Jeff Bolz e920c523e3 vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338) ...

This uses the new VK_EXT_shader_ocp_microscaling_types extension to do fp4 type
promotions, and also uses the float8 extension to do ue4m3 promotions for
nvfp4. It's reasonable to assume that an implementation that supports fp4 will
also support fp8, so we don't need to handle all possible combinations of
support.

2026-07-13 08:44:17 -05:00

..

feature-tests

vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338)

2026-07-13 08:44:17 -05:00

acc.comp

vulkan: support GGML_OP_SET (#19584)

2026-02-14 06:36:38 +01:00

add1.comp

vulkan: implement ADD1, ARANGE, FILL, SOFTPLUS, STEP, ROUND, CEIL, FLOOR, TRUNC (#17319)

2025-11-19 17:29:45 +01:00

add_id.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

add.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

arange.comp

vulkan: implement ADD1, ARANGE, FILL, SOFTPLUS, STEP, ROUND, CEIL, FLOOR, TRUNC (#17319)

2025-11-19 17:29:45 +01:00

argmax.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

argsort_large.comp

vulkan: support larger argsort (#17313)

2025-11-19 17:25:50 +01:00

argsort.comp

vulkan: support larger argsort (#17313)

2025-11-19 17:25:50 +01:00

CMakeLists.txt

vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338)

2026-07-13 08:44:17 -05:00

col2im_1d.comp

vulkan: add col2im_1d op (#24425)

2026-06-16 06:34:43 +02:00

concat.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

contig_copy.comp

vulkan: add cpy bf16 -> f32 pipelines (#22677)

2026-05-17 11:31:20 +02:00

conv2d_dw.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

conv2d_mm.comp

vulkan: Workaround compiler bug in conv2d coopmat2 path (#24924)

2026-06-26 11:53:32 +02:00

conv3d_mm.comp

vulkan: Workaround compiler bug in conv2d coopmat2 path (#24924)

2026-06-26 11:53:32 +02:00

conv_transpose_1d.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

copy_from_quant.comp

vulkan: Support GGML_TYPE_NVFP4 (#21455)

2026-04-14 11:34:23 +02:00

copy_to_quant.comp

vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (#21572)

2026-04-14 15:17:45 +02:00

copy_transpose.comp

vulkan: Add copy_transpose shader (#17371)

2025-11-19 16:50:43 +01:00

copy.comp

vulkan: add cpy bf16 -> f32 pipelines (#22677)

2026-05-17 11:31:20 +02:00

count_equal.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

count_experts.comp

vulkan: preprocess mul_mat_id experts and discard workgroups more quickly (#18352)

2025-12-26 16:12:58 -06:00

cumsum_multipass1.comp

vulkan: Optimize GGML_OP_CUMSUM (#18417)

2026-01-02 15:32:30 -06:00

cumsum_multipass2.comp

vulkan: Optimize GGML_OP_CUMSUM (#18417)

2026-01-02 15:32:30 -06:00

cumsum.comp

vulkan: Optimize GGML_OP_CUMSUM (#18417)

2026-01-02 15:32:30 -06:00

dequant_f32.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_funcs_cm2.glsl

vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338)

2026-07-13 08:44:17 -05:00

dequant_funcs.glsl

vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338)

2026-07-13 08:44:17 -05:00

dequant_head.glsl

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq1_m.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq1_s.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq2_s.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq2_xs.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq2_xxs.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq3_s.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq3_xxs.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq4_nl.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_iq4_xs.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_mxfp4.comp

Vulkan MMQ Integer Dot Refactor and K-Quant support (#16536)

2025-10-29 14:39:03 +01:00

dequant_nvfp4.comp

vulkan: Support GGML_TYPE_NVFP4 (#21455)

2026-04-14 11:34:23 +02:00

dequant_q1_0.comp

vulkan: Support Q1_0 (#21539)

2026-04-10 08:35:27 +02:00

dequant_q2_k.comp

Vulkan MMQ Integer Dot Refactor and K-Quant support (#16536)

