summaryrefslogtreecommitdiff
path: root/media/ffvpx/libavcodec/x86/h264_intrapred.asm
diff options
context:
space:
mode:
Diffstat (limited to 'media/ffvpx/libavcodec/x86/h264_intrapred.asm')
-rw-r--r--media/ffvpx/libavcodec/x86/h264_intrapred.asm102
1 files changed, 71 insertions, 31 deletions
diff --git a/media/ffvpx/libavcodec/x86/h264_intrapred.asm b/media/ffvpx/libavcodec/x86/h264_intrapred.asm
index c88d91b49e..f3aa3172f0 100644
--- a/media/ffvpx/libavcodec/x86/h264_intrapred.asm
+++ b/media/ffvpx/libavcodec/x86/h264_intrapred.asm
@@ -49,7 +49,7 @@ cextern pw_17
cextern pw_32
;-----------------------------------------------------------------------------
-; void ff_pred16x16_vertical_8(uint8_t *src, int stride)
+; void ff_pred16x16_vertical_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmx
@@ -85,7 +85,7 @@ cglobal pred16x16_vertical_8, 2,3
REP_RET
;-----------------------------------------------------------------------------
-; void ff_pred16x16_horizontal_8(uint8_t *src, int stride)
+; void ff_pred16x16_horizontal_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED16x16_H 0
@@ -126,7 +126,7 @@ INIT_XMM ssse3
PRED16x16_H
;-----------------------------------------------------------------------------
-; void ff_pred16x16_dc_8(uint8_t *src, int stride)
+; void ff_pred16x16_dc_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED16x16_DC 0
@@ -188,7 +188,7 @@ INIT_XMM ssse3
PRED16x16_DC
;-----------------------------------------------------------------------------
-; void ff_pred16x16_tm_vp8_8(uint8_t *src, int stride)
+; void ff_pred16x16_tm_vp8_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED16x16_TM 0
@@ -268,8 +268,45 @@ cglobal pred16x16_tm_vp8_8, 2,6,6
jg .loop
REP_RET
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+cglobal pred16x16_tm_vp8_8, 2, 4, 5, dst, stride, stride3, iteration
+ sub dstq, strideq
+ pmovzxbw m0, [dstq]
+ vpbroadcastb xm1, [r0-1]
+ pmovzxbw m1, xm1
+ psubw m0, m1
+ mov iterationd, 4
+ lea stride3q, [strideq*3]
+.loop:
+ vpbroadcastb xm1, [dstq+strideq*1-1]
+ vpbroadcastb xm2, [dstq+strideq*2-1]
+ vpbroadcastb xm3, [dstq+stride3q-1]
+ vpbroadcastb xm4, [dstq+strideq*4-1]
+ pmovzxbw m1, xm1
+ pmovzxbw m2, xm2
+ pmovzxbw m3, xm3
+ pmovzxbw m4, xm4
+ paddw m1, m0
+ paddw m2, m0
+ paddw m3, m0
+ paddw m4, m0
+ vpackuswb m1, m1, m2
+ vpackuswb m3, m3, m4
+ vpermq m1, m1, q3120
+ vpermq m3, m3, q3120
+ movdqa [dstq+strideq*1], xm1
+ vextracti128 [dstq+strideq*2], m1, 1
+ movdqa [dstq+stride3q*1], xm3
+ vextracti128 [dstq+strideq*4], m3, 1
+ lea dstq, [dstq+strideq*4]
+ dec iterationd
+ jg .loop
+ REP_RET
+%endif
+
;-----------------------------------------------------------------------------
-; void ff_pred16x16_plane_*_8(uint8_t *src, int stride)
+; void ff_pred16x16_plane_*_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro H264_PRED16x16_PLANE 1
@@ -550,7 +587,7 @@ H264_PRED16x16_PLANE rv40
H264_PRED16x16_PLANE svq3
;-----------------------------------------------------------------------------
-; void ff_pred8x8_plane_8(uint8_t *src, int stride)
+; void ff_pred8x8_plane_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro H264_PRED8x8_PLANE 0
@@ -724,7 +761,7 @@ INIT_XMM ssse3
H264_PRED8x8_PLANE
;-----------------------------------------------------------------------------
-; void ff_pred8x8_vertical_8(uint8_t *src, int stride)
+; void ff_pred8x8_vertical_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmx
@@ -741,7 +778,7 @@ cglobal pred8x8_vertical_8, 2,2
RET
;-----------------------------------------------------------------------------
-; void ff_pred8x8_horizontal_8(uint8_t *src, int stride)
+; void ff_pred8x8_horizontal_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED8x8_H 0
