ref: 3de8a958314fa33224b6f9a065f20aa57434d341
parent: 2add72d9bcec97855691b90e62b8f3f1634e3544
parent: 98fcccfe9751894ace9693a39ba0609fe5ea904d
author: Fritz Koenig <frkoenig@google.com>
date: Mon Jul 26 02:05:30 EDT 2010
Merge changes I896fe6f9,I90d8b167 * changes: Change the x86 idct functions to do reconstruction at the same time Combine idct and reconstruction steps
--- a/vp8/common/arm/idct_arm.h
+++ b/vp8/common/arm/idct_arm.h
@@ -15,7 +15,6 @@
#if HAVE_ARMV6
extern prototype_idct(vp8_short_idct4x4llm_1_v6);
extern prototype_idct(vp8_short_idct4x4llm_v6_dual);
-extern prototype_idct_scalar(vp8_dc_only_idct_armv6);
extern prototype_second_order(vp8_short_inv_walsh4x4_1_armv6);
extern prototype_second_order(vp8_short_inv_walsh4x4_armv6);
@@ -25,9 +24,6 @@
#undef vp8_idct_idct16
#define vp8_idct_idct16 vp8_short_idct4x4llm_v6_dual
-#undef vp8_idct_idct1_scalar
-#define vp8_idct_idct1_scalar vp8_dc_only_idct_armv6
-
#undef vp8_idct_iwalsh1
#define vp8_idct_iwalsh1 vp8_short_inv_walsh4x4_1_armv6
@@ -38,7 +34,6 @@
#if HAVE_ARMV7
extern prototype_idct(vp8_short_idct4x4llm_1_neon);
extern prototype_idct(vp8_short_idct4x4llm_neon);
-extern prototype_idct_scalar(vp8_dc_only_idct_neon);
extern prototype_second_order(vp8_short_inv_walsh4x4_1_neon);
extern prototype_second_order(vp8_short_inv_walsh4x4_neon);
@@ -47,9 +42,6 @@
#undef vp8_idct_idct16
#define vp8_idct_idct16 vp8_short_idct4x4llm_neon
-
-#undef vp8_idct_idct1_scalar
-#define vp8_idct_idct1_scalar vp8_dc_only_idct_neon
#undef vp8_idct_iwalsh1
#define vp8_idct_iwalsh1 vp8_short_inv_walsh4x4_1_neon
--- a/vp8/common/arm/systemdependent.c
+++ b/vp8/common/arm/systemdependent.c
@@ -43,7 +43,6 @@
rtcd->idct.idct1 = vp8_short_idct4x4llm_1_neon;
rtcd->idct.idct16 = vp8_short_idct4x4llm_neon;
- rtcd->idct.idct1_scalar = vp8_dc_only_idct_neon;
rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_neon;
rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_neon;
@@ -75,7 +74,6 @@
rtcd->idct.idct1 = vp8_short_idct4x4llm_1_v6;
rtcd->idct.idct16 = vp8_short_idct4x4llm_v6_dual;
- rtcd->idct.idct1_scalar = vp8_dc_only_idct_armv6;
rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_armv6;
rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_armv6;
--- a/vp8/common/generic/systemdependent.c
+++ b/vp8/common/generic/systemdependent.c
@@ -32,7 +32,7 @@
rtcd->idct.idct1 = vp8_short_idct4x4llm_1_c;
rtcd->idct.idct16 = vp8_short_idct4x4llm_c;
- rtcd->idct.idct1_scalar = vp8_dc_only_idct_c;
+ rtcd->idct.idct1_scalar_add = vp8_dc_only_idct_add_c;
rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_c;
rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_c;
--- a/vp8/common/idct.h
+++ b/vp8/common/idct.h
@@ -18,8 +18,10 @@
#define prototype_idct(sym) \
void sym(short *input, short *output, int pitch)
-#define prototype_idct_scalar(sym) \
- void sym(short input, short *output, int pitch)
+#define prototype_idct_scalar_add(sym) \
+ void sym(short input, \
