aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec
diff options
context:
space:
mode:
authorRonald S. Bultje <rsbultje@gmail.com>2013-03-10 15:50:53 -0700
committerMichael Niedermayer <michaelni@gmx.at>2013-03-13 04:02:30 +0100
commit89f16ded9b747de8870ddee87176e105a5d75dcd (patch)
tree0a1b753a3cb3fe74c9bddab2cf924156e09cc4ab /libavcodec
parent3ced55d51c2e65b37e50d500dff88bcd80e01b9c (diff)
downloadffmpeg-89f16ded9b747de8870ddee87176e105a5d75dcd.tar.gz
Move ppc half-pel assembly from dsputil to hpeldsp.
Diffstat (limited to 'libavcodec')
-rw-r--r--libavcodec/hpeldsp.c2
-rw-r--r--libavcodec/ppc/Makefile1
-rw-r--r--libavcodec/ppc/dsputil_altivec.c421
-rw-r--r--libavcodec/ppc/hpeldsp_altivec.c464
4 files changed, 467 insertions, 421 deletions
diff --git a/libavcodec/hpeldsp.c b/libavcodec/hpeldsp.c
index de3cd76747..4bc7a8064b 100644
--- a/libavcodec/hpeldsp.c
+++ b/libavcodec/hpeldsp.c
@@ -58,7 +58,9 @@ av_cold void ff_hpeldsp_init(HpelDSPContext* c, int flags)
if (ARCH_ARM) ff_hpeldsp_init_arm (c, flags);
if (HAVE_VIS) ff_hpeldsp_init_vis (c, flags);
if (ARCH_ALPHA) ff_hpeldsp_init_alpha (c, flags);
+#endif
if (ARCH_PPC) ff_hpeldsp_init_ppc (c, flags);
+#if 0
if (ARCH_SH4) ff_hpeldsp_init_sh4 (c, flags);
if (ARCH_BFIN) ff_hpeldsp_init_bfin (c, flags);
#endif
diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile
index a2ce9ebafb..febbb0a231 100644
--- a/libavcodec/ppc/Makefile
+++ b/libavcodec/ppc/Makefile
@@ -3,6 +3,7 @@ OBJS += ppc/dsputil_ppc.o \
OBJS-$(CONFIG_H264CHROMA) += ppc/h264chroma_init.o
OBJS-$(CONFIG_H264QPEL) += ppc/h264_qpel.o
+OBJS-$(CONFIG_HPELDSP) += ppc/hpeldsp_altivec.o
OBJS-$(CONFIG_VORBIS_DECODER) += ppc/vorbisdsp_altivec.o
OBJS-$(CONFIG_VP3DSP) += ppc/vp3dsp_altivec.o
diff --git a/libavcodec/ppc/dsputil_altivec.c b/libavcodec/ppc/dsputil_altivec.c
index 75a6a6ab91..f36e394fb2 100644
--- a/libavcodec/ppc/dsputil_altivec.c
+++ b/libavcodec/ppc/dsputil_altivec.c
@@ -607,354 +607,6 @@ static void add_bytes_altivec(uint8_t *dst, uint8_t *src, int w) {
}
}
-/* next one assumes that ((line_size % 16) == 0) */
-void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
-{
- register vector unsigned char pixelsv1, pixelsv2;
- register vector unsigned char pixelsv1B, pixelsv2B;
- register vector unsigned char pixelsv1C, pixelsv2C;
- register vector unsigned char pixelsv1D, pixelsv2D;
-
- register vector unsigned char perm = vec_lvsl(0, pixels);
- int i;
- register ptrdiff_t line_size_2 = line_size << 1;
- register ptrdiff_t line_size_3 = line_size + line_size_2;
- register ptrdiff_t line_size_4 = line_size << 2;
-
-// hand-unrolling the loop by 4 gains about 15%
-// mininum execution time goes from 74 to 60 cycles
-// it's faster than -funroll-loops, but using
-// -funroll-loops w/ this is bad - 74 cycles again.