2025-10-29 14:39:03 +01:00

dequant_q3_k.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_q4_0.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_q4_1.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_q4_k.comp

Vulkan MMQ Integer Dot Refactor and K-Quant support (#16536)

2025-10-29 14:39:03 +01:00

dequant_q5_0.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_q5_1.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_q5_k.comp

Vulkan MMQ Integer Dot Refactor and K-Quant support (#16536)

2025-10-29 14:39:03 +01:00

dequant_q6_k.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dequant_q8_0.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

diag_mask_inf.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

diag.comp

vulkan: support non-contig unary/glu ops (#24215)

2026-06-13 08:44:15 -05:00

div.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

dot_product_funcs.glsl

vulkan: add v_dot2_f32_f16 support in matrix-matrix multiplication and Flash Attention (#24123)

2026-06-09 13:27:04 +02:00

fill.comp

vulkan: implement ADD1, ARANGE, FILL, SOFTPLUS, STEP, ROUND, CEIL, FLOOR, TRUNC (#17319)

2025-11-19 17:29:45 +01:00

flash_attn_base.glsl

vulkan: add Flash Attention support for BFloat16 KV cache (#23420)

2026-05-30 10:39:31 +02:00

flash_attn_cm1.comp

vulkan: Apply bias before softmax in FA, to avoid overflow (#24909)

2026-06-23 22:34:00 -05:00

flash_attn_cm2.comp

vulkan: add Flash Attention support for BFloat16 KV cache (#23420)

2026-05-30 10:39:31 +02:00

flash_attn_dequant.glsl

vulkan: add Flash Attention support for BFloat16 KV cache (#23420)

2026-05-30 10:39:31 +02:00

flash_attn_mask_opt.comp

vulkan: fix OOB check in flash_attn_mask_opt (#20296)

2026-03-12 06:35:49 +01:00

flash_attn_mmq_funcs.glsl

vulkan: Support asymmetric FA in scalar/mmq/coopmat1 paths (#22589)

2026-05-11 12:49:03 +02:00

flash_attn_split_k_reduce.comp

vulkan: support flash attention GQA/split_k with small batches (#18938)

2026-01-21 17:43:43 +01:00

flash_attn.comp

vulkan: Apply bias before softmax in FA, to avoid overflow (#24909)

2026-06-23 22:34:00 -05:00

fwht.comp

vulkan: add fwht support for Intel with shmem reduction (#23964)

2026-06-05 19:44:40 +02:00

gated_delta_net.comp

Remove padding and multiple D2D copies for MTP (#24086)

2026-06-10 23:21:16 +05:30

geglu_erf.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

geglu_quick.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

geglu.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

generic_binary_head.glsl

vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (#21572)

2026-04-14 15:17:45 +02:00

generic_head.glsl

vulkan: support GGML_UNARY_OP_XIELU (#18062)

2025-12-21 10:17:58 +01:00

generic_unary_head.glsl

vulkan: support non-contig unary/glu ops (#24215)

2026-06-13 08:44:15 -05:00

get_rows_back.comp

vulkan: Support GET_ROWS_BACK (#24883)

2026-06-23 15:39:37 +02:00

get_rows_quant.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

get_rows.comp

vulkan: support get_rows for i32 (#17941)

2025-12-13 10:12:53 +01:00

glu_head.glsl

vulkan: support non-contig unary/glu ops (#24215)

2026-06-13 08:44:15 -05:00

glu_main.glsl

vulkan: support non-contig unary/glu ops (#24215)

2026-06-13 08:44:15 -05:00

group_norm.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

im2col_3d.comp

vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (#21572)

2026-04-14 15:17:45 +02:00

im2col.comp

vulkan: optimize operations in the IM2COL shader (#22685)

2026-05-20 17:15:13 +02:00

l2_norm.comp

vulkan: support all backend tests for SQR/SQRT/SIN/COS/CLAMP/LEAKY_RELU/NORM (#24582)

2026-06-23 09:48:24 -05:00

log.comp

vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (#21572)