+ unsigned char *pred, unsigned char *output, \
+ int pitch, int stride)
#if ARCH_X86 || ARCH_X86_64
#include "x86/idct_x86.h"
@@ -39,10 +41,10 @@
#endif
extern prototype_idct(vp8_idct_idct16);
-#ifndef vp8_idct_idct1_scalar
-#define vp8_idct_idct1_scalar vp8_dc_only_idct_c
+#ifndef vp8_idct_idct1_scalar_add
+#define vp8_idct_idct1_scalar_add vp8_dc_only_idct_add_c
#endif
-extern prototype_idct_scalar(vp8_idct_idct1_scalar);
+extern prototype_idct_scalar_add(vp8_idct_idct1_scalar_add);
#ifndef vp8_idct_iwalsh1
@@ -56,14 +58,14 @@
extern prototype_second_order(vp8_idct_iwalsh16);
typedef prototype_idct((*vp8_idct_fn_t));
-typedef prototype_idct_scalar((*vp8_idct_scalar_fn_t));
+typedef prototype_idct_scalar_add((*vp8_idct_scalar_add_fn_t));
typedef prototype_second_order((*vp8_second_order_fn_t));
typedef struct
{
- vp8_idct_fn_t idct1;
- vp8_idct_fn_t idct16;
- vp8_idct_scalar_fn_t idct1_scalar;
+ vp8_idct_fn_t idct1;
+ vp8_idct_fn_t idct16;
+ vp8_idct_scalar_add_fn_t idct1_scalar_add;
vp8_second_order_fn_t iwalsh1;
vp8_second_order_fn_t iwalsh16;
--- a/vp8/common/idctllm.c
+++ b/vp8/common/idctllm.c
@@ -104,23 +104,30 @@
}
}
-
-void vp8_dc_only_idct_c(short input_dc, short *output, int pitch)
+void vp8_dc_only_idct_add_c(short input_dc, unsigned char *pred_ptr, unsigned char *dst_ptr, int pitch, int stride)
{
- int i;
- int a1;
- short *op = output;
- int shortpitch = pitch >> 1;
- a1 = ((input_dc + 4) >> 3);
+ int a1 = ((input_dc + 4) >> 3);
+ int r, c;
- for (i = 0; i < 4; i++)
+ for (r = 0; r < 4; r++)
{
- op[0] = a1;
- op[1] = a1;
- op[2] = a1;
- op[3] = a1;
- op += shortpitch;
+ for (c = 0; c < 4; c++)
+ {
+ int a = a1 + pred_ptr[c] ;
+
+ if (a < 0)
+ a = 0;
+
+ if (a > 255)
+ a = 255;
+
+ dst_ptr[c] = (unsigned char) a ;
+ }
+
+ dst_ptr += stride;
+ pred_ptr += pitch;
}
+
}
void vp8_short_inv_walsh4x4_c(short *input, short *output)
--- a/vp8/common/x86/idct_x86.h
+++ b/vp8/common/x86/idct_x86.h
@@ -22,7 +22,7 @@
#if HAVE_MMX
extern prototype_idct(vp8_short_idct4x4llm_1_mmx);
extern prototype_idct(vp8_short_idct4x4llm_mmx);
-extern prototype_idct_scalar(vp8_dc_only_idct_mmx);
+extern prototype_idct_scalar_add(vp8_dc_only_idct_add_mmx);
extern prototype_second_order(vp8_short_inv_walsh4x4_mmx);
extern prototype_second_order(vp8_short_inv_walsh4x4_1_mmx);
@@ -34,8 +34,8 @@
#undef vp8_idct_idct16
#define vp8_idct_idct16 vp8_short_idct4x4llm_mmx
-#undef vp8_idct_idct1_scalar
-#define vp8_idct_idct1_scalar vp8_dc_only_idct_mmx
+#undef vp8_idct_idct1_scalar_add
+#define vp8_idct_idct1_scalar_add vp8_dc_only_idct_add_mmx
#undef vp8_idct_iwalsh16
#define vp8_idct_iwalsh16 vp8_short_inv_walsh4x4_mmx
--- a/vp8/common/x86/idctllm_mmx.asm
+++ b/vp8/common/x86/idctllm_mmx.asm
@@ -220,35 +220,61 @@
pop rbp
ret
-;void dc_only_idct_mmx(short input_dc, short *output, int pitch)
-global sym(vp8_dc_only_idct_mmx)
-sym(vp8_dc_only_idct_mmx):
+;void vp8_dc_only_idct_add_mmx(short input_dc, unsigned char *pred_ptr, unsigned char *dst_ptr, int pitch, int stride)
