/*
* Copyright (C) 2023 Linux Studio Plugins Project
* (C) 2023 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_AVX_COMPLEX_H_
#define PRIVATE_DSP_ARCH_X86_AVX_COMPLEX_H_
#ifndef PRIVATE_DSP_ARCH_X86_AVX_IMPL
#error "This header should not be included directly"
#endif /* PRIVATE_DSP_ARCH_X86_AVX_IMPL */
namespace lsp
{
namespace avx
{
#define FMA_OFF(a, b) a
#define FMA_ON(a, b) b
#define FMA_PASS(x) __ASM_EMIT(x)
#define FMA_SKIP(x)
IF_ARCH_X86(
static uint32_t complex_div_const[] __lsp_aligned32 =
{
LSP_DSP_VEC8(0x3f800000), // 1.0
LSP_DSP_VEC8(0x80000000) // signbit
};
)
#define COMPLEX_MUL3_CORE(DST, SRC1, SRC2, SEL) \
__ASM_EMIT ("xor %[off], %[off]") \
/* x8 blocks */ \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("jb 2f") \
__ASM_EMIT ("1:") \
__ASM_EMIT32("mov %[" SRC1 "_re], %[ptr_re]") \
__ASM_EMIT32("mov %[" SRC1 "_im], %[ptr_im]") \
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_re], %[off]), %%ymm4") /* ymm4 = ar */ \
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_im], %[off]), %%ymm5") /* ymm5 = ai */ \
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%ymm4") /* ymm4 = ar */ \
__ASM_EMIT32("vmovups 0x00(%[ptr_im], %[off]), %%ymm5") /* ymm5 = ai */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_re], %[off]), %%ymm0") /* ymm0 = br */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_im], %[off]), %%ymm1") /* ymm1 = bi */ \
__ASM_EMIT ("vmulps %%ymm0, %%ymm5, %%ymm2") /* ymm2 = ai*br */ \
__ASM_EMIT ("vmulps %%ymm1, %%ymm5, %%ymm3") /* ymm3 = ai*bi */ \
__ASM_EMIT (SEL("vmulps %%ymm0, %%ymm4, %%ymm0", "")) /* ymm0 = ar*br */ \
__ASM_EMIT (SEL("vmulps %%ymm1, %%ymm4, %%ymm1", "")) /* ymm1 = ar*bi */ \
__ASM_EMIT32("mov %[" DST "_re], %[ptr_re]") \
__ASM_EMIT32("mov %[" DST "_im], %[ptr_im]") \
__ASM_EMIT (SEL("vsubps %%ymm3, %%ymm0, %%ymm0", "vfmsub132ps %%ymm4, %%ymm3, %%ymm0")) /* ymm0 = ar*br - ai*bi = r */ \
__ASM_EMIT (SEL("vaddps %%ymm2, %%ymm1, %%ymm1", "vfmadd132ps %%ymm4, %%ymm2, %%ymm1")) /* ymm1 = ar*bi + ai*br = i */ \
__ASM_EMIT32("vmovups %%ymm0, 0x00(%[ptr_re], %[off])") \
__ASM_EMIT32("vmovups %%ymm1, 0x00(%[ptr_im], %[off])") \
__ASM_EMIT64("vmovups %%ymm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT64("vmovups %%ymm1, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x20, %[off]") \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("jae 1b") \
__ASM_EMIT ("2:") \
/* 4x block */ \
__ASM_EMIT32("addl $4, %[count]") \
__ASM_EMIT64("add $4, %[count]") \
__ASM_EMIT ("jl 4f") \
__ASM_EMIT32("mov %[" SRC1 "_re], %[ptr_re]") \
__ASM_EMIT32("mov %[" SRC1 "_im], %[ptr_im]") \
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_re], %[off]), %%xmm4") /* xmm4 = ar */ \
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_im], %[off]), %%xmm5") /* xmm5 = ai */ \
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%xmm4") /* xmm4 = ar */ \
__ASM_EMIT32("vmovups 0x00(%[ptr_im], %[off]), %%xmm5") /* xmm5 = ai */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_re], %[off]), %%xmm0") /* xmm0 = br */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_im], %[off]), %%xmm1") /* xmm1 = bi */ \
__ASM_EMIT ("vmulps %%xmm0, %%xmm5, %%xmm2") /* xmm2 = ai*br */ \
__ASM_EMIT ("vmulps %%xmm1, %%xmm5, %%xmm3") /* xmm3 = ai*bi */ \
__ASM_EMIT (SEL("vmulps %%xmm0, %%xmm4, %%xmm0", "")) /* xmm0 = ar*br */ \
__ASM_EMIT (SEL("vmulps %%xmm1, %%xmm4, %%xmm1", "")) /* xmm1 = ar*bi */ \
__ASM_EMIT32("mov %[" DST "_re], %[ptr_re]") \
__ASM_EMIT32("mov %[" DST "_im], %[ptr_im]") \
__ASM_EMIT (SEL("vsubps %%xmm3, %%xmm0, %%xmm0", "vfmsub132ps %%xmm4, %%xmm3, %%xmm0")) /* xmm0 = ar*br - ai*bi = r */ \
__ASM_EMIT (SEL("vaddps %%xmm2, %%xmm1, %%xmm1", "vfmadd132ps %%xmm4, %%xmm2, %%xmm1")) /* xmm1 = ar*bi + ai*br = i */ \
__ASM_EMIT32("vmovups %%xmm0, 0x00(%[ptr_re], %[off])") \
__ASM_EMIT32("vmovups %%xmm1, 0x00(%[ptr_im], %[off])") \
__ASM_EMIT64("vmovups %%xmm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT64("vmovups %%xmm1, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x10, %[off]") \
__ASM_EMIT32("subl $4, %[count]") \
__ASM_EMIT64("sub $4, %[count]") \
__ASM_EMIT ("4:") \
/* 1x blocks */ \
