yum-archive/TaSTT-Whisper

High-performance GPGPU inference of OpenAI's Whisper automatic speech recognition (ASR) model

git clone https://git.yummers.dev/yum-archive/TaSTT-Whisper

KonstantinPerformance improvement, `softMax` shader27dfc34

master
branchesmasterall refs →
1.0 KiB52 linesraw
1// This header is generated by a tool
2#pragma once
3#include <stdint.h>
4
5namespace DirectCompute
6{
7	enum struct eComputeShader: uint16_t
8	{
9		add = 0,
10		addInPlace = 1,
11		addRepeat = 2,
12		addRepeatEx = 3,
13		addRepeatGelu = 4,
14		addRepeatScale = 5,
15		addRows = 6,
16		convolutionMain = 7,
17		convolutionMain2 = 8,
18		convolutionMain2Fixed = 9,
19		convolutionPrep1 = 10,
20		convolutionPrep2 = 11,
21		copyConvert = 12,
22		copyTranspose = 13,
23		diagMaskInf = 14,
24		flashAttention = 15,
25		flashAttentionCompat1 = 16,
26		flashAttentionCompat2 = 17,
27		flashAttentionCompat3 = 18,
28		fmaRepeat1 = 19,
29		fmaRepeat2 = 20,
30		matReshapePanels = 21,
31		mulMatByRow = 22,
32		mulMatByRowTiled = 23,
33		mulMatByRowTiledEx = 24,
34		mulMatByScalar = 25,
35		mulMatDotMain = 26,
36		mulMatDotReshape = 27,
37		mulMatMadMain = 28,
38		mulMatTiled = 29,
39		mulMatTiledEx = 30,
40		norm = 31,
41		normCompat = 32,
42		normFixed = 33,
43		scaleInPlace = 34,
44		softMax = 35,
45		softMaxCompat = 36,
46		softMaxFixed = 37,
47		softMaxLong = 38,
48		zeroMemory = 39,
49	};
50
51	const char* computeShaderName( eComputeShader cs );
52}