+global sym(vp8_dc_only_idct_add_mmx)
+sym(vp8_dc_only_idct_add_mmx):
push rbp
mov rbp, rsp
- SHADOW_ARGS_TO_STACK 3
+ SHADOW_ARGS_TO_STACK 5
GET_GOT rbx
+ push rsi
+ push rdi
; end prolog
- movd mm0, arg(0) ;input_dc
+ mov rsi, arg(1) ;s -- prediction
+ mov rdi, arg(2) ;d -- destination
+ movsxd rax, dword ptr arg(4) ;stride
+ movsxd rdx, dword ptr arg(3) ;pitch
+ pxor mm0, mm0
- paddw mm0, [fours GLOBAL]
- mov rdx, arg(1) ;output
+ movd mm5, arg(0) ;input_dc
- psraw mm0, 3
- movsxd rax, dword ptr arg(2) ;pitch
+ paddw mm5, [fours GLOBAL]
- punpcklwd mm0, mm0
- punpckldq mm0, mm0
+ psraw mm5, 3
- movq [rdx], mm0
- movq [rdx+rax], mm0
+ punpcklwd mm5, mm5
+ punpckldq mm5, mm5
- movq [rdx+rax*2], mm0
- add rdx, rax
+ movd mm1, [rsi]
+ punpcklbw mm1, mm0
+ paddsw mm1, mm5
+ packuswb mm1, mm0 ; pack and unpack to saturate
+ movd [rdi], mm1
- movq [rdx+rax*2], mm0
+ movd mm2, [rsi+rdx]
+ punpcklbw mm2, mm0
+ paddsw mm2, mm5
+ packuswb mm2, mm0 ; pack and unpack to saturate
+ movd [rdi+rax], mm2
+ movd mm3, [rsi+2*rdx]
+ punpcklbw mm3, mm0
+ paddsw mm3, mm5
+ packuswb mm3, mm0 ; pack and unpack to saturate
+ movd [rdi+2*rax], mm3
+
+ add rdi, rax
+ add rsi, rdx
+ movd mm4, [rsi+2*rdx]
+ punpcklbw mm4, mm0
+ paddsw mm4, mm5
+ packuswb mm4, mm0 ; pack and unpack to saturate
+ movd [rdi+2*rax], mm4
+
; begin epilog
+ pop rdi
+ pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
--- a/vp8/common/x86/x86_systemdependent.c
+++ b/vp8/common/x86/x86_systemdependent.c
@@ -42,7 +42,7 @@
{
rtcd->idct.idct1 = vp8_short_idct4x4llm_1_mmx;
rtcd->idct.idct16 = vp8_short_idct4x4llm_mmx;
- rtcd->idct.idct1_scalar = vp8_dc_only_idct_mmx;
+ rtcd->idct.idct1_scalar_add = vp8_dc_only_idct_add_mmx;
rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_mmx;
rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_mmx;
--- a/vp8/decoder/arm/dequantize_arm.h
+++ b/vp8/decoder/arm/dequantize_arm.h
@@ -14,32 +14,14 @@
#if HAVE_ARMV6
extern prototype_dequant_block(vp8_dequantize_b_v6);
-extern prototype_dequant_idct(vp8_dequant_idct_v6);
-extern prototype_dequant_idct_dc(vp8_dequant_dc_idct_v6);
#undef vp8_dequant_block
#define vp8_dequant_block vp8_dequantize_b_v6
-#undef vp8_dequant_idct
-#define vp8_dequant_idct vp8_dequant_idct_v6
-
-#undef vp8_dequant_idct_dc
-#define vp8_dequant_idct_dc vp8_dequant_dc_idct_v6
-#endif
-
#if HAVE_ARMV7
extern prototype_dequant_block(vp8_dequantize_b_neon);
-extern prototype_dequant_idct(vp8_dequant_idct_neon);
-extern prototype_dequant_idct_dc(vp8_dequant_dc_idct_neon);
#undef vp8_dequant_block
#define vp8_dequant_block vp8_dequantize_b_neon
-
-#undef vp8_dequant_idct
-#define vp8_dequant_idct vp8_dequant_idct_neon
-
-#undef vp8_dequant_idct_dc
-#define vp8_dequant_idct_dc vp8_dequant_dc_idct_neon
-#endif
#endif
--- a/vp8/decoder/arm/dsystemdependent.c
+++ b/vp8/decoder/arm/dsystemdependent.c
@@ -23,8 +23,6 @@
pbi->mb.rtcd = &pbi->common.rtcd;
#if HAVE_ARMV7
pbi->dequant.block = vp8_dequantize_b_neon;