@@ -769,7 +806,7 @@ INIT_MMX ssse3
PRED8x8_H
;-----------------------------------------------------------------------------
-; void ff_pred8x8_top_dc_8_mmxext(uint8_t *src, int stride)
+; void ff_pred8x8_top_dc_8_mmxext(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
cglobal pred8x8_top_dc_8, 2,5
@@ -803,7 +840,7 @@ cglobal pred8x8_top_dc_8, 2,5
RET
;-----------------------------------------------------------------------------
-; void ff_pred8x8_dc_8_mmxext(uint8_t *src, int stride)
+; void ff_pred8x8_dc_8_mmxext(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
@@ -864,7 +901,7 @@ cglobal pred8x8_dc_8, 2,5
RET
;-----------------------------------------------------------------------------
-; void ff_pred8x8_dc_rv40_8(uint8_t *src, int stride)
+; void ff_pred8x8_dc_rv40_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
@@ -901,7 +938,7 @@ cglobal pred8x8_dc_rv40_8, 2,7
REP_RET
;-----------------------------------------------------------------------------
-; void ff_pred8x8_tm_vp8_8(uint8_t *src, int stride)
+; void ff_pred8x8_tm_vp8_8(uint8_t *src, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED8x8_TM 0
@@ -1014,7 +1051,7 @@ cglobal pred8x8_tm_vp8_8, 2,3,6
;-----------------------------------------------------------------------------
; void ff_pred8x8l_top_dc_8(uint8_t *src, int has_topleft, int has_topright,
-; int stride)
+; ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED8x8L_TOP_DC 0
cglobal pred8x8l_top_dc_8, 4,4
@@ -1070,7 +1107,7 @@ PRED8x8L_TOP_DC
;-----------------------------------------------------------------------------
; void ff_pred8x8l_dc_8(uint8_t *src, int has_topleft, int has_topright,
-; int stride)
+; ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED8x8L_DC 0
@@ -1174,7 +1211,7 @@ PRED8x8L_DC
;-----------------------------------------------------------------------------
; void ff_pred8x8l_horizontal_8(uint8_t *src, int has_topleft,
-; int has_topright, int stride)
+; int has_topright, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED8x8L_HORIZONTAL 0
@@ -1246,7 +1283,7 @@ PRED8x8L_HORIZONTAL
;-----------------------------------------------------------------------------
; void ff_pred8x8l_vertical_8(uint8_t *src, int has_topleft, int has_topright,
-; int stride)
+; ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED8x8L_VERTICAL 0
@@ -1297,7 +1334,7 @@ PRED8x8L_VERTICAL
;-----------------------------------------------------------------------------
; void ff_pred8x8l_down_left_8(uint8_t *src, int has_topleft,
-; int has_topright, int stride)
+; int has_topright, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
@@ -1498,7 +1535,7 @@ PRED8x8L_DOWN_LEFT
;-----------------------------------------------------------------------------
; void ff_pred8x8l_down_right_8_mmxext(uint8_t *src, int has_topleft,
-; int has_topright, int stride)
+; int has_topright, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
@@ -1750,7 +1787,7 @@ PRED8x8L_DOWN_RIGHT
;-----------------------------------------------------------------------------
; void ff_pred8x8l_vertical_right_8(uint8_t *src, int has_topleft,
-; int has_topright, int stride)
+; int has_topright, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
@@ -1978,7 +2015,7 @@ PRED8x8L_VERTICAL_RIGHT
;-----------------------------------------------------------------------------
; void ff_pred8x8l_vertical_left_8(uint8_t *src, int has_topleft,
-; int has_topright, int stride)
+; int has_topright, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED8x8L_VERTICAL_LEFT 0
@@ -2068,7 +2105,7 @@ PRED8x8L_VERTICAL_LEFT
;-----------------------------------------------------------------------------