-// all this is on a 7450, tuning for the 7450
- for (i = 0; i < h; i += 4) {
- pixelsv1 = vec_ld( 0, pixels);
- pixelsv2 = vec_ld(15, pixels);
- pixelsv1B = vec_ld(line_size, pixels);
- pixelsv2B = vec_ld(15 + line_size, pixels);
- pixelsv1C = vec_ld(line_size_2, pixels);
- pixelsv2C = vec_ld(15 + line_size_2, pixels);
- pixelsv1D = vec_ld(line_size_3, pixels);
- pixelsv2D = vec_ld(15 + line_size_3, pixels);
- vec_st(vec_perm(pixelsv1, pixelsv2, perm),
- 0, (unsigned char*)block);
- vec_st(vec_perm(pixelsv1B, pixelsv2B, perm),
- line_size, (unsigned char*)block);
- vec_st(vec_perm(pixelsv1C, pixelsv2C, perm),
- line_size_2, (unsigned char*)block);
- vec_st(vec_perm(pixelsv1D, pixelsv2D, perm),
- line_size_3, (unsigned char*)block);
- pixels+=line_size_4;
- block +=line_size_4;
- }
-}
-
-/* next one assumes that ((line_size % 16) == 0) */
-#define op_avg(a,b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
-void ff_avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
-{
- register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
- register vector unsigned char perm = vec_lvsl(0, pixels);
- int i;
-
- for (i = 0; i < h; i++) {
- pixelsv1 = vec_ld( 0, pixels);
- pixelsv2 = vec_ld(16,pixels);
- blockv = vec_ld(0, block);
- pixelsv = vec_perm(pixelsv1, pixelsv2, perm);
- blockv = vec_avg(blockv,pixelsv);
- vec_st(blockv, 0, (unsigned char*)block);
- pixels+=line_size;
- block +=line_size;
- }
-}
-
-/* next one assumes that ((line_size % 8) == 0) */
-static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
-{
- register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
- int i;
-
- for (i = 0; i < h; i++) {
- /* block is 8 bytes-aligned, so we're either in the
- left block (16 bytes-aligned) or in the right block (not) */
- int rightside = ((unsigned long)block & 0x0000000F);
-
- blockv = vec_ld(0, block);
- pixelsv1 = vec_ld( 0, pixels);
- pixelsv2 = vec_ld(16, pixels);
- pixelsv = vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels));
-
- if (rightside) {
- pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1));
- } else {
- pixelsv = vec_perm(blockv, pixelsv, vcprm(s0,s1,2,3));
- }
-
- blockv = vec_avg(blockv, pixelsv);
-
- vec_st(blockv, 0, block);
-
- pixels += line_size;
- block += line_size;
- }
-}
-
-/* next one assumes that ((line_size % 8) == 0) */
-static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
-{
- register int i;
- register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
- register vector unsigned char blockv, temp1, temp2;
- register vector unsigned short pixelssum1, pixelssum2, temp3;
- register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
- register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
-
- temp1 = vec_ld(0, pixels);
- temp2 = vec_ld(16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
- if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
- }
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
- pixelssum1 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- pixelssum1 = vec_add(pixelssum1, vctwo);
-
- for (i = 0; i < h ; i++) {
- int rightside = ((unsigned long)block & 0x0000000F);
- blockv = vec_ld(0, block);
-
- temp1 = vec_ld(line_size, pixels);
- temp2 = vec_ld(line_size + 16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
- if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
- }
-
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
- pixelssum2 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- temp3 = vec_add(pixelssum1, pixelssum2);
- temp3 = vec_sra(temp3, vctwo);
- pixelssum1 = vec_add(pixelssum2, vctwo);
- pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
-
- if (rightside) {
- blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
- } else {
- blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
- }
-
- vec_st(blockv, 0, block);
-
- block += line_size;
- pixels += line_size;
- }
-}
-
-/* next one assumes that ((line_size % 8) == 0) */
-static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
-{
- register int i;
- register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
- register vector unsigned char blockv, temp1, temp2;
- register vector unsigned short pixelssum1, pixelssum2, temp3;
- register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
- register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
- register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
-
- temp1 = vec_ld(0, pixels);
- temp2 = vec_ld(16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
- if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
- }
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