- pbi->dequant.idct = vp8_dequant_idct_neon;
- pbi->dequant.idct_dc = vp8_dequant_dc_idct_neon;
pbi->dboolhuff.start = vp8dx_start_decode_c;
pbi->dboolhuff.fill = vp8dx_bool_decoder_fill_c;
pbi->dboolhuff.debool = vp8dx_decode_bool_c;
@@ -32,8 +30,6 @@
#elif HAVE_ARMV6
pbi->dequant.block = vp8_dequantize_b_v6;
- pbi->dequant.idct = vp8_dequant_idct_v6;
- pbi->dequant.idct_dc = vp8_dequant_dc_idct_v6;
pbi->dboolhuff.start = vp8dx_start_decode_c;
pbi->dboolhuff.fill = vp8dx_bool_decoder_fill_c;
pbi->dboolhuff.debool = vp8dx_decode_bool_c;
--- a/vp8/decoder/decodframe.c
+++ b/vp8/decoder/decodframe.c
@@ -126,7 +126,6 @@
}
}
-
static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd)
{
/* If the MV points so far into the UMV border that no visible pixels
@@ -182,8 +181,48 @@
}
-static void reconstruct_mb(VP8D_COMP *pbi, MACROBLOCKD *xd)
+void vp8_decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd)
{
+ int eobtotal = 0;
+ MV orig_mvs[24];
+ int i, do_clamp = xd->mbmi.need_to_clamp_mvs;
+
+ if (xd->mbmi.mb_skip_coeff)
+ {
+ vp8_reset_mb_tokens_context(xd);
+ }
+ else
+ {
+ eobtotal = vp8_decode_mb_tokens(pbi, xd);
+ }
+
+ /* Perform temporary clamping of the MV to be used for prediction */
+ if (do_clamp)
+ {
+ if (xd->mbmi.mode == SPLITMV)
+ for (i=0; i<24; i++)
+ orig_mvs[i] = xd->block[i].bmi.mv.as_mv;
+ else
+ {
+ orig_mvs[0] = xd->mbmi.mv.as_mv;
+ orig_mvs[1] = xd->block[16].bmi.mv.as_mv;
+ }
+ clamp_mvs(xd);
+ }
+
+ xd->mode_info_context->mbmi.dc_diff = 1;
+
+ if (xd->mbmi.mode != B_PRED && xd->mbmi.mode != SPLITMV && eobtotal == 0)
+ {
+ xd->mode_info_context->mbmi.dc_diff = 0;
+ skip_recon_mb(pbi, xd);
+ return;
+ }
+
+ if (xd->segmentation_enabled)
+ mb_init_dequantizer(pbi, xd);
+
+ // do prediction
if (xd->frame_type == KEY_FRAME || xd->mbmi.ref_frame == INTRA_FRAME)
{
vp8_build_intra_predictors_mbuv(xd);
@@ -191,29 +230,19 @@
if (xd->mbmi.mode != B_PRED)
{
vp8_build_intra_predictors_mby_ptr(xd);
- vp8_recon16x16mb(RTCD_VTABLE(recon), xd);
+ } else {
+ vp8_intra_prediction_down_copy(xd);
}
- else
- {
- vp8_recon_intra4x4mb(RTCD_VTABLE(recon), xd);
- }
}
else
{
vp8_build_inter_predictors_mb(xd);
- vp8_recon16x16mb(RTCD_VTABLE(recon), xd);
}
-}
-
-static void de_quantand_idct(VP8D_COMP *pbi, MACROBLOCKD *xd)
-{
- int i;
- BLOCKD *b = &xd->block[24];
-
-
+ // dequantization and idct
if (xd->mbmi.mode != B_PRED && xd->mbmi.mode != SPLITMV)
{
+ BLOCKD *b = &xd->block[24];
DEQUANT_INVOKE(&pbi->dequant, block)(b);
// do 2nd order transform on the dc block
@@ -243,105 +272,66 @@
if (b->eob > 1)
{
- DEQUANT_INVOKE(&pbi->dequant, idct_dc)(b->qcoeff, &b->dequant[0][0], b->diff, 32, xd->block[24].diff[i]);
+ DEQUANT_INVOKE(&pbi->dequant, idct_dc_add)(b->qcoeff, &b->dequant[0][0], b->predictor, *(b->base_dst) + b->dst, 16, b->dst_stride,
+ xd->block[24].diff[i]);
}
else
{
- IDCT_INVOKE(RTCD_VTABLE(idct), idct1_scalar)(xd->block[24].diff[i], b->diff, 32);