__ASM_EMIT32("addl $3, %[count]") \
__ASM_EMIT64("add $3, %[count]") \
__ASM_EMIT ("jl 6f") \
__ASM_EMIT ("5:") \
__ASM_EMIT32("mov %[" SRC1 "_re], %[ptr_re]") \
__ASM_EMIT32("mov %[" SRC1 "_im], %[ptr_im]") \
__ASM_EMIT64("vmovss 0x00(%[" SRC1 "_re], %[off]), %%xmm4") /* xmm4 = ar */ \
__ASM_EMIT64("vmovss 0x00(%[" SRC1 "_im], %[off]), %%xmm5") /* xmm5 = ai */ \
__ASM_EMIT32("vmovss 0x00(%[ptr_re], %[off]), %%xmm4") /* xmm4 = ar */ \
__ASM_EMIT32("vmovss 0x00(%[ptr_im], %[off]), %%xmm5") /* xmm5 = ai */ \
__ASM_EMIT ("vmovss 0x00(%[" SRC2 "_re], %[off]), %%xmm0") /* xmm0 = br */ \
__ASM_EMIT ("vmovss 0x00(%[" SRC2 "_im], %[off]), %%xmm1") /* xmm1 = bi */ \
__ASM_EMIT ("vmulps %%xmm0, %%xmm5, %%xmm2") /* xmm2 = ai*br */ \
__ASM_EMIT ("vmulps %%xmm1, %%xmm5, %%xmm3") /* xmm3 = ai*bi */ \
__ASM_EMIT (SEL("vmulss %%xmm0, %%xmm4, %%xmm0", "")) /* xmm0 = ar*br */ \
__ASM_EMIT (SEL("vmulss %%xmm1, %%xmm4, %%xmm1", "")) /* xmm1 = ar*bi */ \
__ASM_EMIT32("mov %[" DST "_re], %[ptr_re]") \
__ASM_EMIT32("mov %[" DST "_im], %[ptr_im]") \
__ASM_EMIT (SEL("vsubss %%xmm3, %%xmm0, %%xmm0", "vfmsub132ss %%xmm4, %%xmm3, %%xmm0")) /* xmm0 = ar*br - ai*bi = r */ \
__ASM_EMIT (SEL("vaddss %%xmm2, %%xmm1, %%xmm1", "vfmadd132ss %%xmm4, %%xmm2, %%xmm1")) /* xmm1 = ar*bi + ai*br = i */ \
__ASM_EMIT32("vmovss %%xmm0, 0x00(%[ptr_re], %[off])") \
__ASM_EMIT32("vmovss %%xmm1, 0x00(%[ptr_im], %[off])") \
__ASM_EMIT64("vmovss %%xmm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT64("vmovss %%xmm1, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x04, %[off]") \
__ASM_EMIT32("decl %[count]") \
__ASM_EMIT64("dec %[count]") \
__ASM_EMIT ("jge 5b") \
__ASM_EMIT ("6:")
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)
{
IF_ARCH_I386(float *ptr_re, *ptr_im);
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_MUL3_CORE("dst", "src1", "src2", FMA_OFF)
: __IF_32(
[ptr_re] "=&r" (ptr_re), [ptr_im] "=&r" (ptr_im),
)
[count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] X86_GREG (dst_re), [dst_im] X86_GREG (dst_im),
[src1_re] X86_GREG (src1_re), [src1_im] X86_GREG (src1_im),
[src2_re] "r" (src2_re), [src2_im] "r" (src2_im)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5"
);
}
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)
{
IF_ARCH_I386(float *ptr_re, *ptr_im);
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_MUL3_CORE("dst", "src1", "src2", FMA_ON)
: __IF_32(
[ptr_re] "=&r" (ptr_re), [ptr_im] "=&r" (ptr_im),
)
[count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] X86_GREG (dst_re), [dst_im] X86_GREG (dst_im),
[src1_re] X86_GREG (src1_re), [src1_im] X86_GREG (src1_im),
[src2_re] "r" (src2_re), [src2_im] "r" (src2_im)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5"
);
}
#undef COMPLEX_MUL3_CORE
#define COMPLEX_MUL2_CORE(DST, SRC, SEL) \
__ASM_EMIT ("xor %[off], %[off]") \
/* x8 blocks */ \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("jb 2f") \
__ASM_EMIT ("1:") \
__ASM_EMIT ("vmovups 0x00(%[" DST "_re], %[off]), %%ymm4") /* ymm4 = ar */ \
__ASM_EMIT ("vmovups 0x00(%[" DST "_im], %[off]), %%ymm5") /* ymm5 = ai */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = br */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm1") /* ymm1 = bi */ \
__ASM_EMIT ("vmulps %%ymm0, %%ymm5, %%ymm2") /* ymm2 = ai*br */ \
__ASM_EMIT ("vmulps %%ymm1, %%ymm5, %%ymm3") /* ymm3 = ai*bi */ \
__ASM_EMIT (SEL("vmulps %%ymm0, %%ymm4, %%ymm0", "")) /* ymm0 = ar*br */ \
__ASM_EMIT (SEL("vmulps %%ymm1, %%ymm4, %%ymm1", "")) /* ymm1 = ar*bi */ \
__ASM_EMIT (SEL("vsubps %%ymm3, %%ymm0, %%ymm0", "vfmsub132ps %%ymm4, %%ymm3, %%ymm0")) /* ymm0 = ar*br - ai*bi = r */ \
__ASM_EMIT (SEL("vaddps %%ymm2, %%ymm1, %%ymm1", "vfmadd132ps %%ymm4, %%ymm2, %%ymm1")) /* ymm1 = ar*bi + ai*br = i */ \
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT ("vmovups %%ymm1, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x20, %[off]") \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("jae 1b") \
__ASM_EMIT ("2:") \
/* 4x block */ \
__ASM_EMIT32("addl $4, %[count]") \
__ASM_EMIT64("add $4, %[count]") \
__ASM_EMIT ("jl 4f") \
__ASM_EMIT ("vmovups 0x00(%[" DST "_re], %[off]), %%xmm4") /* xmm4 = ar */ \
__ASM_EMIT ("vmovups 0x00(%[" DST "_im], %[off]), %%xmm5") /* xmm5 = ai */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = br */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%xmm1") /* xmm1 = bi */ \
__ASM_EMIT ("vmulps %%xmm0, %%xmm5, %%xmm2") /* xmm2 = ai*br */ \
__ASM_EMIT ("vmulps %%xmm1, %%xmm5, %%xmm3") /* xmm3 = ai*bi */ \
__ASM_EMIT (SEL("vmulps %%xmm0, %%xmm4, %%xmm0", "")) /* xmm0 = ar*br */ \
__ASM_EMIT (SEL("vmulps %%xmm1, %%xmm4, %%xmm1", "")) /* xmm1 = ar*bi */ \
__ASM_EMIT (SEL("vsubps %%xmm3, %%xmm0, %%xmm0", "vfmsub132ps %%xmm4, %%xmm3, %%xmm0")) /* xmm0 = ar*br - ai*bi = r */ \
__ASM_EMIT (SEL("vaddps %%xmm2, %%xmm1, %%xmm1", "vfmadd132ps %%xmm4, %%xmm2, %%xmm1")) /* xmm1 = ar*bi + ai*br = i */ \
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT ("vmovups %%xmm1, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x10, %[off]") \
