/*
* 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_SSE_IMPL
#error "This header should not be included directly"
#endif /* PRIVATE_DSP_ARCH_X86_SSE_IMPL */
namespace lsp
{
namespace sse
{
static inline void FFT_SCRAMBLE_SELF_DIRECT_NAME(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank)
{
// Calculate number of items
size_t items = (1 << rank) - 1;
for (size_t i = 1; i < items; ++i)
{
size_t j = reverse_bits(FFT_TYPE(i), rank); /* Reverse the order of the bits */
if (i >= j)
continue;
/* Copy the values from the reversed position */
ARCH_X86_ASM
(
__ASM_EMIT("movss (%[dst_re], %[i], 4), %%xmm0")
__ASM_EMIT("movss (%[dst_im], %[i], 4), %%xmm1")
__ASM_EMIT("movss (%[dst_re], %[j], 4), %%xmm2")
__ASM_EMIT("movss (%[dst_im], %[j], 4), %%xmm3")
__ASM_EMIT("movss %%xmm2, (%[dst_re], %[i], 4)")
__ASM_EMIT("movss %%xmm3, (%[dst_im], %[i], 4)")
__ASM_EMIT("movss %%xmm0, (%[dst_re], %[j], 4)")
__ASM_EMIT("movss %%xmm1, (%[dst_im], %[j], 4)")
:
: [dst_re] "r"(dst_re), [dst_im] "r"(dst_im), [i] "r"(i), [j] "r"(j)
: "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3"
);
}
// Perform butterfly 8x
items = 1 << (rank - 3);
// Perform 4-element butterflies
ARCH_X86_ASM
(
/* Loop */
__ASM_EMIT(".align 16")
__ASM_EMIT("1:")
/* Load data to registers */
__ASM_EMIT("movups 0x00(%[dst_re]), %%xmm0") /* xmm0 = r0 r1 r2 r3 */
__ASM_EMIT("movups 0x10(%[dst_re]), %%xmm4") /* xmm4 = r4 r5 r6 r7 */
__ASM_EMIT("movups 0x00(%[dst_im]), %%xmm2") /* xmm2 = i0 i1 i2 i3 */
__ASM_EMIT("movups 0x10(%[dst_im]), %%xmm6") /* xmm6 = i4 i5 i6 i7 */
/* 1st-order 4x butterfly */
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = r0 r1 r2 r3 */
__ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = i0 i1 i2 i3 */
__ASM_EMIT("shufps $0x88, %%xmm4, %%xmm0") /* xmm0 = r0 r2 r4 r6 */
__ASM_EMIT("shufps $0x88, %%xmm6, %%xmm2") /* xmm2 = i0 i2 i4 i6 */
__ASM_EMIT("shufps $0xdd, %%xmm4, %%xmm1") /* xmm1 = r1 r3 r5 r7 */
__ASM_EMIT("shufps $0xdd, %%xmm6, %%xmm3") /* xmm3 = i1 i3 i5 i7 */
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = r0 r2 r4 r6 */
__ASM_EMIT("movaps %%xmm2, %%xmm6") /* xmm5 = i0 i2 i4 i6 */
__ASM_EMIT("addps %%xmm1, %%xmm0") /* xmm0 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6' */
__ASM_EMIT("addps %%xmm3, %%xmm2") /* xmm2 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6' */
__ASM_EMIT("subps %%xmm1, %%xmm4") /* xmm4 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7' */
__ASM_EMIT("subps %%xmm3, %%xmm6") /* xmm6 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7' */
/* 2nd-order 4x butterfly */
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = r0' r2' r4' r6' */
__ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = i0' i2' i4' i6' */
__ASM_EMIT("shufps $0x88, %%xmm4, %%xmm0") /* xmm0 = r0' r4' r1' r5' */
__ASM_EMIT("shufps $0xdd, %%xmm6, %%xmm1") /* xmm1 = r2' r6' i3' i7' */
__ASM_EMIT("shufps $0xdd, %%xmm4, %%xmm3") /* xmm3 = i2' i6' r3' r7' */
__ASM_EMIT("shufps $0x88, %%xmm6, %%xmm2") /* xmm2 = i0' i4' i1' i5' */
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = r0' r4' r1' r5' */