+ IDCT_INVOKE(RTCD_VTABLE(idct), idct1_scalar_add)(xd->block[24].diff[i], b->predictor, *(b->base_dst) + b->dst, 16, b->dst_stride);
}
}
-
- for (i = 16; i < 24; i++)
+ }
+ else if ((xd->frame_type == KEY_FRAME || xd->mbmi.ref_frame == INTRA_FRAME) && xd->mbmi.mode == B_PRED)
+ {
+ for (i = 0; i < 16; i++)
{
- b = &xd->block[i];
+ BLOCKD *b = &xd->block[i];
+ vp8_predict_intra4x4(b, b->bmi.mode, b->predictor);
+
if (b->eob > 1)
{
- DEQUANT_INVOKE(&pbi->dequant, idct)(b->qcoeff, &b->dequant[0][0], b->diff, 16);
+ DEQUANT_INVOKE(&pbi->dequant, idct_add)(b->qcoeff, &b->dequant[0][0], b->predictor, *(b->base_dst) + b->dst, 16, b->dst_stride);
}
else
{
- IDCT_INVOKE(RTCD_VTABLE(idct), idct1_scalar)(b->qcoeff[0] * b->dequant[0][0], b->diff, 16);
+ IDCT_INVOKE(RTCD_VTABLE(idct), idct1_scalar_add)(b->qcoeff[0] * b->dequant[0][0], b->predictor, *(b->base_dst) + b->dst, 16, b->dst_stride);
((int *)b->qcoeff)[0] = 0;
}
}
+
}
else
{
- for (i = 0; i < 24; i++)
+ for (i = 0; i < 16; i++)
{
+ BLOCKD *b = &xd->block[i];
- b = &xd->block[i];
-
if (b->eob > 1)
{
- DEQUANT_INVOKE(&pbi->dequant, idct)(b->qcoeff, &b->dequant[0][0], b->diff, (32 - (i & 16)));
+ DEQUANT_INVOKE(&pbi->dequant, idct_add)(b->qcoeff, &b->dequant[0][0], b->predictor, *(b->base_dst) + b->dst, 16, b->dst_stride);
}
else
{
- IDCT_INVOKE(RTCD_VTABLE(idct), idct1_scalar)(b->qcoeff[0] * b->dequant[0][0], b->diff, (32 - (i & 16)));
+ IDCT_INVOKE(RTCD_VTABLE(idct), idct1_scalar_add)(b->qcoeff[0] * b->dequant[0][0], b->predictor, *(b->base_dst) + b->dst, 16, b->dst_stride);
((int *)b->qcoeff)[0] = 0;
}
}
}
-}
-void vp8_decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd)
-{
- int eobtotal = 0;
- MV orig_mvs[24];
- int i, do_clamp = xd->mbmi.need_to_clamp_mvs;
-
- if (xd->mbmi.mb_skip_coeff)
+ for (i = 16; i < 24; i++)
{
- vp8_reset_mb_tokens_context(xd);
- }
- else
- {
- eobtotal = vp8_decode_mb_tokens(pbi, xd);
- }
- /* Perform temporary clamping of the MV to be used for prediction */
- if (do_clamp)
- {
- if (xd->mbmi.mode == SPLITMV)
- for (i=0; i<24; i++)
- orig_mvs[i] = xd->block[i].bmi.mv.as_mv;
- else
+ BLOCKD *b = &xd->block[i];
+
+ if (b->eob > 1)
{
- orig_mvs[0] = xd->mbmi.mv.as_mv;
- orig_mvs[1] = xd->block[16].bmi.mv.as_mv;
+ DEQUANT_INVOKE(&pbi->dequant, idct_add)(b->qcoeff, &b->dequant[0][0], b->predictor, *(b->base_dst) + b->dst, 8, b->dst_stride);
}
- clamp_mvs(xd);
- }
-
- xd->mode_info_context->mbmi.dc_diff = 1;
-
- if (xd->mbmi.mode != B_PRED && xd->mbmi.mode != SPLITMV && eobtotal == 0)
- {
- xd->mode_info_context->mbmi.dc_diff = 0;
- skip_recon_mb(pbi, xd);
- }
- else
- {
- if (xd->segmentation_enabled)
- mb_init_dequantizer(pbi, xd);
-
- de_quantand_idct(pbi, xd);
- reconstruct_mb(pbi, xd);
- }
-
-
- /* Restore the original MV so as not to affect the entropy context. */
- if (do_clamp)
- {
- if (xd->mbmi.mode == SPLITMV)
- for (i=0; i<24; i++)
- xd->block[i].bmi.mv.as_mv = orig_mvs[i];
else
{
- xd->mbmi.mv.as_mv = orig_mvs[0];
- xd->block[16].bmi.mv.as_mv = orig_mvs[1];