__ASM_EMIT32("subl $4, %[count]") \
__ASM_EMIT64("sub $4, %[count]") \
__ASM_EMIT ("4:") \
/* 1x blocks */ \
__ASM_EMIT32("addl $3, %[count]") \
__ASM_EMIT64("add $3, %[count]") \
__ASM_EMIT ("jl 6f") \
__ASM_EMIT ("5:") \
__ASM_EMIT ("vmovss 0x00(%[" DST "_re], %[off]), %%xmm4") /* xmm4 = ar */ \
__ASM_EMIT ("vmovss 0x00(%[" DST "_im], %[off]), %%xmm5") /* xmm5 = ai */ \
__ASM_EMIT ("vmovss 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = br */ \
__ASM_EMIT ("vmovss 0x00(%[" SRC "_im], %[off]), %%xmm1") /* xmm1 = bi */ \
__ASM_EMIT ("vmulps %%xmm0, %%xmm5, %%xmm2") /* xmm2 = ai*br */ \
__ASM_EMIT ("vmulps %%xmm1, %%xmm5, %%xmm3") /* xmm3 = ai*bi */ \
__ASM_EMIT (SEL("vmulss %%xmm0, %%xmm4, %%xmm0", "")) /* xmm0 = ar*br */ \
__ASM_EMIT (SEL("vmulss %%xmm1, %%xmm4, %%xmm1", "")) /* xmm1 = ar*bi */ \
__ASM_EMIT (SEL("vsubss %%xmm3, %%xmm0, %%xmm0", "vfmsub132ss %%xmm4, %%xmm3, %%xmm0")) /* xmm0 = ar*br - ai*bi = r */ \
__ASM_EMIT (SEL("vaddss %%xmm2, %%xmm1, %%xmm1", "vfmadd132ss %%xmm4, %%xmm2, %%xmm1")) /* xmm1 = ar*bi + ai*br = i */ \
__ASM_EMIT ("vmovss %%xmm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT ("vmovss %%xmm1, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x04, %[off]") \
__ASM_EMIT32("decl %[count]") \
__ASM_EMIT64("dec %[count]") \
__ASM_EMIT ("jge 5b") \
__ASM_EMIT ("6:")
void complex_mul2(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_MUL2_CORE("dst", "src", FMA_OFF)
: [count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[src_re] "r" (src_re), [src_im] "r" (src_im)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5"
);
}
void complex_mul2_fma3(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_MUL2_CORE("dst", "src", FMA_ON)
: [count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[src_re] "r" (src_re), [src_im] "r" (src_im)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5"
);
}
#undef COMPLEX_MUL2_CORE
#define COMPLEX_MOD_CORE(DST, SRC, SEL) \
__ASM_EMIT ("xor %[off], %[off]") \
/* x16 blocks */ \
__ASM_EMIT ("sub $16, %[count]") \
__ASM_EMIT ("jb 2f") \
__ASM_EMIT ("1:") \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = r */ \
__ASM_EMIT ("vmovups 0x20(%[" SRC "_re], %[off]), %%ymm1") \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm2") /* ymm2 = i */ \
__ASM_EMIT ("vmovups 0x20(%[" SRC "_im], %[off]), %%ymm3") \
__ASM_EMIT ("vmulps %%ymm0, %%ymm0, %%ymm0") /* ymm0 = r*r */ \
__ASM_EMIT ("vmulps %%ymm1, %%ymm1, %%ymm1") \
__ASM_EMIT (SEL("vmulps %%ymm2, %%ymm2, %%ymm2", "")) /* ymm2 = i*i */ \
__ASM_EMIT (SEL("vmulps %%ymm3, %%ymm3, %%ymm3", "")) \
__ASM_EMIT (SEL("vaddps %%ymm2, %%ymm0, %%ymm0", "vfmadd231ps %%ymm2, %%ymm2, %%ymm0")) /* ymm0 = r*r + i*i */ \
__ASM_EMIT (SEL("vaddps %%ymm3, %%ymm1, %%ymm1", "vfmadd231ps %%ymm3, %%ymm3, %%ymm1")) \
__ASM_EMIT ("vsqrtps %%ymm0, %%ymm0") /* ymm0 = sqrt(r*r + i*i) */ \
__ASM_EMIT ("vsqrtps %%ymm1, %%ymm1") \
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "], %[off])") \
__ASM_EMIT ("vmovups %%ymm1, 0x20(%[" DST "], %[off])") \
__ASM_EMIT ("add $0x40, %[off]") \
__ASM_EMIT ("sub $16, %[count]") \
__ASM_EMIT ("jae 1b") \
__ASM_EMIT ("2:") \
/* x8 block */ \
__ASM_EMIT ("add $8, %[count]") \
__ASM_EMIT ("jl 4f") \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = r */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm2") /* ymm2 = i */ \
__ASM_EMIT ("vmulps %%ymm0, %%ymm0, %%ymm0") /* ymm0 = r*r */ \
__ASM_EMIT (SEL("vmulps %%ymm2, %%ymm2, %%ymm2", "")) /* ymm2 = i*i */ \
__ASM_EMIT (SEL("vaddps %%ymm2, %%ymm0, %%ymm0", "vfmadd231ps %%ymm2, %%ymm2, %%ymm0")) /* ymm0 = r*r + i*i */ \
__ASM_EMIT ("vsqrtps %%ymm0, %%ymm0") /* ymm0 = sqrt(r*r + i*i) */ \
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "], %[off])") \
__ASM_EMIT ("add $0x20, %[off]") \
__ASM_EMIT ("sub $8, %[count]") \
__ASM_EMIT ("4:") \
/* x4 block */ \
__ASM_EMIT ("add $4, %[count]") \
__ASM_EMIT ("jl 6f") \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = r */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = i */ \
__ASM_EMIT ("vmulps %%xmm0, %%xmm0, %%xmm0") /* xmm0 = r*r */ \
__ASM_EMIT (SEL("vmulps %%xmm2, %%xmm2, %%xmm2", "")) /* xmm2 = i*i */ \
__ASM_EMIT (SEL("vaddps %%xmm2, %%xmm0, %%xmm0", "vfmadd231ps %%xmm2, %%xmm2, %%xmm0")) /* xmm0 = r*r + i*i */ \
__ASM_EMIT ("vsqrtps %%xmm0, %%xmm0") /* xmm0 = sqrt(r*r + i*i) */ \
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DST "], %[off])") \
__ASM_EMIT ("add $0x10, %[off]") \
__ASM_EMIT ("sub $4, %[count]") \
__ASM_EMIT ("6:") \
/* x1 block */ \
