* ac3_srfft_sse.c: accelerated SSE ac3 fft functions
*****************************************************************************
* Copyright (C) 1999, 2000, 2001 VideoLAN
- * $Id: ac3_srfft_sse.c,v 1.7 2001/10/31 11:55:53 reno Exp $
+ * $Id: ac3_srfft_sse.c,v 1.8 2001/11/09 10:02:31 reno Exp $
*
* Authors: Renaud Dartus <reno@videolan.org>
* Aaron Holtzman <aholtzma@engr.uvic.ca>
#include "ac3_imdct.h"
#include "ac3_srfft.h"
+
+static float hsqrt2_sse[] ATTR_ALIGN(16) =
+ { 0.707106781188, 0.707106781188, -0.707106781188, -0.707106781188 };
+
+static float C_1_sse[] ATTR_ALIGN(16) =
+ { -1.0, 1.0, -1.0, 1.0 };
+
+typedef struct {
+ int k;
+ void * C1;
+ } ck_sse_t;
+
+
static void fft_4_sse (complex_t *x);
static void fft_8_sse (complex_t *x);
-static void fft_asmb_sse (int k, complex_t *x, complex_t *wTB,
+static void fft_asmb_sse (ck_sse_t * ck, int k, complex_t *x, complex_t *wTB,
const complex_t *d, const complex_t *d_3);
void _M( fft_64p ) ( complex_t *a )
{
+ ck_sse_t ck;
+ ck.C1 = C_1_sse;
+
fft_8_sse(&a[0]); fft_4_sse(&a[8]); fft_4_sse(&a[12]);
- fft_asmb_sse(2, &a[0], &a[8], &delta16[0], &delta16_3[0]);
+ fft_asmb_sse(&ck, 2, &a[0], &a[8], &delta16[0], &delta16_3[0]);
fft_8_sse(&a[16]), fft_8_sse(&a[24]);
- fft_asmb_sse(4, &a[0], &a[16],&delta32[0], &delta32_3[0]);
+ fft_asmb_sse(&ck, 4, &a[0], &a[16],&delta32[0], &delta32_3[0]);
fft_8_sse(&a[32]); fft_4_sse(&a[40]); fft_4_sse(&a[44]);
- fft_asmb_sse(2, &a[32], &a[40], &delta16[0], &delta16_3[0]);
+ fft_asmb_sse(&ck, 2, &a[32], &a[40], &delta16[0], &delta16_3[0]);
fft_8_sse(&a[48]); fft_4_sse(&a[56]); fft_4_sse(&a[60]);
- fft_asmb_sse(2, &a[48], &a[56], &delta16[0], &delta16_3[0]);
+ fft_asmb_sse(&ck, 2, &a[48], &a[56], &delta16[0], &delta16_3[0]);
- fft_asmb_sse(8, &a[0], &a[32],&delta64[0], &delta64_3[0]);
+ fft_asmb_sse(&ck, 8, &a[0], &a[32],&delta64[0], &delta64_3[0]);
}
void _M( fft_128p ) ( complex_t *a )
{
+ ck_sse_t ck;
+ ck.C1 = C_1_sse;
+
fft_8_sse(&a[0]); fft_4_sse(&a[8]); fft_4_sse(&a[12]);
- fft_asmb_sse(2, &a[0], &a[8], &delta16[0], &delta16_3[0]);
+ fft_asmb_sse(&ck, 2, &a[0], &a[8], &delta16[0], &delta16_3[0]);
fft_8_sse(&a[16]), fft_8_sse(&a[24]);
- fft_asmb_sse(4, &a[0], &a[16],&delta32[0], &delta32_3[0]);
+ fft_asmb_sse(&ck, 4, &a[0], &a[16],&delta32[0], &delta32_3[0]);
fft_8_sse(&a[32]); fft_4_sse(&a[40]); fft_4_sse(&a[44]);
- fft_asmb_sse(2, &a[32], &a[40], &delta16[0], &delta16_3[0]);
+ fft_asmb_sse(&ck, 2, &a[32], &a[40], &delta16[0], &delta16_3[0]);