2026-04-14 15:17:45 +02:00

mul_mat_split_k_reduce.comp

vulkan: optimize and reenable split_k (#10637)

2024-12-03 20:29:54 +01:00

mul_mat_vec_base.glsl

vulkan: split mul_mat into multiple dispatches to avoid overflow (#19509)

2026-02-18 10:47:10 +01:00

mul_mat_vec_iface.glsl

vulkan: unify type macros to use Vx instead of _VECx (#21605)

2026-04-09 07:31:51 +02:00

mul_mat_vec_iq1_m.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec_iq1_s.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec_iq2_s.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec_iq2_xs.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec_iq2_xxs.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec_iq3_s.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec_iq3_xxs.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec_nc.comp

vulkan: Fuse mul_mat_id+add_id+mul and mul_mat+add+add. (#17287)

2025-11-15 19:54:23 +01:00

mul_mat_vec_p021.comp

vulkan: Fuse mul_mat_id+add_id+mul and mul_mat+add+add. (#17287)

2025-11-15 19:54:23 +01:00

mul_mat_vec_q2_k.comp

vulkan: unify type macros to use Vx instead of _VECx (#21605)

2026-04-09 07:31:51 +02:00

mul_mat_vec_q3_k.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec_q4_k.comp

vulkan: unify type macros to use Vx instead of _VECx (#21605)

2026-04-09 07:31:51 +02:00

mul_mat_vec_q5_k.comp

vulkan: unify type macros to use Vx instead of _VECx (#21605)

2026-04-09 07:31:51 +02:00

mul_mat_vec_q6_k.comp

vulkan: Use VK_EXT_shader_64bit_indexing to handle large mat_mul(_id) (#18678)

2026-01-12 12:32:13 +01:00

mul_mat_vec.comp

vulkan: Switch MUL_MAT_VEC to 4 K per iteration for F16/32 (#22887)

2026-05-27 17:19:23 +02:00

mul_mat_vecq_funcs.glsl

vulkan: Block-load Q3_K/Q6_K block data and subtract on 32b ints (#23056)

2026-06-01 11:46:48 +02:00

mul_mat_vecq.comp

vulkan: opt mul_mat_vecq for mi50 (#22933)

2026-06-26 13:49:24 +02:00

mul_mm_cm2.comp

vulkan: make mul_mm ALIGNED a spec constant (#24689)

2026-06-23 14:26:17 +02:00

mul_mm_funcs.glsl

vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338)

2026-07-13 08:44:17 -05:00

mul_mm_id_funcs.glsl

vulkan: fix data race in mul_mat_id shader (#19790)

2026-02-24 07:43:12 +01:00

mul_mm.comp

vulkan: make mul_mm ALIGNED a spec constant (#24689)

2026-06-23 14:26:17 +02:00

mul_mmq_funcs.glsl

vulkan: unify type macros to use Vx instead of _VECx (#21605)

2026-04-09 07:31:51 +02:00

mul_mmq_shmem_types.glsl

vulkan: Support asymmetric FA in scalar/mmq/coopmat1 paths (#22589)

2026-05-11 12:49:03 +02:00

mul_mmq.comp

vulkan: fix MMQ shader push constants and multi-dispatch (#19732)

2026-02-19 14:59:16 +01:00

mul.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

multi_add.comp

vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (#21572)

2026-04-14 15:17:45 +02:00

norm.comp

vulkan: support all backend tests for SQR/SQRT/SIN/COS/CLAMP/LEAKY_RELU/NORM (#24582)

2026-06-23 09:48:24 -05:00

opt_step_adamw.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

opt_step_sgd.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

pad.comp

ggml : add circular tiling support to pad, for Vulkan, CUDA, and CPU (used for making seamless textures) (#16985)

2025-12-06 15:07:02 +01:00

pool2d.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

quantize_q8_1.comp

vulkan: handle quantize_q8_1 overflowing the max workgroup count (#18515)

2026-01-05 11:30:14 +01:00

reglu.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

repeat_back.comp

vulkan: support non-contig unary/glu ops (#24215)