__ASM_EMIT ("add $3, %[count]") \
__ASM_EMIT ("jl 8f") \
__ASM_EMIT ("7:") \
__ASM_EMIT ("vmovss 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = r */ \
__ASM_EMIT ("vmovss 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = i */ \
__ASM_EMIT ("vmulss %%xmm0, %%xmm0, %%xmm0") /* xmm0 = r*r */ \
__ASM_EMIT (SEL("vmulss %%xmm2, %%xmm2, %%xmm2", "")) /* xmm2 = i*i */ \
__ASM_EMIT (SEL("vaddss %%xmm2, %%xmm0, %%xmm0", "vfmadd231ps %%xmm2, %%xmm2, %%xmm0")) /* xmm0 = r*r + i*i */ \
__ASM_EMIT ("vsqrtss %%xmm0, %%xmm0, %%xmm0") /* xmm0 = sqrt(r*r + i*i) */ \
__ASM_EMIT ("vmovss %%xmm0, 0x00(%[" DST "], %[off])") \
__ASM_EMIT ("add $0x04, %[off]") \
__ASM_EMIT ("dec %[count]") \
__ASM_EMIT ("jge 7b") \
__ASM_EMIT ("8:") \
void complex_mod(float *dst, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_MOD_CORE("dst", "src", FMA_OFF)
: [count] "+r" (count), [off] "=&r" (off)
: [dst] "r" (dst), [src_re] "r" (src_re), [src_im] "r" (src_im)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3"
);
}
void complex_mod_fma3(float *dst, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_MOD_CORE("dst", "src", FMA_ON)
: [count] "+r" (count), [off] "=&r" (off)
: [dst] "r" (dst), [src_re] "r" (src_re), [src_im] "r" (src_im)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3"
);
}
#undef COMPLEX_MOD_CORE
#define COMPLEX_DIV3_CORE(DSTR, DSTI, TR, TI, BR, BI, __ASM_NOFMA, __ASM_FMA) \
__ASM_EMIT ("xor %[off], %[off]") \
/* x8 blocks */ \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("jb 2f") \
__ASM_EMIT ("1:") \
__ASM_EMIT32("mov %[" TR "], %[ptr_re]") \
__ASM_EMIT32("mov %[" TI "], %[ptr_im]") \
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%ymm2") /* ymm2 = br */ \
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%ymm3") /* ymm3 = bi */ \
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%ymm0") /* ymm0 = tr */ \
__ASM_EMIT32("vmovups 0x00(%[ptr_im], %[off]), %%ymm1") /* ymm1 = ti */ \
__ASM_EMIT64("vmovups 0x00(%[" TR "], %[off]), %%ymm0") /* ymm0 = tr */ \
__ASM_EMIT64("vmovups 0x00(%[" TI "], %[off]), %%ymm1") /* ymm1 = ti */ \
__ASM_EMIT32("mov %[" DSTR "], %[ptr_re]") \
__ASM_EMIT32("mov %[" DSTI "], %[ptr_im]") \
__ASM_EMIT ("vmulps %%ymm2, %%ymm2, %%ymm7") /* ymm7 = br*br */ \
__ASM_NOFMA ("vmulps %%ymm3, %%ymm3, %%ymm6") /* ymm6 = bi*bi */ \
__ASM_EMIT ("vmulps %%ymm3, %%ymm1, %%ymm4") /* ymm4 = ti*bi */ \
__ASM_EMIT ("vmulps %%ymm3, %%ymm0, %%ymm5") /* ymm5 = tr*bi */ \
__ASM_NOFMA ("vaddps %%ymm6, %%ymm7, %%ymm7") /* ymm7 = M = br*br + bi*bi */ \
__ASM_FMA ("vfmadd231ps %%ymm3, %%ymm3, %%ymm7") /* ymm7 = M = br*br + bi*bi */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%ymm6") /* ymm6 = 1 */ \
__ASM_NOFMA ("vmulps %%ymm2, %%ymm0, %%ymm0") /* ymm0 = tr*br */ \
__ASM_NOFMA ("vmulps %%ymm2, %%ymm1, %%ymm1") /* ymm1 = ti*br */ \
__ASM_EMIT ("vdivps %%ymm7, %%ymm6, %%ymm7") /* ymm7 = R = 1/M */ \
__ASM_NOFMA ("vaddps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = r = tr*br + ti*bi */ \
__ASM_NOFMA ("vsubps %%ymm5, %%ymm1, %%ymm1") /* ymm1 = i = ti*br - tr*bi */ \
__ASM_FMA ("vfmadd132ps %%ymm2, %%ymm4, %%ymm0") /* ymm0 = r = tr*br + ti*bi */ \
__ASM_FMA ("vfmsub132ps %%ymm2, %%ymm5, %%ymm1") /* ymm1 = i = tr*br - tr*bi */ \
__ASM_EMIT ("vmulps %%ymm7, %%ymm0, %%ymm0") /* ymm0 = r/R */ \
__ASM_EMIT ("vmulps %%ymm7, %%ymm1, %%ymm1") /* ymm1 = i/R */ \
__ASM_EMIT32("vmovups %%ymm0, 0x00(%[ptr_re], %[off])") \
__ASM_EMIT32("vmovups %%ymm1, 0x00(%[ptr_im], %[off])") \
__ASM_EMIT64("vmovups %%ymm0, 0x00(%[" DSTR "], %[off])") \
__ASM_EMIT64("vmovups %%ymm1, 0x00(%[" DSTI "], %[off])") \
__ASM_EMIT ("add $0x20, %[off]") \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("jae 1b") \
__ASM_EMIT ("2:") \
/* 4x block */ \
__ASM_EMIT32("addl $4, %[count]") \
__ASM_EMIT64("add $4, %[count]") \
__ASM_EMIT ("jl 4f") \
__ASM_EMIT32("mov %[" TR "], %[ptr_re]") \
__ASM_EMIT32("mov %[" TI "], %[ptr_im]") \
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%xmm2") /* xmm2 = br */ \
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%xmm3") /* xmm3 = bi */ \
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%xmm0") /* xmm0 = tr */ \
