port changes on lpf from libvpx/nextgenv2
Manually cherry-picked the following commits:
4b5e462 Upgrade vpx_lpf_{vertical,horizontal}_4 mmx to sse2
3ea537c lpf_8_test: remove unneeded function wrapper
110d377 remove loopfilter 'count' param TODOs
9b44d9d split vpx_highbd_lpf_horizontal_16 in two
1b519fb split vpx_lpf_horizontal_16 in two
e7a23d7 vpx_highbd_lpf_horizontal_4: remove unused count param
5171857 vpx_highbd_lpf_horizontal_8: remove unused count param
3c1019e vpx_highbd_lpf_vertical_4: remove unused count param
72a9f06 vpx_highbd_lpf_vertical_8: remove unused count param
b1e97c6 vpx_lpf_horizontal_4: remove unused count param
ab25e46pgrade vpx_lpf_{vertical,horizontal}_4 mmx to sse2
bd5a5bb vpx_lpf_horizontal_8: remove unused count param
109a47b vpx_lpf_vertical_4: remove unused count param
3722574 vpx_lpf_vertical_8: remove unused count param
47dee37 lpf_8_test: add missing dspr2 tests
4fec4a8 lpf_8_test: add missing vpx_lpf_horizontal_4 tests
c3f2c8a lpf_8_test: add missing vpx_lpf_vertical_4 tests
45a7b5e lpf_8_test: simplify function wrapper generation
Change-Id: I0e9212497bbf30de37b19cd2d6ea63b505abe06d
diff --git a/aom_dsp/aom_dsp.mk b/aom_dsp/aom_dsp.mk
index 896b729..6af8327 100644
--- a/aom_dsp/aom_dsp.mk
+++ b/aom_dsp/aom_dsp.mk
@@ -159,7 +159,6 @@
DSP_SRCS-$(ARCH_X86)$(ARCH_X86_64) += x86/loopfilter_sse2.c
DSP_SRCS-$(HAVE_AVX2) += x86/loopfilter_avx2.c
-DSP_SRCS-$(HAVE_MMX) += x86/loopfilter_mmx.asm
DSP_SRCS-$(HAVE_NEON) += arm/loopfilter_neon.c
ifeq ($(HAVE_NEON_ASM),yes)
diff --git a/aom_dsp/aom_dsp_rtcd_defs.pl b/aom_dsp/aom_dsp_rtcd_defs.pl
index fc2535d..dc9a6b9 100644
--- a/aom_dsp/aom_dsp_rtcd_defs.pl
+++ b/aom_dsp/aom_dsp_rtcd_defs.pl
@@ -499,32 +499,36 @@
specialize qw/aom_lpf_vertical_16_dual sse2 neon_asm dspr2 msa/;
$aom_lpf_vertical_16_dual_neon_asm=aom_lpf_vertical_16_dual_neon;
-add_proto qw/void aom_lpf_vertical_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
+add_proto qw/void aom_lpf_vertical_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
specialize qw/aom_lpf_vertical_8 sse2 neon dspr2 msa/;
add_proto qw/void aom_lpf_vertical_8_dual/, "uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1";
specialize qw/aom_lpf_vertical_8_dual sse2 neon_asm dspr2 msa/;
$aom_lpf_vertical_8_dual_neon_asm=aom_lpf_vertical_8_dual_neon;
-add_proto qw/void aom_lpf_vertical_4/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
-specialize qw/aom_lpf_vertical_4 mmx neon dspr2 msa/;
+add_proto qw/void aom_lpf_vertical_4/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
+specialize qw/aom_lpf_vertical_4 sse2 neon dspr2 msa/;
add_proto qw/void aom_lpf_vertical_4_dual/, "uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1";
specialize qw/aom_lpf_vertical_4_dual sse2 neon dspr2 msa/;
-add_proto qw/void aom_lpf_horizontal_16/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
-specialize qw/aom_lpf_horizontal_16 sse2 avx2 neon_asm dspr2 msa/;
-$aom_lpf_horizontal_16_neon_asm=aom_lpf_horizontal_16_neon;
+add_proto qw/void aom_lpf_horizontal_edge_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
+specialize qw/aom_lpf_horizontal_edge_8 sse2 avx2 neon_asm dspr2 msa/;
+$aom_lpf_horizontal_edge_8_neon_asm=aom_lpf_horizontal_edge_8_neon;
-add_proto qw/void aom_lpf_horizontal_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
+add_proto qw/void aom_lpf_horizontal_edge_16/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
+specialize qw/aom_lpf_horizontal_edge_16 sse2 avx2 neon_asm dspr2 msa/;
+$aom_lpf_horizontal_edge_16_neon_asm=aom_lpf_horizontal_edge_16_neon;
+
+add_proto qw/void aom_lpf_horizontal_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
specialize qw/aom_lpf_horizontal_8 sse2 neon dspr2 msa/;
add_proto qw/void aom_lpf_horizontal_8_dual/, "uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1";
specialize qw/aom_lpf_horizontal_8_dual sse2 neon_asm dspr2 msa/;
$aom_lpf_horizontal_8_dual_neon_asm=aom_lpf_horizontal_8_dual_neon;
-add_proto qw/void aom_lpf_horizontal_4/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