2026-06-13 08:44:15 -05:00

repeat.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

rms_norm_back.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

rms_norm_partials.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

rms_norm.comp

vulkan: fix non-contig rope (#19299)

2026-02-05 08:38:59 +01:00

roll.comp

vulkan: support non-contig unary/glu ops (#24215)

2026-06-13 08:44:15 -05:00

rope_funcs.glsl

vulkan: Support unaligned tensors for ROPE (#22637)

2026-05-17 11:30:16 +02:00

rope_head.glsl

vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (#21572)

2026-04-14 15:17:45 +02:00

rope_multi.comp

vulkan: fix non-contig rope (#19299)

2026-02-05 08:38:59 +01:00

rope_neox.comp

vulkan: fix non-contig rope (#19299)

2026-02-05 08:38:59 +01:00

rope_norm.comp

vulkan: fix non-contig rope (#19299)

2026-02-05 08:38:59 +01:00

rope_params.glsl

vulkan: Support unaligned tensors for ROPE (#22637)

2026-05-17 11:30:16 +02:00

rope_vision.comp

vulkan: fix non-contig rope (#19299)

2026-02-05 08:38:59 +01:00

scale.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

silu_back.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

snake.comp

vulkan: fuse snake activation (mul, sin, sqr, mul, add) (#22855)

2026-05-21 19:39:42 +02:00

soft_max_back.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

soft_max_large1.comp

vulkan: Multi-pass softmax for large number of cols (#17892)

2025-12-13 10:04:29 +01:00

soft_max_large2.comp

vulkan: Multi-pass softmax for large number of cols (#17892)

2025-12-13 10:04:29 +01:00

soft_max_large3.comp

vulkan: Multi-pass softmax for large number of cols (#17892)

2025-12-13 10:04:29 +01:00

soft_max_large_common.glsl

vulkan: Multi-pass softmax for large number of cols (#17892)

2025-12-13 10:04:29 +01:00

soft_max.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

solve_tri.comp

vulkan: support solve_tri with larger N/K values (#17781)

2025-12-06 08:56:45 +01:00

ssm_conv.comp

vulkan: fuse SSM_CONV + BIAS + SILU (#22653)

2026-05-17 10:25:50 +02:00

ssm_scan.comp

vulkan: optimize ssm_scan (#18630)

2026-01-08 15:16:54 +01:00

sub.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

sum_rows.comp

vulkan: Implement GGML_OP_CUMSUM (#17479)

2025-11-26 07:08:10 +01:00

sum_rows.glsl

vulkan: Implement GGML_OP_CUMSUM (#17479)

2025-11-26 07:08:10 +01:00

swiglu_oai.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

swiglu.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

timestep_embedding.comp

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

topk_argsort.comp

vulkan: Reduce temporary memory usage for TOP_K (#17623)

2025-12-02 19:22:04 +01:00

topk_moe.comp

vulkan: fix topk_moe_sigmoid_norm_bias failures in GLM-4.6 (#18582)

2026-01-05 11:51:39 +01:00

topk_nary_search.comp

vulkan: fix top_k bug when there are ties in the input (#17659)

2025-12-05 22:03:19 +01:00

tri.comp

vulkan: support non-contig unary/glu ops (#24215)

2026-06-13 08:44:15 -05:00

types.glsl

vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338)

2026-07-13 08:44:17 -05:00

unary.comp

vulkan: fix step operator for 0 input (#25036)

2026-06-27 10:57:31 +02:00

upscale.comp

vulkan: Support UPSCALE w/antialias (#18327)

2025-12-26 17:00:57 +01:00

utils.glsl

vulkan : incremental shader builds (#16341)

2025-10-04 11:42:56 +02:00

vulkan-shaders-gen.cpp

vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 (#25338)

2026-07-13 08:44:17 -05:00

wkv6.comp

rwkv6: add wkv6 support for Vulkan backend (#10829)

2024-12-16 22:00:46 +01:00

wkv7.comp

llama: Add support for RWKV v7 architecture (#12412)

2025-03-18 07:27:50 +08:00