__ASM_EMIT32("vmovups 0x00(%[ptr_im], %[off]), %%xmm1") /* xmm1 = ti */ \
__ASM_EMIT64("vmovups 0x00(%[" TR "], %[off]), %%xmm0") /* xmm0 = tr */ \
__ASM_EMIT64("vmovups 0x00(%[" TI "], %[off]), %%xmm1") /* xmm1 = ti */ \
__ASM_EMIT32("mov %[" DSTR "], %[ptr_re]") \
__ASM_EMIT32("mov %[" DSTI "], %[ptr_im]") \
__ASM_EMIT ("vmulps %%xmm2, %%xmm2, %%xmm7") /* xmm7 = br*br */ \
__ASM_NOFMA ("vmulps %%xmm3, %%xmm3, %%xmm6") /* xmm6 = bi*bi */ \
__ASM_EMIT ("vmulps %%xmm3, %%xmm1, %%xmm4") /* xmm4 = ti*bi */ \
__ASM_EMIT ("vmulps %%xmm3, %%xmm0, %%xmm5") /* xmm5 = tr*bi */ \
__ASM_NOFMA ("vaddps %%xmm6, %%xmm7, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \
__ASM_FMA ("vfmadd231ps %%xmm3, %%xmm3, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \
__ASM_NOFMA ("vmulps %%xmm2, %%xmm0, %%xmm0") /* xmm0 = tr*br */ \
__ASM_NOFMA ("vmulps %%xmm2, %%xmm1, %%xmm1") /* xmm1 = ti*br */ \
__ASM_EMIT ("vdivps %%xmm7, %%xmm6, %%xmm7") /* xmm7 = R = 1/M */ \
__ASM_NOFMA ("vaddps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \
__ASM_NOFMA ("vsubps %%xmm5, %%xmm1, %%xmm1") /* xmm1 = i = ti*br - tr*bi */ \
__ASM_FMA ("vfmadd132ps %%xmm2, %%xmm4, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \
__ASM_FMA ("vfmsub132ps %%xmm2, %%xmm5, %%xmm1") /* xmm1 = i = tr*br - tr*bi */ \
__ASM_EMIT ("vmulps %%xmm7, %%xmm0, %%xmm0") /* xmm0 = r/R */ \
__ASM_EMIT ("vmulps %%xmm7, %%xmm1, %%xmm1") /* xmm1 = i/R */ \
__ASM_EMIT32("vmovups %%xmm0, 0x00(%[ptr_re], %[off])") \
__ASM_EMIT32("vmovups %%xmm1, 0x00(%[ptr_im], %[off])") \
__ASM_EMIT64("vmovups %%xmm0, 0x00(%[" DSTR "], %[off])") \
__ASM_EMIT64("vmovups %%xmm1, 0x00(%[" DSTI "], %[off])") \
__ASM_EMIT32("subl $4, %[count]") \
__ASM_EMIT64("sub $4, %[count]") \
__ASM_EMIT ("add $0x10, %[off]") \
__ASM_EMIT ("4:") \
/* 1x blocks */ \
__ASM_EMIT32("addl $3, %[count]") \
__ASM_EMIT64("add $3, %[count]") \
__ASM_EMIT ("jl 6f") \
__ASM_EMIT ("5:") \
__ASM_EMIT32("mov %[" TR "], %[ptr_re]") \
__ASM_EMIT32("mov %[" TI "], %[ptr_im]") \
__ASM_EMIT ("vmovss 0x00(%[" BR "], %[off]), %%xmm2") /* xmm2 = br */ \
__ASM_EMIT ("vmovss 0x00(%[" BI "], %[off]), %%xmm3") /* xmm3 = bi */ \
__ASM_EMIT32("vmovss 0x00(%[ptr_re], %[off]), %%xmm0") /* xmm0 = tr */ \
__ASM_EMIT32("vmovss 0x00(%[ptr_im], %[off]), %%xmm1") /* xmm1 = ti */ \
__ASM_EMIT64("vmovss 0x00(%[" TR "], %[off]), %%xmm0") /* xmm0 = tr */ \
__ASM_EMIT64("vmovss 0x00(%[" TI "], %[off]), %%xmm1") /* xmm1 = ti */ \
__ASM_EMIT32("mov %[" DSTR "], %[ptr_re]") \
__ASM_EMIT32("mov %[" DSTI "], %[ptr_im]") \
__ASM_EMIT ("vmulss %%xmm2, %%xmm2, %%xmm7") /* xmm7 = br*br */ \
__ASM_NOFMA ("vmulss %%xmm3, %%xmm3, %%xmm6") /* xmm6 = bi*bi */ \
__ASM_EMIT ("vmulss %%xmm3, %%xmm1, %%xmm4") /* xmm4 = ti*bi */ \
__ASM_EMIT ("vmulss %%xmm3, %%xmm0, %%xmm5") /* xmm5 = tr*bi */ \
__ASM_NOFMA ("vaddss %%xmm6, %%xmm7, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \
__ASM_FMA ("vfmadd231ss %%xmm3, %%xmm3, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \
__ASM_NOFMA ("vmulss %%xmm2, %%xmm0, %%xmm0") /* xmm0 = tr*br */ \
__ASM_NOFMA ("vmulss %%xmm2, %%xmm1, %%xmm1") /* xmm1 = ti*br */ \
__ASM_EMIT ("vdivss %%xmm7, %%xmm6, %%xmm7") /* xmm7 = R = 1/M */ \
__ASM_NOFMA ("vaddss %%xmm4, %%xmm0, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \
__ASM_NOFMA ("vsubss %%xmm5, %%xmm1, %%xmm1") /* xmm1 = i = ti*br - tr*bi */ \
__ASM_FMA ("vfmadd132ss %%xmm2, %%xmm4, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \
__ASM_FMA ("vfmsub132ss %%xmm2, %%xmm5, %%xmm1") /* xmm1 = i = tr*br - tr*bi */ \
__ASM_EMIT ("vmulss %%xmm7, %%xmm0, %%xmm0") /* xmm0 = r/R */ \
__ASM_EMIT ("vmulss %%xmm7, %%xmm1, %%xmm1") /* xmm1 = i/R */ \
__ASM_EMIT32("vmovss %%xmm0, 0x00(%[ptr_re], %[off])") \
__ASM_EMIT32("vmovss %%xmm1, 0x00(%[ptr_im], %[off])") \
__ASM_EMIT64("vmovss %%xmm0, 0x00(%[" DSTR "], %[off])") \
__ASM_EMIT64("vmovss %%xmm1, 0x00(%[" DSTI "], %[off])") \
__ASM_EMIT ("add $0x04, %[off]") \
__ASM_EMIT32("decl %[count]") \
__ASM_EMIT64("dec %[count]") \
__ASM_EMIT ("jge 5b") \
__ASM_EMIT ("6:")
void complex_div3(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)
{
IF_ARCH_I386(float *ptr_re, *ptr_im);
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_DIV3_CORE("dst_re", "dst_im", "src1_re", "src1_im", "src2_re", "src2_im", FMA_PASS, FMA_SKIP)
: __IF_32(
[ptr_re] "=&r" (ptr_re), [ptr_im] "=&r" (ptr_im),
)
[count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] X86_GREG (dst_re), [dst_im] X86_GREG (dst_im),
[src1_re] X86_GREG (src1_re), [src1_im] X86_GREG (src1_im),
[src2_re] "r" (src2_re), [src2_im] "r" (src2_im),
[CC] "o" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
void complex_div3_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)
{
IF_ARCH_I386(float *ptr_re, *ptr_im);
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_DIV3_CORE("dst_re", "dst_im", "src1_re", "src1_im", "src2_re", "src2_im", FMA_SKIP, FMA_PASS)
: __IF_32(
[ptr_re] "=&r" (ptr_re), [ptr_im] "=&r" (ptr_im),
)
[count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] X86_GREG (dst_re), [dst_im] X86_GREG (dst_im),
