/* * Copyright (C) 2020 Linux Studio Plugins Project * (C) 2020 Vladimir Sadovnikov * * This file is part of lsp-dsp-lib * Created on: 31 мар. 2020 г. * * lsp-dsp-lib is free software: you can redistribute it and/or modify * it under the terms of the GNU Lesser General Public License as published by * the Free Software Foundation, either version 3 of the License, or * any later version. * * lsp-dsp-lib is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public License * along with lsp-dsp-lib. If not, see . */ #include #include #include #include #include #define MIN_RANK 8 #define MAX_RANK 16 namespace lsp { namespace generic { void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); void add2(float *dst, const float *src, size_t count); void fastconv_parse(float *dst, const float *src, size_t rank); void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); } IF_ARCH_X86( namespace sse { void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); void add2(float *dst, const float *src, size_t count); void fastconv_parse(float *dst, const float *src, size_t rank); void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); } namespace avx { void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); void add2(float *dst, const float *src, size_t count); void fastconv_parse(float *dst, const float *src, size_t rank); void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); void direct_fft_fma3(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void reverse_fft_fma3(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void complex_mul3_fma3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); void fastconv_parse_fma3(float *dst, const float *src, size_t rank); void fastconv_parse_apply_fma3(float *dst, float *tmp, const float *c, const float *src, size_t rank); } ) IF_ARCH_ARM( namespace neon_d32 { void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); void add2(float *dst, const float *src, size_t count); void fastconv_parse(float *dst, const float *src, size_t rank); void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); } ) IF_ARCH_AARCH64( namespace asimd { void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); void add2(float *dst, const float *src, size_t count); void fastconv_parse(float *dst, const float *src, size_t rank); void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); } ) typedef void (* direct_fft_t)(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); typedef void (* reverse_fft_t)(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); typedef void (* complex_mul3_t)(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); typedef void (* add2_t)(float *dst, const float *src, size_t count); typedef void (* fastconv_parse_t)(float *dst, const float *src, size_t rank); typedef void (* fastconv_parse_apply_t)(float *dst, float *tmp, const float *c, const float *src, size_t rank); } //----------------------------------------------------------------------------- // Performance test for complex multiplication PTEST_BEGIN("dsp.fft", fastconv, 10, 1000) void call( const char *label, float *out, float *tmp, float *tmp2, float *conv, const float *in, const float *cv, size_t rank, direct_fft_t direct, complex_mul3_t cmul, reverse_fft_t reverse, add2_t add2 ) { if (!(PTEST_SUPPORTED(direct) && (PTEST_SUPPORTED(cmul)) && (PTEST_SUPPORTED(reverse)) && (PTEST_SUPPORTED(add2)))) return; char buf[80]; snprintf(buf, sizeof(buf), "%s x %d", label, int(1 << rank)); printf("Testing %s samples (rank = %d)...\n", buf, int(rank)); size_t bin_size = 1 << rank; // Prepare data dsp::fill_zero(out, bin_size); direct(conv, &conv[bin_size], cv, &cv[bin_size], rank); // Run test PTEST_LOOP(buf, direct(tmp, &tmp[bin_size], in, &in[bin_size], rank); cmul(tmp, &tmp[bin_size], tmp, &tmp[bin_size], conv, &conv[bin_size], bin_size); reverse(tmp2, &tmp2[bin_size], tmp, &tmp[bin_size], rank); add2(out, tmp2, bin_size); ) } void call( const char *label, float *out, float *tmp, float *conv, const float *in, const float *cv, size_t rank, fastconv_parse_t parse, fastconv_parse_apply_t apply ) { if (!(PTEST_SUPPORTED(parse) && (PTEST_SUPPORTED(apply)))) return; char buf[80]; snprintf(buf, sizeof(buf), "%s x %d", label, int(1 << rank)); printf("Testing %s samples (rank = %d)...\n", buf, int(rank)); // Prepare data dsp::fill_zero(out, 1 << rank); parse(conv, cv, rank); PTEST_LOOP(buf, apply(out, tmp, conv, in, rank); ); } PTEST_MAIN { size_t fft_size = 1 << MAX_RANK; size_t alloc = fft_size * 2 * 2 // in + cv size + fft_size * 2 // conv size + fft_size * 2 // tmp size + fft_size * 2 // tmp2 size + fft_size; // out size uint8_t *data = NULL; float *in = alloc_aligned(data, alloc, 64); dsp::fill_zero(in, alloc); float *cv = &in[fft_size * 2]; float *conv = &cv[fft_size * 2]; float *tmp = &conv[fft_size * 2]; float *tmp2 = &tmp[fft_size * 2]; float *out = &tmp2[fft_size * 2]; for (size_t i=0; i < (fft_size / 2); ++i) { in[i] = randf(0.0f, 1.0f); cv[i] = randf(0.0f, 1.0f); } for (size_t rank=MIN_RANK; rank <= MAX_RANK; ++rank) { call("generic::fft", out, tmp, tmp2, conv, in, cv, rank, generic::direct_fft, generic::complex_mul3, generic::reverse_fft, generic::add2); call("generic::fastconv_fft", out, tmp, conv, in, cv, rank, generic::fastconv_parse, generic::fastconv_parse_apply); IF_ARCH_X86( call("sse::fft", out, tmp, tmp2, conv, in, cv, rank, sse::direct_fft, sse::complex_mul3, sse::reverse_fft, sse::add2); call("avx::fft", out, tmp, tmp2, conv, in, cv, rank, avx::direct_fft, avx::complex_mul3, avx::reverse_fft, avx::add2); call("avx::fft_fma3", out, tmp, tmp2, conv, in, cv, rank, avx::direct_fft_fma3, avx::complex_mul3_fma3, avx::reverse_fft_fma3, avx::add2); call("sse::fastconv_fft", out, tmp, conv, in, cv, rank, sse::fastconv_parse, sse::fastconv_parse_apply); call("avx::fastconv_fft", out, tmp, conv, in, cv, rank, avx::fastconv_parse, avx::fastconv_parse_apply); call("avx::fastconv_fft_fma3", out, tmp, conv, in, cv, rank, avx::fastconv_parse_fma3, avx::fastconv_parse_apply_fma3); ) IF_ARCH_ARM( call("neon_d32::fft", out, tmp, tmp2, conv, in, cv, rank, neon_d32::direct_fft, neon_d32::complex_mul3, neon_d32::reverse_fft, neon_d32::add2); call("neon_d32::fastconv_fft", out, tmp, conv, in, cv, rank, neon_d32::fastconv_parse, neon_d32::fastconv_parse_apply); ) IF_ARCH_AARCH64( call("asimd::fft", out, tmp, tmp2, conv, in, cv, rank, asimd::direct_fft, asimd::complex_mul3, asimd::reverse_fft, asimd::add2); call("asimd::fastconv_fft", out, tmp, conv, in, cv, rank, asimd::fastconv_parse, asimd::fastconv_parse_apply); ) PTEST_SEPARATOR; } free_aligned(data); } PTEST_END