fft_8_sse(&a[48]); fft_4_sse(&a[56]); fft_4_sse(&a[60]);
- fft_asmb_sse(2, &a[48], &a[56], &delta16[0], &delta16_3[0]);
+ fft_asmb_sse(&ck, 2, &a[48], &a[56], &delta16[0], &delta16_3[0]);
- fft_asmb_sse(8, &a[0], &a[32],&delta64[0], &delta64_3[0]);
+ fft_asmb_sse(&ck, 8, &a[0], &a[32],&delta64[0], &delta64_3[0]);
fft_8_sse(&a[64]); fft_4_sse(&a[72]); fft_4_sse(&a[76]);
/* fft_16(&a[64]); */
- fft_asmb_sse(2, &a[64], &a[72], &delta16[0], &delta16_3[0]);
+ fft_asmb_sse(&ck, 2, &a[64], &a[72], &delta16[0], &delta16_3[0]);
fft_8_sse(&a[80]); fft_8_sse(&a[88]);
/* fft_32(&a[64]); */
- fft_asmb_sse(4, &a[64], &a[80],&delta32[0], &delta32_3[0]);
+ fft_asmb_sse(&ck, 4, &a[64], &a[80],&delta32[0], &delta32_3[0]);
fft_8_sse(&a[96]); fft_4_sse(&a[104]), fft_4_sse(&a[108]);
/* fft_16(&a[96]); */
- fft_asmb_sse(2, &a[96], &a[104], &delta16[0], &delta16_3[0]);
+ fft_asmb_sse(&ck, 2, &a[96], &a[104], &delta16[0], &delta16_3[0]);
fft_8_sse(&a[112]), fft_8_sse(&a[120]);
/* fft_32(&a[96]); */
- fft_asmb_sse(4, &a[96], &a[112], &delta32[0], &delta32_3[0]);
+ fft_asmb_sse(&ck, 4, &a[96], &a[112], &delta32[0], &delta32_3[0]);
/* fft_128(&a[0]); */
- fft_asmb_sse(16, &a[0], &a[64], &delta128[0], &delta128_3[0]);
+ fft_asmb_sse(&ck, 16, &a[0], &a[64], &delta128[0], &delta128_3[0]);
}
-static float hsqrt2_sse[] ATTR_ALIGN(16) =
- { 0.707106781188, 0.707106781188, -0.707106781188, -0.707106781188 };
-
-static float C_1_sse[] ATTR_ALIGN(16) =
- { -1.0, 1.0, -1.0, 1.0 };
-
static void fft_4_sse (complex_t *x)
{
__asm__ __volatile__ (
: "a" (x), "c" (hsqrt2_sse), "d" (C_1_sse));
}
-static void fft_asmb_sse (int k, complex_t *x, complex_t *wTB,
+static void fft_asmb_sse (ck_sse_t * ck, int k, complex_t *x, complex_t *wTB,
const complex_t *d, const complex_t *d_3)
{
+ ck->k = k;
+
__asm__ __volatile__ (
".align 16\n"
"pushl %%ebp\n"
"movl %%esp, %%ebp\n"
- "subl $4, %%esp\n"
+ "subl $8, %%esp\n"
"pushl %%eax\n"
"pushl %%ebx\n"
"pushl %%esi\n"
"pushl %%edi\n"
- "movl %%ecx, -4(%%ebp)\n" /* k */
- "shll $4, %%ecx\n" /* 16k */ ///
+ "movl 4(%%ecx), %%ebx\n"
+ "movl %%ebx, -4(%%ebp)\n"
+ "movl (%%ecx), %%ecx\n"
+
+ "movl %%ecx, -8(%%ebp)\n" /* k */
"addl $8, %%edx\n"
"addl $8, %%esi\n"
+ "shll $4, %%ecx\n" /* 16k */
/* TRANSZERO and TRANS */
".align 16\n"
"movhlps %%xmm5, %%xmm7\n" /* wT[1].im * d[1].im | wT[1].re * d[1].im */
"movlhps %%xmm6, %%xmm5\n" /* wB[1].im * d3[1].re | wB[1].re * d3[1].re | wT[1].im * d[1].re | wT[1].re * d[1].re */