[src1_re] X86_GREG (src1_re), [src1_im] X86_GREG (src1_im),
[src2_re] "r" (src2_re), [src2_im] "r" (src2_im),
[CC] "o" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
#undef COMPLEX_DIV3_CORE
#define COMPLEX_DIV2_CORE(DSTR, DSTI, TR, TI, BR, BI, __ASM_NOFMA, __ASM_FMA) \
__ASM_EMIT ("xor %[off], %[off]") \
/* x8 blocks */ \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("jb 2f") \
__ASM_EMIT ("1:") \
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%ymm2") /* ymm2 = br */ \
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%ymm3") /* ymm3 = bi */ \
__ASM_EMIT ("vmovups 0x00(%[" TR "], %[off]), %%ymm0") /* ymm0 = tr */ \
__ASM_EMIT ("vmovups 0x00(%[" TI "], %[off]), %%ymm1") /* ymm1 = ti */ \
__ASM_EMIT ("vmulps %%ymm2, %%ymm2, %%ymm7") /* ymm7 = br*br */ \
__ASM_NOFMA ("vmulps %%ymm3, %%ymm3, %%ymm6") /* ymm6 = bi*bi */ \
__ASM_EMIT ("vmulps %%ymm3, %%ymm1, %%ymm4") /* ymm4 = ti*bi */ \
__ASM_EMIT ("vmulps %%ymm3, %%ymm0, %%ymm5") /* ymm5 = tr*bi */ \
__ASM_NOFMA ("vaddps %%ymm6, %%ymm7, %%ymm7") /* ymm7 = M = br*br + bi*bi */ \
__ASM_FMA ("vfmadd231ps %%ymm3, %%ymm3, %%ymm7") /* ymm7 = M = br*br + bi*bi */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%ymm6") /* ymm6 = 1 */ \
__ASM_NOFMA ("vmulps %%ymm2, %%ymm0, %%ymm0") /* ymm0 = tr*br */ \
__ASM_NOFMA ("vmulps %%ymm2, %%ymm1, %%ymm1") /* ymm1 = ti*br */ \
__ASM_EMIT ("vdivps %%ymm7, %%ymm6, %%ymm7") /* ymm7 = R = 1/M */ \
__ASM_NOFMA ("vaddps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = r = tr*br + ti*bi */ \
__ASM_NOFMA ("vsubps %%ymm5, %%ymm1, %%ymm1") /* ymm1 = i = ti*br - tr*bi */ \
__ASM_FMA ("vfmadd132ps %%ymm2, %%ymm4, %%ymm0") /* ymm0 = r = tr*br + ti*bi */ \
__ASM_FMA ("vfmsub132ps %%ymm2, %%ymm5, %%ymm1") /* ymm1 = i = tr*br - tr*bi */ \
__ASM_EMIT ("vmulps %%ymm7, %%ymm0, %%ymm0") /* ymm0 = r/R */ \
__ASM_EMIT ("vmulps %%ymm7, %%ymm1, %%ymm1") /* ymm1 = i/R */ \
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DSTR "], %[off])") \
__ASM_EMIT ("vmovups %%ymm1, 0x00(%[" DSTI "], %[off])") \
__ASM_EMIT ("add $0x20, %[off]") \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("jae 1b") \
__ASM_EMIT ("2:") \
/* 4x block */ \
__ASM_EMIT32("addl $4, %[count]") \
__ASM_EMIT64("add $4, %[count]") \
__ASM_EMIT ("jl 4f") \
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%xmm2") /* xmm2 = br */ \
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%xmm3") /* xmm3 = bi */ \
__ASM_EMIT ("vmovups 0x00(%[" TR "], %[off]), %%xmm0") /* xmm0 = tr */ \
__ASM_EMIT ("vmovups 0x00(%[" TI "], %[off]), %%xmm1") /* xmm1 = ti */ \
__ASM_EMIT ("vmulps %%xmm2, %%xmm2, %%xmm7") /* xmm7 = br*br */ \
__ASM_NOFMA ("vmulps %%xmm3, %%xmm3, %%xmm6") /* xmm6 = bi*bi */ \
__ASM_EMIT ("vmulps %%xmm3, %%xmm1, %%xmm4") /* xmm4 = ti*bi */ \
__ASM_EMIT ("vmulps %%xmm3, %%xmm0, %%xmm5") /* xmm5 = tr*bi */ \
__ASM_NOFMA ("vaddps %%xmm6, %%xmm7, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \
__ASM_FMA ("vfmadd231ps %%xmm3, %%xmm3, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \
__ASM_NOFMA ("vmulps %%xmm2, %%xmm0, %%xmm0") /* xmm0 = tr*br */ \
__ASM_NOFMA ("vmulps %%xmm2, %%xmm1, %%xmm1") /* xmm1 = ti*br */ \
__ASM_EMIT ("vdivps %%xmm7, %%xmm6, %%xmm7") /* xmm7 = R = 1/M */ \
__ASM_NOFMA ("vaddps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \
__ASM_NOFMA ("vsubps %%xmm5, %%xmm1, %%xmm1") /* xmm1 = i = ti*br - tr*bi */ \
__ASM_FMA ("vfmadd132ps %%xmm2, %%xmm4, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \
__ASM_FMA ("vfmsub132ps %%xmm2, %%xmm5, %%xmm1") /* xmm1 = i = tr*br - tr*bi */ \
__ASM_EMIT ("vmulps %%xmm7, %%xmm0, %%xmm0") /* xmm0 = r/R */ \
__ASM_EMIT ("vmulps %%xmm7, %%xmm1, %%xmm1") /* xmm1 = i/R */ \
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DSTR "], %[off])") \
__ASM_EMIT ("vmovups %%xmm1, 0x00(%[" DSTI "], %[off])") \
__ASM_EMIT32("subl $4, %[count]") \
__ASM_EMIT64("sub $4, %[count]") \
__ASM_EMIT ("add $0x10, %[off]") \
__ASM_EMIT ("4:") \
/* 1x blocks */ \