- pixelssum1 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- pixelssum1 = vec_add(pixelssum1, vcone);
-
- for (i = 0; i < h ; i++) {
- int rightside = ((unsigned long)block & 0x0000000F);
- blockv = vec_ld(0, block);
-
- temp1 = vec_ld(line_size, pixels);
- temp2 = vec_ld(line_size + 16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
- if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
- }
-
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
- pixelssum2 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- temp3 = vec_add(pixelssum1, pixelssum2);
- temp3 = vec_sra(temp3, vctwo);
- pixelssum1 = vec_add(pixelssum2, vcone);
- pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
-
- if (rightside) {
- blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
- } else {
- blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
- }
-
- vec_st(blockv, 0, block);
-
- block += line_size;
- pixels += line_size;
- }
-}
-
-/* next one assumes that ((line_size % 16) == 0) */
-static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
-{
- register int i;
- register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
- register vector unsigned char blockv, temp1, temp2;
- register vector unsigned short temp3, temp4,
- pixelssum1, pixelssum2, pixelssum3, pixelssum4;
- register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
- register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
-
- temp1 = vec_ld(0, pixels);
- temp2 = vec_ld(16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
- if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
- }
- pixelsv3 = vec_mergel(vczero, pixelsv1);
- pixelsv4 = vec_mergel(vczero, pixelsv2);
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
- pixelssum3 = vec_add((vector unsigned short)pixelsv3,
- (vector unsigned short)pixelsv4);
- pixelssum3 = vec_add(pixelssum3, vctwo);
- pixelssum1 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- pixelssum1 = vec_add(pixelssum1, vctwo);
-
- for (i = 0; i < h ; i++) {
- blockv = vec_ld(0, block);
-
- temp1 = vec_ld(line_size, pixels);
- temp2 = vec_ld(line_size + 16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
- if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
- }
-
- pixelsv3 = vec_mergel(vczero, pixelsv1);
- pixelsv4 = vec_mergel(vczero, pixelsv2);
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
-
- pixelssum4 = vec_add((vector unsigned short)pixelsv3,
- (vector unsigned short)pixelsv4);
- pixelssum2 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- temp4 = vec_add(pixelssum3, pixelssum4);
- temp4 = vec_sra(temp4, vctwo);
- temp3 = vec_add(pixelssum1, pixelssum2);
- temp3 = vec_sra(temp3, vctwo);
-
- pixelssum3 = vec_add(pixelssum4, vctwo);
- pixelssum1 = vec_add(pixelssum2, vctwo);
-
- blockv = vec_packsu(temp3, temp4);
-
- vec_st(blockv, 0, block);
-
- block += line_size;
- pixels += line_size;
- }
-}
-
-/* next one assumes that ((line_size % 16) == 0) */
-static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
-{
- register int i;
- register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
- register vector unsigned char blockv, temp1, temp2;
- register vector unsigned short temp3, temp4,
- pixelssum1, pixelssum2, pixelssum3, pixelssum4;
- register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
- register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
- register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
-
- temp1 = vec_ld(0, pixels);
- temp2 = vec_ld(16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
- if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
- }
- pixelsv3 = vec_mergel(vczero, pixelsv1);
- pixelsv4 = vec_mergel(vczero, pixelsv2);
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
- pixelssum3 = vec_add((vector unsigned short)pixelsv3,
- (vector unsigned short)pixelsv4);
- pixelssum3 = vec_add(pixelssum3, vcone);
- pixelssum1 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- pixelssum1 = vec_add(pixelssum1, vcone);
-
- for (i = 0; i < h ; i++) {
- blockv = vec_ld(0, block);
-
- temp1 = vec_ld(line_size, pixels);
- temp2 = vec_ld(line_size + 16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
- if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
- }
-
- pixelsv3 = vec_mergel(vczero, pixelsv1);
- pixelsv4 = vec_mergel(vczero, pixelsv2);
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
-
- pixelssum4 = vec_add((vector unsigned short)pixelsv3,
- (vector unsigned short)pixelsv4);
- pixelssum2 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- temp4 = vec_add(pixelssum3, pixelssum4);
- temp4 = vec_sra(temp4, vctwo);
- temp3 = vec_add(pixelssum1, pixelssum2);
- temp3 = vec_sra(temp3, vctwo);
-
- pixelssum3 = vec_add(pixelssum4, vcone);