+ IDCT_INVOKE(RTCD_VTABLE(idct), idct1_scalar_add)(b->qcoeff[0] * b->dequant[0][0], b->predictor, *(b->base_dst) + b->dst, 8, b->dst_stride);
+ ((int *)b->qcoeff)[0] = 0;
}
}
}
--- a/vp8/decoder/dequantize.c
+++ b/vp8/decoder/dequantize.c
@@ -32,8 +32,12 @@
}
}
-void vp8_dequant_idct_c(short *input, short *dq, short *output, int pitch)
+void vp8_dequant_idct_add_c(short *input, short *dq, unsigned char *pred, unsigned char *dest, int pitch, int stride)
{
+ // output needs to be at least pitch * 4 for vp8_short_idct4x4llm_c to work properly
+ short output[16*4];
+ short *diff_ptr = output;
+ int r, c;
int i;
for (i = 0; i < 16; i++)
@@ -41,13 +45,38 @@
input[i] = dq[i] * input[i];
}
- vp8_short_idct4x4llm_c(input, output, pitch);
+ vp8_short_idct4x4llm_c(input, output, pitch*2);
+
vpx_memset(input, 0, 32);
+
+ for (r = 0; r < 4; r++)
+ {
+ for (c = 0; c < 4; c++)
+ {
+ int a = diff_ptr[c] + pred[c];
+
+ if (a < 0)
+ a = 0;
+
+ if (a > 255)
+ a = 255;
+
+ dest[c] = (unsigned char) a;
+ }
+
+ dest += stride;
+ diff_ptr += pitch;
+ pred += pitch;
+ }
}
-void vp8_dequant_dc_idct_c(short *input, short *dq, short *output, int pitch, int Dc)
+void vp8_dequant_dc_idct_add_c(short *input, short *dq, unsigned char *pred, unsigned char *dest, int pitch, int stride, int Dc)
{
int i;
+ // output needs to be at least pitch * 4 for vp8_short_idct4x4llm_c to work properly
+ short output[16*4];
+ short *diff_ptr = output;
+ int r, c;
input[0] = (short)Dc;
@@ -56,6 +85,27 @@
input[i] = dq[i] * input[i];
}
- vp8_short_idct4x4llm_c(input, output, pitch);
+ vp8_short_idct4x4llm_c(input, output, pitch*2);
+
vpx_memset(input, 0, 32);
+
+ for (r = 0; r < 4; r++)
+ {
+ for (c = 0; c < 4; c++)
+ {
+ int a = diff_ptr[c] + pred[c];
+
+ if (a < 0)
+ a = 0;
+
+ if (a > 255)
+ a = 255;
+
+ dest[c] = (unsigned char) a;
+ }
+
+ dest += stride;
+ diff_ptr += pitch;
+ pred += pitch;
+ }
}
--- a/vp8/decoder/dequantize.h
+++ b/vp8/decoder/dequantize.h
@@ -16,11 +16,16 @@
#define prototype_dequant_block(sym) \
void sym(BLOCKD *x)
-#define prototype_dequant_idct(sym) \
- void sym(short *input, short *dq, short *output, int pitch)
+#define prototype_dequant_idct_add(sym) \
+ void sym(short *input, short *dq, \
+ unsigned char *pred, unsigned char *output, \
+ int pitch, int stride)
-#define prototype_dequant_idct_dc(sym) \
- void sym(short *input, short *dq, short *output, int pitch, int dc)
+#define prototype_dequant_idct_dc_add(sym) \
+ void sym(short *input, short *dq, \
+ unsigned char *pred, unsigned char *output, \
+ int pitch, int stride, \
+ int dc)
#if ARCH_X86 || ARCH_X86_64
#include "x86/dequantize_x86.h"
@@ -35,25 +40,26 @@
#endif
extern prototype_dequant_block(vp8_dequant_block);
-#ifndef vp8_dequant_idct
-#define vp8_dequant_idct vp8_dequant_idct_c
+#ifndef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_c
#endif
-extern prototype_dequant_idct(vp8_dequant_idct);
+extern prototype_dequant_idct_add(vp8_dequant_idct_add);
-#ifndef vp8_dequant_idct_dc