__ASM_EMIT32("addl $3, %[count]") \
__ASM_EMIT64("add $3, %[count]") \
__ASM_EMIT ("jl 6f") \
__ASM_EMIT ("5:") \
__ASM_EMIT ("vmovss 0x00(%[" BR "], %[off]), %%xmm2") /* xmm2 = br */ \
__ASM_EMIT ("vmovss 0x00(%[" BI "], %[off]), %%xmm3") /* xmm3 = bi */ \
__ASM_EMIT ("vmovss 0x00(%[" TR "], %[off]), %%xmm0") /* xmm0 = tr */ \
__ASM_EMIT ("vmovss 0x00(%[" TI "], %[off]), %%xmm1") /* xmm1 = ti */ \
__ASM_EMIT ("vmulss %%xmm2, %%xmm2, %%xmm7") /* xmm7 = br*br */ \
__ASM_NOFMA ("vmulss %%xmm3, %%xmm3, %%xmm6") /* xmm6 = bi*bi */ \
__ASM_EMIT ("vmulss %%xmm3, %%xmm1, %%xmm4") /* xmm4 = ti*bi */ \
__ASM_EMIT ("vmulss %%xmm3, %%xmm0, %%xmm5") /* xmm5 = tr*bi */ \
__ASM_NOFMA ("vaddss %%xmm6, %%xmm7, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \
__ASM_FMA ("vfmadd231ss %%xmm3, %%xmm3, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \
__ASM_NOFMA ("vmulss %%xmm2, %%xmm0, %%xmm0") /* xmm0 = tr*br */ \
__ASM_NOFMA ("vmulss %%xmm2, %%xmm1, %%xmm1") /* xmm1 = ti*br */ \
__ASM_EMIT ("vdivss %%xmm7, %%xmm6, %%xmm7") /* xmm7 = R = 1/M */ \
__ASM_NOFMA ("vaddss %%xmm4, %%xmm0, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \
__ASM_NOFMA ("vsubss %%xmm5, %%xmm1, %%xmm1") /* xmm1 = i = ti*br - tr*bi */ \
__ASM_FMA ("vfmadd132ss %%xmm2, %%xmm4, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \
__ASM_FMA ("vfmsub132ss %%xmm2, %%xmm5, %%xmm1") /* xmm1 = i = tr*br - tr*bi */ \
__ASM_EMIT ("vmulss %%xmm7, %%xmm0, %%xmm0") /* xmm0 = r/R */ \
__ASM_EMIT ("vmulss %%xmm7, %%xmm1, %%xmm1") /* xmm1 = i/R */ \
__ASM_EMIT ("vmovss %%xmm0, 0x00(%[" DSTR "], %[off])") \
__ASM_EMIT ("vmovss %%xmm1, 0x00(%[" DSTI "], %[off])") \
__ASM_EMIT ("add $0x04, %[off]") \
__ASM_EMIT32("decl %[count]") \
__ASM_EMIT64("dec %[count]") \
__ASM_EMIT ("jge 5b") \
__ASM_EMIT ("6:")
void complex_div2(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_DIV2_CORE("dst_re", "dst_im", "dst_re", "dst_im", "src_re", "src_im", FMA_PASS, FMA_SKIP)
: [count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[src_re] "r" (src_re), [src_im] "r" (src_im),
[CC] "m" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
void complex_div2_fma3(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_DIV2_CORE("dst_re", "dst_im", "dst_re", "dst_im", "src_re", "src_im", FMA_SKIP, FMA_PASS)
: [count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[src_re] "r" (src_re), [src_im] "r" (src_im),
[CC] "m" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
void complex_rdiv2(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_DIV2_CORE("dst_re", "dst_im", "src_re", "src_im", "dst_re", "dst_im", FMA_PASS, FMA_SKIP)
: [count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[src_re] "r" (src_re), [src_im] "r" (src_im),
[CC] "m" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
void complex_rdiv2_fma3(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_DIV2_CORE("dst_re", "dst_im", "src_re", "src_im", "dst_re", "dst_im", FMA_SKIP, FMA_PASS)
: [count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[src_re] "r" (src_re), [src_im] "r" (src_im),
[CC] "m" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
#undef COMPLEX_DIV2_CORE
#define COMPLEX_RCP_CORE(DST, SRC, SEL) \
__ASM_EMIT ("xor %[off], %[off]") \
/* x16 blocks */ \
__ASM_EMIT32("subl $16, %[count]") \
__ASM_EMIT64("sub $16, %[count]") \
__ASM_EMIT ("jb 2f") \
__ASM_EMIT ("1:") \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = ar */ \
__ASM_EMIT ("vmovups 0x20(%[" SRC "_re], %[off]), %%ymm1") \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm2") /* ymm2 = ai */ \
__ASM_EMIT ("vmovups 0x20(%[" SRC "_im], %[off]), %%ymm3") \
__ASM_EMIT ("vmulps %%ymm0, %%ymm0, %%ymm4") /* ymm4 = ar*ar */ \
__ASM_EMIT ("vmulps %%ymm1, %%ymm1, %%ymm5") \
__ASM_EMIT (SEL("vmulps %%ymm2, %%ymm2, %%ymm6", "")) /* ymm6 = ai*ai */ \
__ASM_EMIT (SEL("vmulps %%ymm3, %%ymm3, %%ymm7", "")) \
__ASM_EMIT (SEL("vaddps %%ymm6, %%ymm4, %%ymm4", "vfmadd231ps %%ymm2, %%ymm2, %%ymm4")) /* ymm4 = R = ar*ar+ai*ai */ \
__ASM_EMIT (SEL("vaddps %%ymm7, %%ymm5, %%ymm5", "vfmadd231ps %%ymm3, %%ymm3, %%ymm5")) \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%ymm6") /* ymm6 = 1 */ \
__ASM_EMIT ("vxorps 0x20 + %[CC], %%ymm2, %%ymm2") /* ymm1 = -ai */ \
__ASM_EMIT ("vxorps 0x20 + %[CC], %%ymm3, %%ymm3") \
__ASM_EMIT ("vdivps %%ymm4, %%ymm6, %%ymm4") /* ymm4 = 1/R */ \
__ASM_EMIT ("vdivps %%ymm5, %%ymm6, %%ymm5") \
__ASM_EMIT ("vmulps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = ar/R */ \
__ASM_EMIT ("vmulps %%ymm5, %%ymm1, %%ymm1") \
__ASM_EMIT ("vmulps %%ymm4, %%ymm2, %%ymm2") /* ymm1 = -ai/R */ \
__ASM_EMIT ("vmulps %%ymm5, %%ymm3, %%ymm3") \
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT ("vmovups %%ymm1, 0x20(%[" DST "_re], %[off])") \
__ASM_EMIT ("vmovups %%ymm2, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("vmovups %%ymm3, 0x20(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x40, %[off]") \
__ASM_EMIT32("subl $16, %[count]") \
__ASM_EMIT64("sub $16, %[count]") \
__ASM_EMIT ("jae 1b") \
__ASM_EMIT ("2:") \
/* 8x block */ \