- pixelssum1 = vec_add(pixelssum2, vcone);
-
- blockv = vec_packsu(temp3, temp4);
-
- vec_st(blockv, 0, block);
-
- block += line_size;
- pixels += line_size;
- }
-}
-
static int hadamard8_diff8x8_altivec(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){
int sum;
register const vector unsigned char vzero =
@@ -1284,69 +936,6 @@ static int hadamard8_diff16_altivec(/*MpegEncContext*/ void *s, uint8_t *dst, ui
return score;
}
-/* next one assumes that ((line_size % 8) == 0) */
-static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
-{
- register int i;
- register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
- register vector unsigned char blockv, temp1, temp2, blocktemp;
- register vector unsigned short pixelssum1, pixelssum2, temp3;
-
- register const vector unsigned char vczero = (const vector unsigned char)
- vec_splat_u8(0);
- register const vector unsigned short vctwo = (const vector unsigned short)
- vec_splat_u16(2);
-
- temp1 = vec_ld(0, pixels);
- temp2 = vec_ld(16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
- if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
- }
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
- pixelssum1 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- pixelssum1 = vec_add(pixelssum1, vctwo);
-
- for (i = 0; i < h ; i++) {
- int rightside = ((unsigned long)block & 0x0000000F);
- blockv = vec_ld(0, block);
-
- temp1 = vec_ld(line_size, pixels);
- temp2 = vec_ld(line_size + 16, pixels);
- pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
- if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
- pixelsv2 = temp2;
- } else {
- pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
- }
-
- pixelsv1 = vec_mergeh(vczero, pixelsv1);
- pixelsv2 = vec_mergeh(vczero, pixelsv2);
- pixelssum2 = vec_add((vector unsigned short)pixelsv1,
- (vector unsigned short)pixelsv2);
- temp3 = vec_add(pixelssum1, pixelssum2);
- temp3 = vec_sra(temp3, vctwo);
- pixelssum1 = vec_add(pixelssum2, vctwo);
- pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
-
- if (rightside) {
- blocktemp = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
- } else {
- blocktemp = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
- }
-
- blockv = vec_avg(blocktemp, blockv);
- vec_st(blockv, 0, block);
-
- block += line_size;
- pixels += line_size;
- }
-}
-
av_cold void ff_dsputil_init_altivec(DSPContext *c, AVCodecContext *avctx)
{
const int high_bit_depth = avctx->bits_per_raw_sample > 8;
@@ -1367,16 +956,6 @@ av_cold void ff_dsputil_init_altivec(DSPContext *c, AVCodecContext *avctx)
if (!high_bit_depth) {
c->get_pixels = get_pixels_altivec;
c->clear_block = clear_block_altivec;
- c->put_pixels_tab[0][0] = ff_put_pixels16_altivec;
- /* the two functions do the same thing, so use the same code */
- c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_altivec;
- c->avg_pixels_tab[0][0] = ff_avg_pixels16_altivec;
- c->avg_pixels_tab[1][0] = avg_pixels8_altivec;
- c->avg_pixels_tab[1][3] = avg_pixels8_xy2_altivec;
- c->put_pixels_tab[1][3] = put_pixels8_xy2_altivec;
- c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_altivec;
- c->put_pixels_tab[0][3] = put_pixels16_xy2_altivec;
- c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_altivec;
}
c->hadamard8_diff[0] = hadamard8_diff16_altivec;
diff --git a/libavcodec/ppc/hpeldsp_altivec.c b/libavcodec/ppc/hpeldsp_altivec.c
new file mode 100644
index 0000000000..4309d39851
--- /dev/null
+++ b/libavcodec/ppc/hpeldsp_altivec.c
@@ -0,0 +1,464 @@
+/*
+ * Copyright (c) 2002 Brian Foley
+ * Copyright (c) 2002 Dieter Shirley
+ * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+#include "libavutil/cpu.h"
+#include "libavcodec/hpeldsp.h"
+
+#if HAVE_ALTIVEC
+#if HAVE_ALTIVEC_H
+#include <altivec.h>
+#endif
+#include "libavutil/ppc/types_altivec.h"
+#include "libavutil/ppc/util_altivec.h"
+#include "dsputil_altivec.h"
+
+/* next one assumes that ((line_size % 16) == 0) */
+void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
+{
+ register vector unsigned char pixelsv1, pixelsv2;
+ register vector unsigned char pixelsv1B, pixelsv2B;
+ register vector unsigned char pixelsv1C, pixelsv2C;
+ register vector unsigned char pixelsv1D, pixelsv2D;
+
+ register vector unsigned char perm = vec_lvsl(0, pixels);
+ int i;
+ register ptrdiff_t line_size_2 = line_size << 1;
+ register ptrdiff_t line_size_3 = line_size + line_size_2;
+ register ptrdiff_t line_size_4 = line_size << 2;
+
+// hand-unrolling the loop by 4 gains about 15%
+// mininum execution time goes from 74 to 60 cycles
+// it's faster than -funroll-loops, but using
+// -funroll-loops w/ this is bad - 74 cycles again.