-#define vp8_dequant_idct_dc vp8_dequant_dc_idct_c
+#ifndef vp8_dequant_idct_dc_add
+#define vp8_dequant_idct_dc_add vp8_dequant_dc_idct_add_c
#endif
-extern prototype_dequant_idct_dc(vp8_dequant_idct_dc);
+extern prototype_dequant_idct_dc_add(vp8_dequant_idct_dc_add);
-
typedef prototype_dequant_block((*vp8_dequant_block_fn_t));
-typedef prototype_dequant_idct((*vp8_dequant_idct_fn_t));
-typedef prototype_dequant_idct_dc((*vp8_dequant_idct_dc_fn_t));
+
+typedef prototype_dequant_idct_add((*vp8_dequant_idct_add_fn_t));
+typedef prototype_dequant_idct_dc_add((*vp8_dequant_idct_dc_add_fn_t));
+
typedef struct
{
- vp8_dequant_block_fn_t block;
- vp8_dequant_idct_fn_t idct;
- vp8_dequant_idct_dc_fn_t idct_dc;
+ vp8_dequant_block_fn_t block;
+ vp8_dequant_idct_add_fn_t idct_add;
+ vp8_dequant_idct_dc_add_fn_t idct_dc_add;
} vp8_dequant_rtcd_vtable_t;
#if CONFIG_RUNTIME_CPU_DETECT
--- a/vp8/decoder/generic/dsystemdependent.c
+++ b/vp8/decoder/generic/dsystemdependent.c
@@ -21,8 +21,8 @@
#if CONFIG_RUNTIME_CPU_DETECT
pbi->mb.rtcd = &pbi->common.rtcd;
pbi->dequant.block = vp8_dequantize_b_c;
- pbi->dequant.idct = vp8_dequant_idct_c;
- pbi->dequant.idct_dc = vp8_dequant_dc_idct_c;
+ pbi->dequant.idct_add = vp8_dequant_idct_add_c;
+ pbi->dequant.idct_dc_add = vp8_dequant_dc_idct_add_c;
pbi->dboolhuff.start = vp8dx_start_decode_c;
pbi->dboolhuff.fill = vp8dx_bool_decoder_fill_c;
#if 0 //For use with RTCD, when implemented
--- a/vp8/decoder/x86/dequantize_mmx.asm
+++ b/vp8/decoder/x86/dequantize_mmx.asm
@@ -50,12 +50,12 @@
ret
-;void dequant_idct_mmx(short *input, short *dq, short *output, int pitch)
-global sym(vp8_dequant_idct_mmx)
-sym(vp8_dequant_idct_mmx):
+;void dequant_idct_add_mmx(short *input, short *dq, unsigned char *pred, unsigned char *dest, int pitch, int stride)
+global sym(vp8_dequant_idct_add_mmx)
+sym(vp8_dequant_idct_add_mmx):
push rbp
mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
+ SHADOW_ARGS_TO_STACK 6
GET_GOT rbx
push rsi
push rdi
@@ -77,7 +77,8 @@
movq mm3, [rax+24]
pmullw mm3, [rdx+24]
- mov rdx, arg(2) ;output
+ mov rdx, arg(3) ;dest
+ mov rsi, arg(2) ;pred
pxor mm7, mm7
@@ -88,7 +89,8 @@
movq [rax+24],mm7
- movsxd rax, dword ptr arg(3) ;pitch
+ movsxd rax, dword ptr arg(4) ;pitch
+ movsxd rdi, dword ptr arg(5) ;stride
psubw mm0, mm2 ; b1= 0-2
paddw mm2, mm2 ;
@@ -207,14 +209,35 @@
punpckldq mm2, mm4 ; 32 22 12 02
punpckhdq mm5, mm4 ; 33 23 13 03
- movq [rdx], mm0
+ pxor mm7, mm7
- movq [rdx+rax], mm1
- movq [rdx+rax*2], mm2
+ movd mm4, [rsi]
+ punpcklbw mm4, mm7
+ paddsw mm0, mm4
+ packuswb mm0, mm7
+ movd [rdx], mm0
- add rdx, rax
- movq [rdx+rax*2], mm5
+ movd mm4, [rsi+rax]
+ punpcklbw mm4, mm7
+ paddsw mm1, mm4
+ packuswb mm1, mm7
+ movd [rdx+rdi], mm1
+ movd mm4, [rsi+2*rax]
+ punpcklbw mm4, mm7
+ paddsw mm2, mm4
+ packuswb mm2, mm7
+ movd [rdx+rdi*2], mm2
+
+ add rdx, rdi
+ add rsi, rax
+
+ movd mm4, [rsi+2*rax]
+ punpcklbw mm4, mm7
+ paddsw mm5, mm4
+ packuswb mm5, mm7
+ movd [rdx+rdi*2], mm5