__ASM_EMIT32("addl $8, %[count]") \
__ASM_EMIT64("add $8, %[count]") \
__ASM_EMIT ("jl 4f") \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = ar */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm2") /* ymm2 = ai */ \
__ASM_EMIT ("vmulps %%ymm0, %%ymm0, %%ymm4") /* ymm4 = ar*ar */ \
__ASM_EMIT (SEL("vmulps %%ymm2, %%ymm2, %%ymm6", "")) /* ymm6 = ai*ai */ \
__ASM_EMIT (SEL("vaddps %%ymm6, %%ymm4, %%ymm4", "vfmadd231ps %%ymm2, %%ymm2, %%ymm4")) /* ymm4 = R = ar*ar+ai*ai */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%ymm6") /* ymm6 = 1 */ \
__ASM_EMIT ("vxorps 0x20 + %[CC], %%ymm2, %%ymm2") /* ymm1 = -ai */ \
__ASM_EMIT ("vdivps %%ymm4, %%ymm6, %%ymm4") /* ymm4 = 1/R */ \
__ASM_EMIT ("vmulps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = ar/R */ \
__ASM_EMIT ("vmulps %%ymm4, %%ymm2, %%ymm2") /* ymm1 = -ai/R */ \
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT ("vmovups %%ymm2, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x20, %[off]") \
__ASM_EMIT32("subl $8, %[count]") \
__ASM_EMIT64("sub $8, %[count]") \
__ASM_EMIT ("4:") \
/* 4x block */ \
__ASM_EMIT32("addl $4, %[count]") \
__ASM_EMIT64("add $4, %[count]") \
__ASM_EMIT ("jl 6f") \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = ar */ \
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = ai */ \
__ASM_EMIT ("vmulps %%xmm0, %%xmm0, %%xmm4") /* xmm4 = ar*ar */ \
__ASM_EMIT (SEL("vmulps %%xmm2, %%xmm2, %%xmm6", "")) /* xmm6 = ai*ai */ \
__ASM_EMIT (SEL("vaddps %%xmm6, %%xmm4, %%xmm4", "vfmadd231ps %%xmm2, %%xmm2, %%xmm4")) /* xmm4 = R = ar*ar+ai*ai */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \
__ASM_EMIT ("vxorps 0x20 + %[CC], %%xmm2, %%xmm2") /* xmm1 = -ai */ \
__ASM_EMIT ("vdivps %%xmm4, %%xmm6, %%xmm4") /* xmm4 = 1/R */ \
__ASM_EMIT ("vmulps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = ar/R */ \
__ASM_EMIT ("vmulps %%xmm4, %%xmm2, %%xmm2") /* xmm1 = -ai/R */ \
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT ("vmovups %%xmm2, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x10, %[off]") \
__ASM_EMIT32("subl $4, %[count]") \
__ASM_EMIT64("sub $4, %[count]") \
__ASM_EMIT ("6:") \
/* 1x blocks */ \
__ASM_EMIT32("addl $3, %[count]") \
__ASM_EMIT64("add $3, %[count]") \
__ASM_EMIT ("jl 8f") \
__ASM_EMIT ("7:") \
__ASM_EMIT ("vmovss 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = ar */ \
__ASM_EMIT ("vmovss 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = ai */ \
__ASM_EMIT ("vmulss %%xmm0, %%xmm0, %%xmm4") /* xmm4 = ar*ar */ \
__ASM_EMIT (SEL("vmulss %%xmm2, %%xmm2, %%xmm6", "")) /* xmm6 = ai*ai */ \
__ASM_EMIT (SEL("vaddss %%xmm6, %%xmm4, %%xmm4", "vfmadd231ss %%xmm2, %%xmm2, %%xmm4")) /* xmm4 = R = ar*ar+ai*ai */ \
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \
__ASM_EMIT ("vxorps 0x20 + %[CC], %%xmm2, %%xmm2") /* xmm1 = -ai */ \
__ASM_EMIT ("vdivss %%xmm4, %%xmm6, %%xmm4") /* xmm4 = 1/R */ \
__ASM_EMIT ("vmulss %%xmm4, %%xmm0, %%xmm0") /* xmm0 = ar/R */ \
__ASM_EMIT ("vmulss %%xmm4, %%xmm2, %%xmm2") /* xmm1 = -ai/R */ \
__ASM_EMIT ("vmovss %%xmm0, 0x00(%[" DST "_re], %[off])") \
__ASM_EMIT ("vmovss %%xmm2, 0x00(%[" DST "_im], %[off])") \
__ASM_EMIT ("add $0x04, %[off]") \
__ASM_EMIT32("decl %[count]") \
__ASM_EMIT64("dec %[count]") \
__ASM_EMIT ("jge 7b") \
__ASM_EMIT ("8:")
void complex_rcp1(float *dst_re, float *dst_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_RCP_CORE("dst", "dst", FMA_OFF)
: [count] "+r" (count), [off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[CC] "m" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
void complex_rcp1_fma3(float *dst_re, float *dst_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_RCP_CORE("dst", "dst", FMA_ON)
: [count] "+r" (count), [off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[CC] "m" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
void complex_rcp2(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_RCP_CORE("dst", "src", FMA_OFF)
: [count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[src_re] "r" (src_re), [src_im] "r" (src_im),
[CC] "m" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
void complex_rcp2_fma3(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count)
{
IF_ARCH_X86(size_t off);
ARCH_X86_ASM
(
COMPLEX_RCP_CORE("dst", "src", FMA_ON)
: [count] X86_PGREG (count),
[off] "=&r" (off)
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im),
[src_re] "r" (src_re), [src_im] "r" (src_im),
[CC] "m" (complex_div_const)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
#undef COMPLEX_RCP_CORE
#undef FMA_OFF
#undef FMA_ON
#undef FMA_PASS
#undef FMA_SKIP
} /* namespace avx */
} /* namespace lsp */
#endif /* PRIVATE_DSP_ARCH_X86_AVX_COMPLEX_H_ */