agray3 042e95d92f Vectorize load instructions in dmmv f16 CUDA kernel (llama/9816) ...

* Vectorize load instructions in dmmv f16 CUDA kernel

Replaces scalar with vector load instructions, which substantially
improves performance on NVIDIA HBM GPUs, e.g. gives a 1.27X overall
speedup for Meta-Llama-3-8B-Instruct-F16 BS1 inference evaluation on
H100 SXM 80GB HBM3. On GDDR GPUs, there is a slight (1.01X) speedup.

* addressed comment

* Update ggml/src/ggml-cuda/dmmv.cu

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

---------

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

2024-11-01 10:19:05 +02:00

..

template-instances

CUDA: MMQ code deduplication + iquant support (llama/8495)

2024-08-08 22:48:46 +03:00

vendors

musa: enable building fat binaries, enable unified memory, and disable Flash Attention on QY1 (MTT S80) (llama/9526)

2024-09-24 19:45:08 +03:00

acc.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

acc.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

arange.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

arange.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

argmax.cu

ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

2024-10-05 15:23:51 +03:00

argmax.cuh

ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

2024-10-05 15:23:51 +03:00

argsort.cu

ggml : reduce hash table reset cost (llama/8698)

2024-08-08 22:48:46 +03:00

argsort.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

binbcast.cu

ggml/examples: add backend support for numerical optimization (ggml/949)

2024-09-24 19:45:08 +03:00

binbcast.cuh

ggml/examples: add backend support for numerical optimization (ggml/949)

2024-09-24 19:45:08 +03:00

clamp.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

clamp.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

common.cuh

ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

2024-10-05 15:23:51 +03:00

concat.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

concat.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

conv-transpose-1d.cu

feat: cuda implementation for ggml_conv_transpose_1d (ggml/854)

2024-07-08 14:53:55 +03:00

conv-transpose-1d.cuh

feat: cuda implementation for ggml_conv_transpose_1d (ggml/854)

2024-07-08 14:53:55 +03:00

convert.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

convert.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

count-equal.cu

ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

2024-10-05 15:23:51 +03:00

count-equal.cuh

ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

2024-10-05 15:23:51 +03:00

cpy.cu

cuda: add q8_0->f32 cpy operation (llama/9571)

2024-09-24 19:45:08 +03:00

cpy.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

cross-entropy-loss.cu

ggml/examples: add backend support for numerical optimization (ggml/949)

2024-09-24 19:45:08 +03:00

cross-entropy-loss.cuh

ggml/examples: add backend support for numerical optimization (ggml/949)

2024-09-24 19:45:08 +03:00

dequantize.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

diagmask.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

diagmask.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

dmmv.cu

Vectorize load instructions in dmmv f16 CUDA kernel (llama/9816)

2024-11-01 10:19:05 +02:00

dmmv.cuh

cuda : fix dmmv cols requirement to 2*GGML_CUDA_DMMV_X (llama/8800)

2024-08-08 22:48:46 +03:00

fattn-common.cuh

CPU/CUDA: Gemma 2 FlashAttention support (llama/8542)

2024-08-28 13:22:20 +03:00

fattn-tile-f16.cu

ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

2024-10-05 15:23:51 +03:00

fattn-tile-f16.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

fattn-tile-f32.cu

musa: enable building fat binaries, enable unified memory, and disable Flash Attention on QY1 (MTT S80) (llama/9526)

2024-09-24 19:45:08 +03:00

fattn-tile-f32.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

fattn-vec-f16.cuh

ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

2024-10-05 15:23:51 +03:00

fattn-vec-f32.cuh

CPU/CUDA: Gemma 2 FlashAttention support (llama/8542)

2024-08-28 13:22:20 +03:00

fattn-wmma-f16.cuh

CPU/CUDA: Gemma 2 FlashAttention support (llama/8542)

2024-08-28 13:22:20 +03:00

fattn.cu

CUDA: enable Gemma FA for HIP/Pascal (llama/9581)

2024-09-24 19:45:08 +03:00

fattn.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

getrows.cu

ggml : reduce hash table reset cost (llama/8698)

2024-08-08 22:48:46 +03:00

getrows.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

im2col.cu

CUDA: remove bad assert (ggml/972)

2024-10-03 12:22:17 +03:00

im2col.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

mma.cuh

CUDA: optimize and refactor MMQ (llama/8416)

2024-08-08 22:48:46 +03:00

mmq.cu

CUDA: fix --split-mode row race condition (llama/9413)

2024-09-24 19:45:08 +03:00

mmq.cuh

CUDA: fix --split-mode row race condition (llama/9413)

2024-09-24 19:45:08 +03:00

mmvq.cu

ggml : reduce hash table reset cost (llama/8698)

2024-08-08 22:48:46 +03:00

mmvq.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

norm.cu

ggml : add epsilon as a parameter for group_norm (llama/8818)

2024-08-08 22:48:46 +03:00

norm.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

opt-step-adamw.cu

ggml/examples: add backend support for numerical optimization (ggml/949)

2024-09-24 19:45:08 +03:00

opt-step-adamw.cuh

ggml/examples: add backend support for numerical optimization (ggml/949)

2024-09-24 19:45:08 +03:00

out-prod.cu

ggml : fix builds (llama/0)

2024-09-24 19:45:08 +03:00

out-prod.cuh

ggml/examples: add backend support for numerical optimization (ggml/949)

2024-09-24 19:45:08 +03:00

pad.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

pad.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

pool2d.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

pool2d.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

quantize.cu

ggml : reduce hash table reset cost (llama/8698)

2024-08-08 22:48:46 +03:00

quantize.cuh

CUDA: optimize and refactor MMQ (llama/8416)

2024-08-08 22:48:46 +03:00

rope.cu

ggml : move rope type enum to ggml.h (llama/8949)

2024-08-28 13:22:20 +03:00

rope.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

rwkv-wkv.cu

RWKV v6: RWKV_WKV op CUDA implementation (llama/9454)

2024-09-24 19:45:08 +03:00

rwkv-wkv.cuh

RWKV v6: RWKV_WKV op CUDA implementation (llama/9454)

2024-09-24 19:45:08 +03:00

scale.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

scale.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

softmax.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

softmax.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

sum.cu

CUDA: fix sum.cu compilation for CUDA < 11.7 (llama/9562)

2024-09-24 19:45:08 +03:00

sum.cuh

tests: add gradient tests for all backends (ggml/932)

2024-09-24 19:45:08 +03:00

sumrows.cu

feat: ref. cross entropy, add CUDA, fix grad test (ggml/929)

2024-08-28 13:22:20 +03:00

sumrows.cuh

feat: ref. cross entropy, add CUDA, fix grad test (ggml/929)

2024-08-28 13:22:20 +03:00

tsembd.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

tsembd.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

unary.cu

RWKV v6: RWKV_WKV op CUDA implementation (llama/9454)

2024-09-24 19:45:08 +03:00

unary.cuh

RWKV v6: RWKV_WKV op CUDA implementation (llama/9454)

2024-09-24 19:45:08 +03:00

upscale.cu

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

upscale.cuh

whisper : reorganize source code + improve CMake (#2256)

2024-06-26 19:34:09 +03:00

vecdotq.cuh

CUDA: MMQ code deduplication + iquant support (llama/8495)

2024-08-08 22:48:46 +03:00