+
; begin epilog
pop rdi
pop rsi
@@ -224,12 +247,12 @@
ret
-;void dequant_dc_idct_mmx(short *input, short *dq, short *output, int pitch, int Dc)
-global sym(vp8_dequant_dc_idct_mmx)
-sym(vp8_dequant_dc_idct_mmx):
+;void dequant_dc_idct_add_mmx(short *input, short *dq, unsigned char *pred, unsigned char *dest, int pitch, int stride, int Dc)
+global sym(vp8_dequant_dc_idct_add_mmx)
+sym(vp8_dequant_dc_idct_add_mmx):
push rbp
mov rbp, rsp
- SHADOW_ARGS_TO_STACK 5
+ SHADOW_ARGS_TO_STACK 7
GET_GOT rbx
push rsi
push rdi
@@ -238,7 +261,7 @@
mov rax, arg(0) ;input
mov rdx, arg(1) ;dq
- movsxd rcx, dword ptr arg(4) ;Dc
+ movsxd rcx, dword ptr arg(6) ;Dc
movq mm0, [rax ]
pmullw mm0, [rdx]
@@ -252,7 +275,8 @@
movq mm3, [rax+24]
pmullw mm3, [rdx+24]
- mov rdx, arg(2) ;output
+ mov rdx, arg(3) ;dest
+ mov rsi, arg(2) ;pred
pxor mm7, mm7
@@ -262,8 +286,10 @@
movq [rax+16],mm7
movq [rax+24],mm7
+
pinsrw mm0, rcx, 0
- movsxd rax, dword ptr arg(3) ;pitch
+ movsxd rax, dword ptr arg(4) ;pitch
+ movsxd rdi, dword ptr arg(5) ;stride
psubw mm0, mm2 ; b1= 0-2
paddw mm2, mm2 ;
@@ -382,13 +408,34 @@
punpckldq mm2, mm4 ; 32 22 12 02
punpckhdq mm5, mm4 ; 33 23 13 03
- movq [rdx], mm0
+ pxor mm7, mm7
- movq [rdx+rax], mm1
- movq [rdx+rax*2], mm2
+ movd mm4, [rsi]
+ punpcklbw mm4, mm7
+ paddsw mm0, mm4
+ packuswb mm0, mm7
+ movd [rdx], mm0
- add rdx, rax
- movq [rdx+rax*2], mm5
+ movd mm4, [rsi+rax]
+ punpcklbw mm4, mm7
+ paddsw mm1, mm4
+ packuswb mm1, mm7
+ movd [rdx+rdi], mm1
+
+ movd mm4, [rsi+2*rax]
+ punpcklbw mm4, mm7
+ paddsw mm2, mm4
+ packuswb mm2, mm7
+ movd [rdx+rdi*2], mm2
+
+ add rdx, rdi
+ add rsi, rax
+
+ movd mm4, [rsi+2*rax]
+ punpcklbw mm4, mm7
+ paddsw mm5, mm4
+ packuswb mm5, mm7
+ movd [rdx+rdi*2], mm5
; begin epilog
pop rdi
--- a/vp8/decoder/x86/dequantize_x86.h
+++ b/vp8/decoder/x86/dequantize_x86.h
@@ -21,8 +21,8 @@
*/
#if HAVE_MMX
extern prototype_dequant_block(vp8_dequantize_b_mmx);
-extern prototype_dequant_idct(vp8_dequant_idct_mmx);
-extern prototype_dequant_idct_dc(vp8_dequant_dc_idct_mmx);
+extern prototype_dequant_idct_add(vp8_dequant_idct_add_mmx);
+extern prototype_dequant_idct_dc_add(vp8_dequant_dc_idct_add_mmx);
#if !CONFIG_RUNTIME_CPU_DETECT
@@ -29,11 +29,11 @@
#undef vp8_dequant_block
#define vp8_dequant_block vp8_dequantize_b_mmx
-#undef vp8_dequant_idct
-#define vp8_dequant_idct vp8_dequant_idct_mmx
+#undef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_mmx
#undef vp8_dequant_idct_dc
-#define vp8_dequant_idct_dc vp8_dequant_dc_idct_mmx
+#define vp8_dequant_idct_add_dc vp8_dequant_dc_idct_add_mmx
#endif
#endif
--- a/vp8/decoder/x86/x86_dsystemdependent.c
+++ b/vp8/decoder/x86/x86_dsystemdependent.c
@@ -43,8 +43,8 @@
if (flags & HAS_MMX)
{
pbi->dequant.block = vp8_dequantize_b_mmx;
- pbi->dequant.idct = vp8_dequant_idct_mmx;
- pbi->dequant.idct_dc = vp8_dequant_dc_idct_mmx;
+ pbi->dequant.idct_add = vp8_dequant_idct_add_mmx;
+ pbi->dequant.idct_dc_add = vp8_dequant_dc_idct_add_mmx;
}
#endif