; void ff_pred8x8l_horizontal_up_8(uint8_t *src, int has_topleft,
-; int has_topright, int stride)
+; int has_topright, ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED8x8L_HORIZONTAL_UP 0
@@ -2156,7 +2193,7 @@ PRED8x8L_HORIZONTAL_UP
;-----------------------------------------------------------------------------
; void ff_pred8x8l_horizontal_down_8(uint8_t *src, int has_topleft,
-; int has_topright, int stride)
+; int has_topright, ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
@@ -2404,7 +2441,8 @@ INIT_MMX ssse3
PRED8x8L_HORIZONTAL_DOWN
;-------------------------------------------------------------------------------
-; void ff_pred4x4_dc_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+; void ff_pred4x4_dc_8_mmxext(uint8_t *src, const uint8_t *topright,
+; ptrdiff_t stride)
;-------------------------------------------------------------------------------
INIT_MMX mmxext
@@ -2435,7 +2473,7 @@ cglobal pred4x4_dc_8, 3,5
;-----------------------------------------------------------------------------
; void ff_pred4x4_tm_vp8_8_mmxext(uint8_t *src, const uint8_t *topright,
-; int stride)
+; ptrdiff_t stride)
;-----------------------------------------------------------------------------
%macro PRED4x4_TM 0
@@ -2514,7 +2552,7 @@ cglobal pred4x4_tm_vp8_8, 3,3
;-----------------------------------------------------------------------------
; void ff_pred4x4_vertical_vp8_8_mmxext(uint8_t *src, const uint8_t *topright,
-; int stride)
+; ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
@@ -2535,7 +2573,7 @@ cglobal pred4x4_vertical_vp8_8, 3,3
;-----------------------------------------------------------------------------
; void ff_pred4x4_down_left_8_mmxext(uint8_t *src, const uint8_t *topright,
-; int stride)
+; ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
cglobal pred4x4_down_left_8, 3,3
@@ -2562,7 +2600,7 @@ cglobal pred4x4_down_left_8, 3,3
;------------------------------------------------------------------------------
; void ff_pred4x4_vertical_left_8_mmxext(uint8_t *src, const uint8_t *topright,
-; int stride)
+; ptrdiff_t stride)
;------------------------------------------------------------------------------
INIT_MMX mmxext
@@ -2588,7 +2626,7 @@ cglobal pred4x4_vertical_left_8, 3,3
;------------------------------------------------------------------------------
; void ff_pred4x4_horizontal_up_8_mmxext(uint8_t *src, const uint8_t *topright,
-; int stride)
+; ptrdiff_t stride)
;------------------------------------------------------------------------------
INIT_MMX mmxext
@@ -2622,7 +2660,8 @@ cglobal pred4x4_horizontal_up_8, 3,3
;------------------------------------------------------------------------------
; void ff_pred4x4_horizontal_down_8_mmxext(uint8_t *src,
-; const uint8_t *topright, int stride)
+; const uint8_t *topright,
+; ptrdiff_t stride)
;------------------------------------------------------------------------------
INIT_MMX mmxext
@@ -2658,7 +2697,8 @@ cglobal pred4x4_horizontal_down_8, 3,3
;-----------------------------------------------------------------------------
; void ff_pred4x4_vertical_right_8_mmxext(uint8_t *src,
-; const uint8_t *topright, int stride)
+; const uint8_t *topright,
+; ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
@@ -2689,7 +2729,7 @@ cglobal pred4x4_vertical_right_8, 3,3
;-----------------------------------------------------------------------------
; void ff_pred4x4_down_right_8_mmxext(uint8_t *src, const uint8_t *topright,
-; int stride)
+; ptrdiff_t stride)
;-----------------------------------------------------------------------------
INIT_MMX mmxext
generated by cgit v1.2.3 (git 2.26.2) at 2024-11-10 16:14:01 +0000