+// all this is on a 7450, tuning for the 7450
+ for (i = 0; i < h; i += 4) {
+ pixelsv1 = vec_ld( 0, pixels);
+ pixelsv2 = vec_ld(15, pixels);
+ pixelsv1B = vec_ld(line_size, pixels);
+ pixelsv2B = vec_ld(15 + line_size, pixels);
+ pixelsv1C = vec_ld(line_size_2, pixels);
+ pixelsv2C = vec_ld(15 + line_size_2, pixels);
+ pixelsv1D = vec_ld(line_size_3, pixels);
+ pixelsv2D = vec_ld(15 + line_size_3, pixels);
+ vec_st(vec_perm(pixelsv1, pixelsv2, perm),
+ 0, (unsigned char*)block);
+ vec_st(vec_perm(pixelsv1B, pixelsv2B, perm),
+ line_size, (unsigned char*)block);
+ vec_st(vec_perm(pixelsv1C, pixelsv2C, perm),
+ line_size_2, (unsigned char*)block);
+ vec_st(vec_perm(pixelsv1D, pixelsv2D, perm),
+ line_size_3, (unsigned char*)block);
+ pixels+=line_size_4;
+ block +=line_size_4;
+ }
+}
+
+/* next one assumes that ((line_size % 16) == 0) */
+#define op_avg(a,b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
+void ff_avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
+{
+ register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
+ register vector unsigned char perm = vec_lvsl(0, pixels);
+ int i;
+
+ for (i = 0; i < h; i++) {
+ pixelsv1 = vec_ld( 0, pixels);
+ pixelsv2 = vec_ld(16,pixels);
+ blockv = vec_ld(0, block);
+ pixelsv = vec_perm(pixelsv1, pixelsv2, perm);
+ blockv = vec_avg(blockv,pixelsv);
+ vec_st(blockv, 0, (unsigned char*)block);
+ pixels+=line_size;
+ block +=line_size;
+ }
+}
+
+/* next one assumes that ((line_size % 8) == 0) */
+static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
+{
+ register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
+ int i;
+
+ for (i = 0; i < h; i++) {
+ /* block is 8 bytes-aligned, so we're either in the
+ left block (16 bytes-aligned) or in the right block (not) */
+ int rightside = ((unsigned long)block & 0x0000000F);
+
+ blockv = vec_ld(0, block);
+ pixelsv1 = vec_ld( 0, pixels);
+ pixelsv2 = vec_ld(16, pixels);
+ pixelsv = vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels));
+
+ if (rightside) {
+ pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1));
+ } else {
+ pixelsv = vec_perm(blockv, pixelsv, vcprm(s0,s1,2,3));
+ }
+
+ blockv = vec_avg(blockv, pixelsv);
+
+ vec_st(blockv, 0, block);
+
+ pixels += line_size;
+ block += line_size;
+ }
+}
+
+/* next one assumes that ((line_size % 8) == 0) */
+static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
+{
+ register int i;
+ register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
+ register vector unsigned char blockv, temp1, temp2;
+ register vector unsigned short pixelssum1, pixelssum2, temp3;
+ register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
+ register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
+
+ temp1 = vec_ld(0, pixels);
+ temp2 = vec_ld(16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
+ if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
+ }
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum1 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ pixelssum1 = vec_add(pixelssum1, vctwo);
+
+ for (i = 0; i < h ; i++) {
+ int rightside = ((unsigned long)block & 0x0000000F);
+ blockv = vec_ld(0, block);
+
+ temp1 = vec_ld(line_size, pixels);
+ temp2 = vec_ld(line_size + 16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
+ if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
+ }
+
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum2 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ temp3 = vec_add(pixelssum1, pixelssum2);
+ temp3 = vec_sra(temp3, vctwo);
+ pixelssum1 = vec_add(pixelssum2, vctwo);
+ pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
+
+ if (rightside) {
+ blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
+ } else {
+ blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
+ }
+
+ vec_st(blockv, 0, block);
+
+ block += line_size;
+ pixels += line_size;
+ }
+}
+
+/* next one assumes that ((line_size % 8) == 0) */
+static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
+{
+ register int i;
