopmat support

Compile a variant of the scalar mul_mm shader that will promote the bf16
values to float, and use that when either the bf16 extension or the coopmat
extensions aren't available.

* vulkan: bfloat16 fixes (really works without bfloat16 support now)

* vulkan: fix spirv-val failure and reenable -O
This commit is contained in:
Jeff Bolz
2025-05-01 13:49:39 -05:00
committed by GitHub
parent fc727bcdd5
commit 79f26e9e12
13 changed files with 368 additions and 67 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp
View File

@@ -20,9 +20,14 @@ void main() {
const uint a_offset = get_aoffset() + i01*p.nb01 + i11*p.nb02 + i12*p.nb03;
const uint d_offset = get_doffset() + i10*p.nb21 + i11*p.nb22 + i12*p.nb23;
#ifndef OPTIMIZATION_ERROR_WORKAROUND
data_d[d_offset + i00] = D_TYPE(data_a[a_offset + i00]);
#if defined(DATA_A_BF16)
FLOAT_TYPE v = FLOAT_TYPE(bf16_to_fp32(data_a[a_offset + i00]));
#else
data_d[d_offset + i00] = data_a[a_offset + i00];
FLOAT_TYPE v = FLOAT_TYPE(data_a[a_offset + i00]);
#endif
#ifndef OPTIMIZATION_ERROR_WORKAROUND
data_d[d_offset + i00] = D_TYPE(v);
#else
data_d[d_offset + i00] = D_TYPE(v);
#endif
}