1 /* Copyright (c) 2011-2019 Mozilla 2 2023 Amazon */ 3 /* 4 Redistribution and use in source and binary forms, with or without 5 modification, are permitted provided that the following conditions 6 are met: 7 8 - Redistributions of source code must retain the above copyright 9 notice, this list of conditions and the following disclaimer. 10 11 - Redistributions in binary form must reproduce the above copyright 12 notice, this list of conditions and the following disclaimer in the 13 documentation and/or other materials provided with the distribution. 14 15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 16 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 17 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 18 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR 19 CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 20 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 22 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 23 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 24 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 25 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #ifndef DNN_X86_H 29 #define DNN_X86_H 30 31 #include "cpu_support.h" 32 #include "opus_types.h" 33 34 #if defined(OPUS_X86_MAY_HAVE_SSE2) 35 void compute_linear_sse2(const LinearLayer *linear, float *out, const float *in); 36 void compute_activation_sse2(float *output, const float *input, int N, int activation); 37 void compute_conv2d_sse2(const Conv2dLayer *conv, float *out, float *mem, const float *in, int height, int hstride, int activation); 38 #endif 39 40 #if defined(OPUS_X86_MAY_HAVE_SSE4_1) 41 void compute_linear_sse4_1(const LinearLayer *linear, float *out, const float *in); 42 void compute_activation_sse4_1(float *output, const float *input, int N, int activation); 43 void compute_conv2d_sse4_1(const Conv2dLayer *conv, float *out, float *mem, const float *in, int height, int hstride, int activation); 44 #endif 45 46 #if defined(OPUS_X86_MAY_HAVE_AVX2) 47 void compute_linear_avx2(const LinearLayer *linear, float *out, const float *in); 48 void compute_activation_avx2(float *output, const float *input, int N, int activation); 49 void compute_conv2d_avx2(const Conv2dLayer *conv, float *out, float *mem, const float *in, int height, int hstride, int activation); 50 #endif 51 52 53 #if defined(OPUS_X86_PRESUME_AVX2) 54 55 #define OVERRIDE_COMPUTE_LINEAR 56 #define compute_linear(linear, out, in, arch) ((void)(arch),compute_linear_avx2(linear, out, in)) 57 #define OVERRIDE_COMPUTE_ACTIVATION 58 #define compute_activation(output, input, N, activation, arch) ((void)(arch),compute_activation_avx2(output, input, N, activation)) 59 #define OVERRIDE_COMPUTE_CONV2D 60 #define compute_conv2d(conv, out, mem, in, height, hstride, activation, arch) ((void)(arch),compute_conv2d_avx2(conv, out, mem, in, height, hstride, activation)) 61 62 #elif defined(OPUS_X86_PRESUME_SSE4_1) && !defined(OPUS_X86_MAY_HAVE_AVX2) 63 64 #define OVERRIDE_COMPUTE_LINEAR 65 #define compute_linear(linear, out, in, arch) ((void)(arch),compute_linear_sse4_1(linear, out, in)) 66 #define OVERRIDE_COMPUTE_ACTIVATION 67 #define compute_activation(output, input, N, activation, arch) ((void)(arch),compute_activation_sse4_1(output, input, N, activation)) 68 #define OVERRIDE_COMPUTE_CONV2D 69 #define compute_conv2d(conv, out, mem, in, height, hstride, activation, arch) ((void)(arch),compute_conv2d_sse4_1(conv, out, mem, in, height, hstride, activation)) 70 71 #elif defined(OPUS_X86_PRESUME_SSE2) && !defined(OPUS_X86_MAY_HAVE_AVX2) && !defined(OPUS_X86_MAY_HAVE_SSE4_1) 72 73 #define OVERRIDE_COMPUTE_LINEAR 74 #define compute_linear(linear, out, in, arch) ((void)(arch),compute_linear_sse2(linear, out, in)) 75 #define OVERRIDE_COMPUTE_ACTIVATION 76 #define compute_activation(output, input, N, activation, arch) ((void)(arch),compute_activation_sse2(output, input, N, activation)) 77 #define OVERRIDE_COMPUTE_CONV2D 78 #define compute_conv2d(conv, out, mem, in, height, hstride, activation, arch) ((void)(arch),compute_conv2d_sse2(conv, out, mem, in, height, hstride, activation)) 79 80 #elif defined(OPUS_HAVE_RTCD) && (defined(OPUS_X86_MAY_HAVE_AVX2) || defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) 81 82 extern void (*const DNN_COMPUTE_LINEAR_IMPL[OPUS_ARCHMASK + 1])( 83 const LinearLayer *linear, 84 float *out, 85 const float *in 86 ); 87 #define OVERRIDE_COMPUTE_LINEAR 88 #define compute_linear(linear, out, in, arch) \ 89 ((*DNN_COMPUTE_LINEAR_IMPL[(arch) & OPUS_ARCHMASK])(linear, out, in)) 90 91 92 extern void (*const DNN_COMPUTE_ACTIVATION_IMPL[OPUS_ARCHMASK + 1])( 93 float *output, 94 const float *input, 95 int N, 96 int activation 97 ); 98 #define OVERRIDE_COMPUTE_ACTIVATION 99 #define compute_activation(output, input, N, activation, arch) \ 100 ((*DNN_COMPUTE_ACTIVATION_IMPL[(arch) & OPUS_ARCHMASK])(output, input, N, activation)) 101 102 103 extern void (*const DNN_COMPUTE_CONV2D_IMPL[OPUS_ARCHMASK + 1])( 104 const Conv2dLayer *conv, 105 float *out, 106 float *mem, 107 const float *in, 108 int height, 109 int hstride, 110 int activation 111 ); 112 #define OVERRIDE_COMPUTE_CONV2D 113 #define compute_conv2d(conv, out, mem, in, height, hstride, activation, arch) \ 114 ((*DNN_COMPUTE_CONV2D_IMPL[(arch) & OPUS_ARCHMASK])(conv, out, mem, in, height, hstride, activation)) 115 116 117 #endif 118 119 120 121 #endif /* DNN_X86_H */ 122