__ASM_EMIT("movaps %%xmm2, %%xmm6") /* xmm5 = i0' i4' i1' i5' */
__ASM_EMIT("addps %%xmm1, %%xmm0") /* xmm0 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r1" r5" */
__ASM_EMIT("addps %%xmm3, %%xmm2") /* xmm2 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i3" i7" */
__ASM_EMIT("subps %%xmm1, %%xmm4") /* xmm4 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r3" r7" */
__ASM_EMIT("subps %%xmm3, %%xmm6") /* xmm6 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i1" i5" */
/* Reorder and store */
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = r0" r4" r1" r5" */
__ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = i0" i4" i3" i7" */
__ASM_EMIT("shufps $0x88, %%xmm4, %%xmm0") /* xmm0 = r0" r1" r2" r3" */
__ASM_EMIT("shufps $0x88, %%xmm6, %%xmm2") /* xmm2 = i0" i3" i2" i1" */
__ASM_EMIT("shufps $0xdd, %%xmm4, %%xmm1") /* xmm1 = r4" r5" r6" r7" */
__ASM_EMIT("shufps $0xdd, %%xmm6, %%xmm3") /* xmm3 = i4" i7" i6" i5" */
__ASM_EMIT("shufps $0x6c, %%xmm2, %%xmm2") /* xmm2 = i0" i1" i2" i3" */
__ASM_EMIT("shufps $0x6c, %%xmm3, %%xmm3") /* xmm3 = i4" i5" i6" i7" */
__ASM_EMIT("movups %%xmm0, 0x00(%[dst_re])")
__ASM_EMIT("movups %%xmm1, 0x10(%[dst_re])")
__ASM_EMIT("movups %%xmm2, 0x00(%[dst_im])")
__ASM_EMIT("movups %%xmm3, 0x10(%[dst_im])")
/* Move pointers */
__ASM_EMIT("add $0x20, %[dst_re]")
__ASM_EMIT("add $0x20, %[dst_im]")
/* Repeat cycle */
__ASM_EMIT("dec %[items]")
__ASM_EMIT("jnz 1b")
__ASM_EMIT("2:")
: [dst_re] "+r"(dst_re), [dst_im] "+r"(dst_im), [items] "+r"(items)
:
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2"
);
}
static inline void FFT_SCRAMBLE_COPY_DIRECT_NAME(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank)
{
size_t regs = 1 << rank;
for (size_t i=0; i= j)
continue;
/* Copy the values from the reversed position */
ARCH_X86_ASM
(
__ASM_EMIT("movss (%[dst_re], %[i], 4), %%xmm0")
__ASM_EMIT("movss (%[dst_im], %[i], 4), %%xmm1")
__ASM_EMIT("movss (%[dst_re], %[j], 4), %%xmm2")
__ASM_EMIT("movss (%[dst_im], %[j], 4), %%xmm3")
__ASM_EMIT("movss %%xmm2, (%[dst_re], %[i], 4)")
__ASM_EMIT("movss %%xmm3, (%[dst_im], %[i], 4)")
__ASM_EMIT("movss %%xmm0, (%[dst_re], %[j], 4)")
__ASM_EMIT("movss %%xmm1, (%[dst_im], %[j], 4)")
:
: [dst_re] "r"(dst_re), [dst_im] "r"(dst_im), [i] "r"(i), [j] "r"(j)
: "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3"
);
}
// Perform butterfly 4x
items = 1 << (rank - 3);
// Perform 4-element butterflies
ARCH_X86_ASM
(
/* Loop */
__ASM_EMIT(".align 16")
__ASM_EMIT("1:")
/* Load data to registers */
__ASM_EMIT("movups 0x00(%[dst_re]), %%xmm0") /* xmm0 = r0 r1 r2 r3 */
__ASM_EMIT("movups 0x10(%[dst_re]), %%xmm4") /* xmm4 = r4 r5 r6 r7 */
__ASM_EMIT("movups 0x00(%[dst_im]), %%xmm2") /* xmm2 = i0 i1 i2 i3 */
__ASM_EMIT("movups 0x10(%[dst_im]), %%xmm6") /* xmm6 = i4 i5 i6 i7 */
/* 1st-order 4x butterfly */
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = r0 r1 r2 r3 */
__ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = i0 i1 i2 i3 */
__ASM_EMIT("shufps $0x88, %%xmm4, %%xmm0") /* xmm0 = r0 r2 r4 r6 */
__ASM_EMIT("shufps $0x88, %%xmm6, %%xmm2") /* xmm2 = i0 i2 i4 i6 */
__ASM_EMIT("shufps $0xdd, %%xmm4, %%xmm1") /* xmm1 = r1 r3 r5 r7 */
__ASM_EMIT("shufps $0xdd, %%xmm6, %%xmm3") /* xmm3 = i1 i3 i5 i7 */
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = r0 r2 r4 r6 */
__ASM_EMIT("movaps %%xmm2, %%xmm6") /* xmm5 = i0 i2 i4 i6 */
__ASM_EMIT("addps %%xmm1, %%xmm0") /* xmm0 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6' */
__ASM_EMIT("addps %%xmm3, %%xmm2") /* xmm2 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6' */
__ASM_EMIT("subps %%xmm1, %%xmm4") /* xmm4 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7' */
__ASM_EMIT("subps %%xmm3, %%xmm6") /* xmm6 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7' */
/* 2nd-order 4x butterfly */
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = r0' r2' r4' r6' */
__ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = i0' i2' i4' i6' */
__ASM_EMIT("shufps $0x88, %%xmm4, %%xmm0") /* xmm0 = r0' r4' r1' r5' */
__ASM_EMIT("shufps $0xdd, %%xmm6, %%xmm1") /* xmm1 = r2' r6' i3' i7' */
__ASM_EMIT("shufps $0xdd, %%xmm4, %%xmm3") /* xmm3 = i2' i6' r3' r7' */
__ASM_EMIT("shufps $0x88, %%xmm6, %%xmm2") /* xmm2 = i0' i4' i1' i5' */
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = r0' r4' r1' r5' */
__ASM_EMIT("movaps %%xmm2, %%xmm6") /* xmm5 = i0' i4' i1' i5' */
__ASM_EMIT("addps %%xmm1, %%xmm0") /* xmm0 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r3" r7" */
__ASM_EMIT("addps %%xmm3, %%xmm2") /* xmm2 = i0'+i2' i4'+i6' i1'+i3' i5'+i7' = i0" i4" i1" i5" */
__ASM_EMIT("subps %%xmm1, %%xmm4") /* xmm4 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r1" r5" */
__ASM_EMIT("subps %%xmm3, %%xmm6") /* xmm6 = i0'-i2' i4'-i6' i1'-i3' i5'-i7' = i2" i6" i3" i7" */
/* Reorder and store */
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = r0" r4" r3" r7" */
__ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = i0" i4" i1" i5" */
__ASM_EMIT("shufps $0x88, %%xmm4, %%xmm0") /* xmm0 = r0" r3" r2" r1" */
__ASM_EMIT("shufps $0x88, %%xmm6, %%xmm2") /* xmm2 = i0" i1" i2" i3" */
__ASM_EMIT("shufps $0xdd, %%xmm4, %%xmm1") /* xmm1 = r4" r7" r6" r5" */
__ASM_EMIT("shufps $0xdd, %%xmm6, %%xmm3") /* xmm3 = i4" i5" i6" i7" */
__ASM_EMIT("shufps $0x6c, %%xmm0, %%xmm0") /* xmm0 = r0" r1" r2" r3" */
__ASM_EMIT("shufps $0x6c, %%xmm1, %%xmm1") /* xmm1 = r4" r5" r6" r7" */
__ASM_EMIT("movups %%xmm0, 0x00(%[dst_re])")
__ASM_EMIT("movups %%xmm1, 0x10(%[dst_re])")
__ASM_EMIT("movups %%xmm2, 0x00(%[dst_im])")
__ASM_EMIT("movups %%xmm3, 0x10(%[dst_im])")
/* Move pointers */
__ASM_EMIT("add $0x20, %[dst_re]")
__ASM_EMIT("add $0x20, %[dst_im]")
/* Repeat cycle */
__ASM_EMIT("dec %[items]")
__ASM_EMIT("jnz 1b")
__ASM_EMIT("2:")
: [dst_re] "+r"(dst_re), [dst_im] "+r"(dst_im), [items] "+r"(items)
:
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2"
);
}
static inline void FFT_SCRAMBLE_COPY_REVERSE_NAME(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank)
{
size_t regs = 1 << rank;
for (size_t i=0; i