"shufps $0xb1, %%xmm6, %%xmm7\n" /* wB[1].re * d3[1].im | wB[i].im * d3[1].im | wT[1].re * d[1].im | wT[1].im * d[1].im */
+ "movl -4(%%ebp), %%ebx\n"
"movaps (%%ebx), %%xmm4\n"
"mulps %%xmm4, %%xmm7\n"
"addps %%xmm7, %%xmm5\n" /* wB[1] * d3[1] | wT[1] * d[1] */
"movlhps %%xmm5, %%xmm1\n" /* d[1] * wT[1] | wT[0] */
"shufps $0xe4, %%xmm5, %%xmm2\n" /* d3[1] * wB[1] | wB[0] */
"movaps %%xmm1, %%xmm3\n" /* d[1] * wT[1] | wT[0] */
- "leal (%%eax, %%ecx, 2), %%esp\n"
+ "leal (%%eax, %%ecx, 2), %%ebx\n"
"addps %%xmm2, %%xmm1\n" /* u */
"subps %%xmm2, %%xmm3\n" /* v */
"mulps %%xmm4, %%xmm3\n"
"addps %%xmm3, %%xmm5\n"
"subps %%xmm3, %%xmm6\n"
"movaps %%xmm0, (%%eax)\n"
- "movaps %%xmm2, (%%esp)\n"
+ "movaps %%xmm2, (%%ebx)\n"
"movaps %%xmm5, (%%eax, %%ecx)\n"
- "movaps %%xmm6, (%%esp, %%ecx)\n"
+ "movaps %%xmm6, (%%ebx, %%ecx)\n"
"addl $16, %%eax\n"
"addl $16, %%edi\n"
"addl $8, %%edx\n"
"addl $8, %%esi\n"
- "decl -4(%%ebp)\n"
+ "decl -8(%%ebp)\n"
".align 16\n"
".loop:\n"
"mulps %%xmm5, %%xmm4\n" /* wB[0].im * d3[0].im | wB[0].re * d3[0].im | wB[0].im * d3[0].re | wB[0].re * d3[0].re */
"mulps %%xmm7, %%xmm6\n" /* wB[1].im * d3[1].im | wB[1].re * d3[1].im | wB[1].im * d3[1].re | wB[1].re * d3[1].re */
"shufps $0xb1, %%xmm2, %%xmm1\n" /* d[1].im * wT[1].re | d[1].im * wT[1].im | d[0].im * wT[0].re | d[0].im * wT[0].im */
+ "movl -4(%%ebp), %%ebx\n"
"movaps (%%ebx), %%xmm3\n" /* 1.0 | -1.0 | 1.0 | -1.0 */
"movhlps %%xmm4, %%xmm5\n" /* wB[0].im * d3[0].im | wB[0].re * d3[0].im */
"addps %%xmm4, %%xmm0\n" /* u */
"subps %%xmm4, %%xmm1\n" /* v */
"movaps (%%eax), %%xmm6\n" /* x[1] | x[0] */
- "leal (%%eax, %%ecx, 2), %%esp\n"
+ "leal (%%eax, %%ecx, 2), %%ebx\n"
"mulps %%xmm3, %%xmm1\n"
"addl $16, %%edi\n"
"addl $16, %%esi\n"
"addps %%xmm0, %%xmm6\n"
"subps %%xmm0, %%xmm2\n"
"movaps %%xmm6, (%%eax)\n"
- "movaps %%xmm2, (%%esp)\n"
+ "movaps %%xmm2, (%%ebx)\n"
"addps %%xmm1, %%xmm7\n"
"subps %%xmm1, %%xmm4\n"
"addl $16, %%edx\n"
"movaps %%xmm7, (%%eax, %%ecx)\n"
- "movaps %%xmm4, (%%esp, %%ecx)\n"
+ "movaps %%xmm4, (%%ebx, %%ecx)\n"
"addl $16, %%eax\n"
- "decl -4(%%ebp)\n"
+ "decl -8(%%ebp)\n"
"jnz .loop\n"
".align 16\n"
"popl %%ebx\n"
"popl %%eax\n"
- "addl $4, %%esp\n"
-
+ "addl $8, %%esp\n"
+
"leave\n"
- : "=c" (k), "=a" (x), "=D" (wTB)
- : "c" (k), "a" (x), "D" (wTB), "d" (d), "S" (d_3), "b" (C_1_sse) );
+ : "=a" (x), "=D" (wTB)
+ : "c" (ck), "a" (x), "D" (wTB), "d" (d), "S" (d_3) );
}
projects / vlc / commitdiff