/* * 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 . */ #ifndef PRIVATE_DSP_ARCH_X86_SSE2_PMATH_EXP_H_ #define PRIVATE_DSP_ARCH_X86_SSE2_PMATH_EXP_H_ #ifndef PRIVATE_DSP_ARCH_X86_SSE2_IMPL #error "This header should not be included directly" #endif /* PRIVATE_DSP_ARCH_X86_SSE2_IMPL */ namespace lsp { namespace sse2 { /* Code to generate constants: float fact(int n) { return (n > 1) ? n*fact(n-1) : 1; } uint32_t to_hex(float v) { union { uint32_t i; float f; } r; r.f = v; return r.i; } int main() { printf("LSP_DSP_VEC4(0x7fffffff), // sign\n"); printf("LSP_DSP_VEC4(0x%08x), // ln(2)\n", to_hex(M_LN2)); float f7 = fact(7); for (int i=6; i>=1; --i) { float C = f7 / fact(i); printf("LSP_DSP_VEC4(0x%08x), // C%d = %d!/%d! = %.16f \n", to_hex(C), i-1, 7, i, C); } printf("LSP_DSP_VEC4(0x%08x), // 1/7! = %.16f\n", to_hex(1.0f/f7), 1.0f/f7); printf("LSP_DSP_VEC4(0x%08x), // 1.0\n", to_hex(1.0f)); } */ IF_ARCH_X86( static const uint32_t EXP2_CONST[] __lsp_aligned16 = { LSP_DSP_VEC4(0x7fffffff), // sign LSP_DSP_VEC4(0x3f317218), // ln(2) LSP_DSP_VEC4(0x40e00000), // C5 = 7!/6! = 7.0000000000000000 LSP_DSP_VEC4(0x42280000), // C4 = 7!/5! = 42.0000000000000000 LSP_DSP_VEC4(0x43520000), // C3 = 7!/4! = 210.0000000000000000 LSP_DSP_VEC4(0x44520000), // C2 = 7!/3! = 840.0000000000000000 LSP_DSP_VEC4(0x451d8000), // C1 = 7!/2! = 2520.0000000000000000 LSP_DSP_VEC4(0x459d8000), // C0 = 7!/1! = 5040.0000000000000000 LSP_DSP_VEC4(0x39500d01), // 1/7! = 0.0001984127011383 LSP_DSP_VEC4(0x3f800000), // 1.0 LSP_DSP_VEC4(0x7f), // 0x7f }; static const float EXP_LOG2E[] __lsp_aligned16 = { LSP_DSP_VEC4(M_LOG2E) }; ) #define POW2_CORE_X8 \ /* xmm0 = x, xmm4 = y */ \ __ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = x */ \ __ASM_EMIT("movaps %%xmm4, %%xmm6") \ __ASM_EMIT("andps 0x00 + %[E2C], %%xmm2") /* xmm2 = XP = fabs(x) */ \ __ASM_EMIT("andps 0x00 + %[E2C], %%xmm6") \ __ASM_EMIT("psrad $31, %%xmm0") /* xmm0 = [ x < 0 ] */ \ __ASM_EMIT("psrad $31, %%xmm4") \ __ASM_EMIT("cvttps2dq %%xmm2, %%xmm3") /* xmm3 = R = int(XP) */ \ __ASM_EMIT("cvttps2dq %%xmm6, %%xmm7") \ __ASM_EMIT("movaps %%xmm3, %%xmm1") /* xmm1 = R */ \ __ASM_EMIT("movaps %%xmm7, %%xmm5") \ __ASM_EMIT("cvtdq2ps %%xmm3, %%xmm3") /* xmm3 = float(R) */ \ __ASM_EMIT("cvtdq2ps %%xmm7, %%xmm7") \ __ASM_EMIT("paddd 0xa0 + %[E2C], %%xmm1") /* xmm1 = R + 127 */ \ __ASM_EMIT("paddd 0xa0 + %[E2C], %%xmm5") \ __ASM_EMIT("subps %%xmm3, %%xmm2") /* xmm2 = XP - float(R) */ \ __ASM_EMIT("subps %%xmm7, %%xmm6") \ __ASM_EMIT("pslld $23, %%xmm1") /* xmm1 = 1 << (R+127) */ \ __ASM_EMIT("pslld $23, %%xmm5") \ __ASM_EMIT("mulps 0x10 + %[E2C], %%xmm2") /* xmm2 = X = ln(2) * (XP - float(R)) */ \ __ASM_EMIT("mulps 0x10 + %[E2C], %%xmm6") \ __ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = X */ \ __ASM_EMIT("movaps %%xmm6, %%xmm7") \ /* xmm0 = [ x < 0 ], xmm1 = 1 << R, xmm2 = X, xmm3 = X */ \ __ASM_EMIT("addps 0x20 + %[E2C], %%xmm2") /* xmm2 = C5+X */ \ __ASM_EMIT("addps 0x20 + %[E2C], %%xmm6") \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C5+X) */ \ __ASM_EMIT("mulps %%xmm7, %%xmm6") \ __ASM_EMIT("addps 0x30 + %[E2C], %%xmm2") /* xmm2 = C4+(X*(C5+X)) */ \ __ASM_EMIT("addps 0x30 + %[E2C], %%xmm6") \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C4+(X*(C5+X))) */ \ __ASM_EMIT("mulps %%xmm7, %%xmm6") \ __ASM_EMIT("addps 0x40 + %[E2C], %%xmm2") /* xmm2 = C3+X*(C4+(X*(C5+X))) */ \ __ASM_EMIT("addps 0x40 + %[E2C], %%xmm6") \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C3+X*(C4+(X*(C5+X)))) */ \ __ASM_EMIT("mulps %%xmm7, %%xmm6") \ __ASM_EMIT("addps 0x50 + %[E2C], %%xmm2") /* xmm2 = C2+X*(C3+X*(C4+(X*(C5+X)))) */ \ __ASM_EMIT("addps 0x50 + %[E2C], %%xmm6") \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C2+X*(C3+X*(C4+(X*(C5+X))))) */ \ __ASM_EMIT("mulps %%xmm7, %%xmm6") \ __ASM_EMIT("addps 0x60 + %[E2C], %%xmm2") /* xmm2 = C1+X*(C2+(X*C3+X*(C4+(X*(C5+X))))) */ \ __ASM_EMIT("addps 0x60 + %[E2C], %%xmm6") \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X)))))) */ \ __ASM_EMIT("mulps %%xmm7, %%xmm6") \ __ASM_EMIT("addps 0x70 + %[E2C], %%xmm2") /* xmm2 = C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X)))))) */ \ __ASM_EMIT("addps 0x70 + %[E2C], %%xmm6") \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X))))))) */ \ __ASM_EMIT("mulps %%xmm7, %%xmm6") \ __ASM_EMIT("mulps 0x80 + %[E2C], %%xmm2") /* xmm2 = 1/7! * X*(C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X))))))) */ \ __ASM_EMIT("mulps 0x80 + %[E2C], %%xmm6") \ __ASM_EMIT("movaps 0x90 + %[E2C], %%xmm3") /* xmm3 = 1 */ \ __ASM_EMIT("movaps %%xmm3, %%xmm7") \ __ASM_EMIT("addps %%xmm3, %%xmm2") /* xmm2 = 1 + 1/7! * X*(C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X))))))) */ \ __ASM_EMIT("addps %%xmm7, %%xmm6") \ __ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = p = (1 << R) * (1 + 1/7! * X*(C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X)))))))) */ \ __ASM_EMIT("mulps %%xmm5, %%xmm6") \ __ASM_EMIT("divps %%xmm2, %%xmm3") /* xmm3 = 1/p */ \ __ASM_EMIT("divps %%xmm6, %%xmm7") \ /* xmm0 = [ x < 0 ], xmm2 = p, xmm3 = 1/p */ \ __ASM_EMIT("andps %%xmm0, %%xmm3") /* xmm3 = 1/p & [ x < 0] */ \ __ASM_EMIT("andps %%xmm4, %%xmm7") \ __ASM_EMIT("andnps %%xmm2, %%xmm0") /* xmm0 = p & [ x >= 0] */ \ __ASM_EMIT("andnps %%xmm6, %%xmm4") \ __ASM_EMIT("orps %%xmm3, %%xmm0") /* xmm0 = (1/p & [ x < 0]) | (p & [ x >= 0]) */ \ __ASM_EMIT("orps %%xmm7, %%xmm4") #define POW2_CORE_X4 \ /* xmm0 = x */ \ __ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = x */ \ __ASM_EMIT("andps 0x00 + %[E2C], %%xmm2") /* xmm2 = XP = fabs(x) */ \ __ASM_EMIT("psrad $31, %%xmm0") /* xmm0 = [ x < 0 ] */ \ __ASM_EMIT("cvttps2dq %%xmm2, %%xmm3") /* xmm3 = R = int(XP) */ \ __ASM_EMIT("movaps %%xmm3, %%xmm1") /* xmm1 = R */ \ __ASM_EMIT("cvtdq2ps %%xmm3, %%xmm3") /* xmm3 = float(R) */ \ __ASM_EMIT("paddd 0xa0 + %[E2C], %%xmm1") /* xmm1 = R + 127 */ \ __ASM_EMIT("subps %%xmm3, %%xmm2") /* xmm2 = XP - float(R) */ \ __ASM_EMIT("pslld $23, %%xmm1") /* xmm1 = 1 << (R+127) */ \ __ASM_EMIT("mulps 0x10 + %[E2C], %%xmm2") /* xmm2 = X = ln(2) * (XP - float(R)) */ \ __ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = X */ \ /* xmm0 = [ x < 0 ], xmm1 = 1 << R, xmm2 = X, xmm3 = X */ \ __ASM_EMIT("addps 0x20 + %[E2C], %%xmm2") /* xmm2 = C5+X */ \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C5+X) */ \ __ASM_EMIT("addps 0x30 + %[E2C], %%xmm2") /* xmm2 = C4+(X*(C5+X)) */ \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C4+(X*(C5+X))) */ \ __ASM_EMIT("addps 0x40 + %[E2C], %%xmm2") /* xmm2 = C3+X*(C4+(X*(C5+X))) */ \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C3+X*(C4+(X*(C5+X)))) */ \ __ASM_EMIT("addps 0x50 + %[E2C], %%xmm2") /* xmm2 = C2+X*(C3+X*(C4+(X*(C5+X)))) */ \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C2+X*(C3+X*(C4+(X*(C5+X))))) */ \ __ASM_EMIT("addps 0x60 + %[E2C], %%xmm2") /* xmm2 = C1+X*(C2+(X*C3+X*(C4+(X*(C5+X))))) */ \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X)))))) */ \ __ASM_EMIT("addps 0x70 + %[E2C], %%xmm2") /* xmm2 = C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X)))))) */ \ __ASM_EMIT("mulps %%xmm3, %%xmm2") /* xmm2 = X*(C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X))))))) */ \ __ASM_EMIT("mulps 0x80 + %[E2C], %%xmm2") /* xmm2 = 1/7! * X*(C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X))))))) */ \ __ASM_EMIT("movaps 0x90 + %[E2C], %%xmm3") /* xmm3 = 1 */ \ __ASM_EMIT("addps %%xmm3, %%xmm2") /* xmm2 = 1 + 1/7! * X*(C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X))))))) */ \ __ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = p = (1 << R) * (1 + 1/7! * X*(C0+X*(C1+X*(C2+(X*C3+X*(C4+(X*(C5+X)))))))) */ \ __ASM_EMIT("divps %%xmm2, %%xmm3") /* xmm3 = 1/p */ \ /* xmm0 = [ x < 0 ], xmm2 = p, xmm3 = 1/p */ \ __ASM_EMIT("andps %%xmm0, %%xmm3") /* xmm3 = 1/p & [ x < 0] */ \ __ASM_EMIT("andnps %%xmm2, %%xmm0") /* xmm0 = p & [ x >= 0] */ \ __ASM_EMIT("orps %%xmm3, %%xmm0") /* xmm0 = (1/p & [ x < 0]) | (p & [ x >= 0]) */ #define EXP_CORE_X4 \ __ASM_EMIT("mulps %[LOG2E], %%xmm0") \ POW2_CORE_X4 #define EXP_CORE_X8 \ __ASM_EMIT("mulps %[LOG2E], %%xmm0") \ __ASM_EMIT("mulps %[LOG2E], %%xmm4") \ POW2_CORE_X8 void exp1(float *dst, size_t count) { IF_ARCH_X86(float *src); ARCH_X86_ASM( // x8 blocks __ASM_EMIT("sub $8, %[count]") __ASM_EMIT("jb 2f") __ASM_EMIT("1:") __ASM_EMIT("movups 0x00(%[dst]), %%xmm0") __ASM_EMIT("movups 0x10(%[dst]), %%xmm4") EXP_CORE_X8 __ASM_EMIT("movups %%xmm0, 0x00(%[dst])") __ASM_EMIT("movups %%xmm4, 0x10(%[dst])") __ASM_EMIT("add $0x20, %[dst]") __ASM_EMIT("sub $8, %[count]") __ASM_EMIT("jae 1b") __ASM_EMIT("2:") // x4 block __ASM_EMIT("add $4, %[count]") __ASM_EMIT("jl 4f") __ASM_EMIT("movups 0x00(%[dst]), %%xmm0") EXP_CORE_X4 __ASM_EMIT("movups %%xmm0, 0x00(%[dst])") __ASM_EMIT("add $0x10, %[dst]") __ASM_EMIT("sub $4, %[count]") __ASM_EMIT("4:") // Tail: 1x-3x block __ASM_EMIT("add $4, %[count]") __ASM_EMIT("jle 12f") __ASM_EMIT("mov %[dst], %[src]") __ASM_EMIT("test $1, %[count]") __ASM_EMIT("jz 6f") __ASM_EMIT("movss 0x00(%[src]), %%xmm0") __ASM_EMIT("add $4, %[src]") __ASM_EMIT("6:") __ASM_EMIT("test $2, %[count]") __ASM_EMIT("jz 8f") __ASM_EMIT("movhps 0x00(%[src]), %%xmm0") __ASM_EMIT("8:") EXP_CORE_X4 __ASM_EMIT("test $1, %[count]") __ASM_EMIT("jz 10f") __ASM_EMIT("movss %%xmm0, 0x00(%[dst])") __ASM_EMIT("add $4, %[dst]") __ASM_EMIT("10:") __ASM_EMIT("test $2, %[count]") __ASM_EMIT("jz 12f") __ASM_EMIT("movhps %%xmm0, 0x00(%[dst])") __ASM_EMIT("12:") : [dst] "+r" (dst), [src] "=&r" (src), [count] "+r" (count) : [E2C] "o" (EXP2_CONST), [LOG2E] "m" (EXP_LOG2E) : "cc", "memory", "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7" ); } void exp2(float *dst, const float *src, size_t count) { ARCH_X86_ASM( // x8 blocks __ASM_EMIT("sub $8, %[count]") __ASM_EMIT("jb 2f") __ASM_EMIT("1:") __ASM_EMIT("movups 0x00(%[src]), %%xmm0") __ASM_EMIT("movups 0x10(%[src]), %%xmm4") EXP_CORE_X8 __ASM_EMIT("movups %%xmm0, 0x00(%[dst])") __ASM_EMIT("movups %%xmm4, 0x10(%[dst])") __ASM_EMIT("add $0x20, %[src]") __ASM_EMIT("add $0x20, %[dst]") __ASM_EMIT("sub $8, %[count]") __ASM_EMIT("jae 1b") __ASM_EMIT("2:") // x4 block __ASM_EMIT("add $4, %[count]") __ASM_EMIT("jl 4f") __ASM_EMIT("movups 0x00(%[src]), %%xmm0") EXP_CORE_X4 __ASM_EMIT("movups %%xmm0, 0x00(%[dst])") __ASM_EMIT("add $0x10, %[src]") __ASM_EMIT("add $0x10, %[dst]") __ASM_EMIT("sub $4, %[count]") __ASM_EMIT("4:") // Tail: 1x-3x block __ASM_EMIT("add $4, %[count]") __ASM_EMIT("jle 12f") __ASM_EMIT("test $1, %[count]") __ASM_EMIT("jz 6f") __ASM_EMIT("movss 0x00(%[src]), %%xmm0") __ASM_EMIT("add $4, %[src]") __ASM_EMIT("6:") __ASM_EMIT("test $2, %[count]") __ASM_EMIT("jz 8f") __ASM_EMIT("movhps 0x00(%[src]), %%xmm0") __ASM_EMIT("8:") EXP_CORE_X4 __ASM_EMIT("test $1, %[count]") __ASM_EMIT("jz 10f") __ASM_EMIT("movss %%xmm0, 0x00(%[dst])") __ASM_EMIT("add $4, %[dst]") __ASM_EMIT("10:") __ASM_EMIT("test $2, %[count]") __ASM_EMIT("jz 12f") __ASM_EMIT("movhps %%xmm0, 0x00(%[dst])") __ASM_EMIT("12:") : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) : [E2C] "o" (EXP2_CONST), [LOG2E] "m" (EXP_LOG2E) : "cc", "memory", "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7" ); } /* const float *E2C = reinterpret_cast(EXP2_CONST); float XP = fabs(x); int R = XP; float X = (XP - float(R)) * E2C[1]; float p = E2C[9] + E2C[8]*X*( E2C[7] + X*( E2C[6] + X*( E2C[5] + X*( E2C[4] + X*( E2C[3] + X*( E2C[2] + X ))))) ); p *= 1 << R; return (x < 0.0f) ? 1.0f / p : p; */ } } #endif /* PRIVATE_DSP_ARCH_X86_SSE2_PMATH_EXP_H_ */