+ register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
+ register vector unsigned char blockv, temp1, temp2;
+ register vector unsigned short pixelssum1, pixelssum2, temp3;
+ register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
+ register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
+ register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
+
+ temp1 = vec_ld(0, pixels);
+ temp2 = vec_ld(16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
+ if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
+ }
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum1 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ pixelssum1 = vec_add(pixelssum1, vcone);
+
+ for (i = 0; i < h ; i++) {
+ int rightside = ((unsigned long)block & 0x0000000F);
+ blockv = vec_ld(0, block);
+
+ temp1 = vec_ld(line_size, pixels);
+ temp2 = vec_ld(line_size + 16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
+ if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
+ }
+
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum2 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ temp3 = vec_add(pixelssum1, pixelssum2);
+ temp3 = vec_sra(temp3, vctwo);
+ pixelssum1 = vec_add(pixelssum2, vcone);
+ pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
+
+ if (rightside) {
+ blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
+ } else {
+ blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
+ }
+
+ vec_st(blockv, 0, block);
+
+ block += line_size;
+ pixels += line_size;
+ }
+}
+
+/* next one assumes that ((line_size % 16) == 0) */
+static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
+{
+ register int i;
+ register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
+ register vector unsigned char blockv, temp1, temp2;
+ register vector unsigned short temp3, temp4,
+ pixelssum1, pixelssum2, pixelssum3, pixelssum4;
+ register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
+ register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
+
+ temp1 = vec_ld(0, pixels);
+ temp2 = vec_ld(16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
+ if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
+ }
+ pixelsv3 = vec_mergel(vczero, pixelsv1);
+ pixelsv4 = vec_mergel(vczero, pixelsv2);
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum3 = vec_add((vector unsigned short)pixelsv3,
+ (vector unsigned short)pixelsv4);
+ pixelssum3 = vec_add(pixelssum3, vctwo);
+ pixelssum1 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ pixelssum1 = vec_add(pixelssum1, vctwo);
+
+ for (i = 0; i < h ; i++) {
+ blockv = vec_ld(0, block);
+
+ temp1 = vec_ld(line_size, pixels);
+ temp2 = vec_ld(line_size + 16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
+ if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
+ }
+
+ pixelsv3 = vec_mergel(vczero, pixelsv1);
+ pixelsv4 = vec_mergel(vczero, pixelsv2);
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+
+ pixelssum4 = vec_add((vector unsigned short)pixelsv3,
+ (vector unsigned short)pixelsv4);
+ pixelssum2 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ temp4 = vec_add(pixelssum3, pixelssum4);
+ temp4 = vec_sra(temp4, vctwo);
+ temp3 = vec_add(pixelssum1, pixelssum2);
+ temp3 = vec_sra(temp3, vctwo);
+
+ pixelssum3 = vec_add(pixelssum4, vctwo);
+ pixelssum1 = vec_add(pixelssum2, vctwo);
+
+ blockv = vec_packsu(temp3, temp4);
+
+ vec_st(blockv, 0, block);
+
+ block += line_size;
+ pixels += line_size;
+ }
+}
+
+/* next one assumes that ((line_size % 16) == 0) */
+static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
+{
+ register int i;
+ register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
+ register vector unsigned char blockv, temp1, temp2;
+ register vector unsigned short temp3, temp4,
+ pixelssum1, pixelssum2, pixelssum3, pixelssum4;
+ register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
+ register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
+ register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
+
+ temp1 = vec_ld(0, pixels);
