0cc4m 4a6d52efe6 Vulkan: VK_KHR_cooperative_matrix support to speed up prompt processing (llama/10597) ...

* Vulkan: Implement VK_KHR_cooperative_matrix support in the matrix matrix multiplication shader

* Improve performance with better q4_k and q5_k dequant and store unrolling

* Add Vulkan MUL_MAT and MUL_MAT_ID accumulator precision selection

* Rework mulmat shader selection and compilation logic, avoid compiling shaders that won't get used by device

* Vulkan: Implement accumulator switch for specific mul mat mat shaders

* Vulkan: Unroll more loops for more mul mat mat performance

* Vulkan: Add VK_AMD_shader_core_properties2 support to read Compute Unit count for split_k logic

* Disable coopmat support on AMD proprietary driver

* Remove redundant checks

* Add environment variable GGML_VK_DISABLE_COOPMAT to disable VK_KHR_cooperative_matrix support

* Fix rebase typo

* Fix coopmat2 MUL_MAT_ID pipeline selection

2024-12-18 12:52:16 +02:00

..

acc.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

add.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

argsort.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

clamp.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

CMakeLists.txt

vulkan: Add VK_NV_cooperative_matrix2 support for mul_mat and flash attention (llama/10206)

2024-12-18 12:52:16 +02:00

concat.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

contig_copy.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

copy.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

cos.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_f32.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_funcs_cm2.comp

vulkan: Add VK_NV_cooperative_matrix2 support for mul_mat and flash attention (llama/10206)

2024-12-18 12:52:16 +02:00

dequant_funcs.comp

vulkan: further optimize mul_mat_vec using larger loads (llama/10387)

2024-12-08 20:14:35 +02:00

dequant_head.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_iq4_nl.comp

vulkan: copy iq4_nl LUT into shared memory (llama/10409)

2024-12-08 20:14:35 +02:00

dequant_q2_k.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q3_k.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q4_0.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q4_1.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q4_k.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q5_0.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q5_1.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q5_k.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q6_k.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

dequant_q8_0.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

diag_mask_inf.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

div.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

flash_attn_cm2.comp

vulkan: Add VK_NV_cooperative_matrix2 support for mul_mat and flash attention (llama/10206)

2024-12-18 12:52:16 +02:00

gelu_quick.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

gelu.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

generic_binary_head.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

generic_head.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

generic_unary_head.comp

vulkan: Implement "fast divide" (mul+shift) for unary ops like copy (llama/10642)

2024-12-08 20:14:35 +02:00

get_rows_quant.comp

vulkan: copy iq4_nl LUT into shared memory (llama/10409)

2024-12-08 20:14:35 +02:00

get_rows.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

group_norm.comp

vulkan: fix group_norm (llama/10496)

2024-12-08 20:14:35 +02:00

im2col.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

leaky_relu.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

mul_mat_split_k_reduce.comp

vulkan: optimize and reenable split_k (llama/10637)

2024-12-08 20:14:35 +02:00

mul_mat_vec_base.comp

vulkan: skip integer div/mod in get_offsets for batch_idx==0 (llama/10506)

2024-12-08 20:14:35 +02:00

mul_mat_vec_nc.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

mul_mat_vec_p021.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

mul_mat_vec_q2_k.comp

vulkan: optimize Q2_K and Q3_K mul_mat_vec (llama/10459)

2024-12-08 20:14:35 +02:00

mul_mat_vec_q3_k.comp

vulkan: optimize Q2_K and Q3_K mul_mat_vec (llama/10459)

2024-12-08 20:14:35 +02:00

mul_mat_vec_q4_k.comp

vulkan: further optimize mul_mat_vec using larger loads (llama/10387)

2024-12-08 20:14:35 +02:00

mul_mat_vec_q5_k.comp

vulkan: further optimize q5_k mul_mat_vec (llama/10479)

2024-12-08 20:14:35 +02:00

mul_mat_vec_q6_k.comp

vulkan: Dynamic subgroup size support for Q6_K mat_vec (llama/10536)

2024-12-08 20:14:35 +02:00

mul_mat_vec.comp

vulkan: copy iq4_nl LUT into shared memory (llama/10409)

2024-12-08 20:14:35 +02:00

mul_mm_cm2.comp

vulkan: Add VK_NV_cooperative_matrix2 support for mul_mat and flash attention (llama/10206)

2024-12-18 12:52:16 +02:00

mul_mm.comp

Vulkan: VK_KHR_cooperative_matrix support to speed up prompt processing (llama/10597)

2024-12-18 12:52:16 +02:00

mul.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

norm.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

pad.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

pool2d.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

relu.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

repeat.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

rms_norm.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

rope_head.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

rope_neox.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

rope_norm.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

scale.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

silu.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

sin.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

soft_max.comp

vulkan : fix soft_max.comp division by zero (#2633)

2024-12-16 12:34:38 +02:00

square.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

sum_rows.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

tanh.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

timestep_embedding.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

types.comp

vulkan: define all quant data structures in types.comp (llama/10440)

2024-12-08 20:14:35 +02:00

upscale.comp

ggml : build backends as libraries (llama/10256)

2024-11-20 21:00:08 +02:00

vulkan-shaders-gen.cpp

Vulkan: VK_KHR_cooperative_matrix support to speed up prompt processing (llama/10597)

2024-12-18 12:52:16 +02:00