+ temp2 = vec_ld(16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
+ if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
+ }
+ pixelsv3 = vec_mergel(vczero, pixelsv1);
+ pixelsv4 = vec_mergel(vczero, pixelsv2);
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum3 = vec_add((vector unsigned short)pixelsv3,
+ (vector unsigned short)pixelsv4);
+ pixelssum3 = vec_add(pixelssum3, vcone);
+ pixelssum1 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ pixelssum1 = vec_add(pixelssum1, vcone);
+
+ for (i = 0; i < h ; i++) {
+ blockv = vec_ld(0, block);
+
+ temp1 = vec_ld(line_size, pixels);
+ temp2 = vec_ld(line_size + 16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
+ if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
+ }
+
+ pixelsv3 = vec_mergel(vczero, pixelsv1);
+ pixelsv4 = vec_mergel(vczero, pixelsv2);
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+
+ pixelssum4 = vec_add((vector unsigned short)pixelsv3,
+ (vector unsigned short)pixelsv4);
+ pixelssum2 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ temp4 = vec_add(pixelssum3, pixelssum4);
+ temp4 = vec_sra(temp4, vctwo);
+ temp3 = vec_add(pixelssum1, pixelssum2);
+ temp3 = vec_sra(temp3, vctwo);
+
+ pixelssum3 = vec_add(pixelssum4, vcone);
+ pixelssum1 = vec_add(pixelssum2, vcone);
+
+ blockv = vec_packsu(temp3, temp4);
+
+ vec_st(blockv, 0, block);
+
+ block += line_size;
+ pixels += line_size;
+ }
+}
+
+/* next one assumes that ((line_size % 8) == 0) */
+static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
+{
+ register int i;
+ register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
+ register vector unsigned char blockv, temp1, temp2, blocktemp;
+ register vector unsigned short pixelssum1, pixelssum2, temp3;
+
+ register const vector unsigned char vczero = (const vector unsigned char)
+ vec_splat_u8(0);
+ register const vector unsigned short vctwo = (const vector unsigned short)
+ vec_splat_u16(2);
+
+ temp1 = vec_ld(0, pixels);
+ temp2 = vec_ld(16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
+ if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
+ }
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum1 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ pixelssum1 = vec_add(pixelssum1, vctwo);
+
+ for (i = 0; i < h ; i++) {
+ int rightside = ((unsigned long)block & 0x0000000F);
+ blockv = vec_ld(0, block);
+
+ temp1 = vec_ld(line_size, pixels);
+ temp2 = vec_ld(line_size + 16, pixels);
+ pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
+ if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
+ pixelsv2 = temp2;
+ } else {
+ pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
+ }
+
+ pixelsv1 = vec_mergeh(vczero, pixelsv1);
+ pixelsv2 = vec_mergeh(vczero, pixelsv2);
+ pixelssum2 = vec_add((vector unsigned short)pixelsv1,
+ (vector unsigned short)pixelsv2);
+ temp3 = vec_add(pixelssum1, pixelssum2);
+ temp3 = vec_sra(temp3, vctwo);
+ pixelssum1 = vec_add(pixelssum2, vctwo);
+ pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
+
+ if (rightside) {
+ blocktemp = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
+ } else {
+ blocktemp = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
+ }
+
+ blockv = vec_avg(blocktemp, blockv);
+ vec_st(blockv, 0, block);
+
+ block += line_size;
+ pixels += line_size;
+ }
+}
+#endif /* HAVE_ALTIVEC */
+
+void ff_hpeldsp_init_ppc(HpelDSPContext* c, int flags)
+{
+#if HAVE_ALTIVEC
+ int mm_flags = av_get_cpu_flags();
+
+ if (mm_flags & AV_CPU_FLAG_ALTIVEC) {
+ c->put_pixels_tab[0][0] = ff_put_pixels16_altivec;
+ c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_altivec;
+ c->avg_pixels_tab[0][0] = ff_avg_pixels16_altivec;
+ c->avg_pixels_tab[1][0] = avg_pixels8_altivec;
+ c->avg_pixels_tab[1][3] = avg_pixels8_xy2_altivec;
+ c->put_pixels_tab[1][3] = put_pixels8_xy2_altivec;
+ c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_altivec;
+ c->put_pixels_tab[0][3] = put_pixels16_xy2_altivec;
+ c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_altivec;
+ }
+#endif /* HAVE_ALTIVEC */
+}