-specialize qw/aom_lpf_horizontal_4 mmx neon dspr2 msa/;
+add_proto qw/void aom_lpf_horizontal_4/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
+specialize qw/aom_lpf_horizontal_4 sse2 neon dspr2 msa/;
add_proto qw/void aom_lpf_horizontal_4_dual/, "uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1";
specialize qw/aom_lpf_horizontal_4_dual sse2 neon dspr2 msa/;
@@ -553,28 +557,31 @@
add_proto qw/void aom_highbd_lpf_vertical_16_dual/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
specialize qw/aom_highbd_lpf_vertical_16_dual sse2/;
- add_proto qw/void aom_highbd_lpf_vertical_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+ add_proto qw/void aom_highbd_lpf_vertical_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
specialize qw/aom_highbd_lpf_vertical_8 sse2/;
add_proto qw/void aom_highbd_lpf_vertical_8_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
specialize qw/aom_highbd_lpf_vertical_8_dual sse2/;
- add_proto qw/void aom_highbd_lpf_vertical_4/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+ add_proto qw/void aom_highbd_lpf_vertical_4/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
specialize qw/aom_highbd_lpf_vertical_4 sse2/;
add_proto qw/void aom_highbd_lpf_vertical_4_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
specialize qw/aom_highbd_lpf_vertical_4_dual sse2/;
- add_proto qw/void aom_highbd_lpf_horizontal_16/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
- specialize qw/aom_highbd_lpf_horizontal_16 sse2/;
+ add_proto qw/void aom_highbd_lpf_horizontal_edge_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
+ specialize qw/aom_highbd_lpf_horizontal_edge_8 sse2/;
- add_proto qw/void aom_highbd_lpf_horizontal_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+ add_proto qw/void aom_highbd_lpf_horizontal_edge_16/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
+ specialize qw/aom_highbd_lpf_horizontal_edge_16 sse2/;
+
+ add_proto qw/void aom_highbd_lpf_horizontal_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
specialize qw/aom_highbd_lpf_horizontal_8 sse2/;
add_proto qw/void aom_highbd_lpf_horizontal_8_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
specialize qw/aom_highbd_lpf_horizontal_8_dual sse2/;
- add_proto qw/void aom_highbd_lpf_horizontal_4/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+ add_proto qw/void aom_highbd_lpf_horizontal_4/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
specialize qw/aom_highbd_lpf_horizontal_4 sse2/;
add_proto qw/void aom_highbd_lpf_horizontal_4_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
diff --git a/aom_dsp/arm/loopfilter_4_neon.asm b/aom_dsp/arm/loopfilter_4_neon.asm
index ac1be9d..8b54984 100644
--- a/aom_dsp/arm/loopfilter_4_neon.asm
+++ b/aom_dsp/arm/loopfilter_4_neon.asm
@@ -19,37 +19,28 @@
; Currently aom only works on iterations 8 at a time. The aom loop filter
; works on 16 iterations at a time.
-; TODO(fgalligan): See about removing the count code as this function is only
-; called with a count of 1.
;
; void aom_lpf_horizontal_4_neon(uint8_t *s,
; int p /* pitch */,
; const uint8_t *blimit,
; const uint8_t *limit,
-; const uint8_t *thresh,
-; int count)
+; const uint8_t *thresh)
;
; r0 uint8_t *s,
; r1 int p, /* pitch */
; r2 const uint8_t *blimit,
; r3 const uint8_t *limit,
; sp const uint8_t *thresh,
-; sp+4 int count
|aom_lpf_horizontal_4_neon| PROC
push {lr}
vld1.8 {d0[]}, [r2] ; duplicate *blimit
- ldr r12, [sp, #8] ; load count
ldr r2, [sp, #4] ; load thresh
add r1, r1, r1 ; double pitch
- cmp r12, #0
- beq end_aom_lf_h_edge
-
vld1.8 {d1[]}, [r3] ; duplicate *limit
vld1.8 {d2[]}, [r2] ; duplicate *thresh
-count_lf_h_loop
sub r2, r0, r1, lsl #1 ; move src pointer down by 4 lines
add r3, r2, r1, lsr #1 ; set to 3 lines down
@@ -72,47 +63,34 @@
vst1.u8 {d6}, [r2@64], r1 ; store oq0
vst1.u8 {d7}, [r3@64], r1 ; store oq1
- add r0, r0, #8
- subs r12, r12, #1
- bne count_lf_h_loop
-
-end_aom_lf_h_edge
pop {pc}
ENDP ; |aom_lpf_horizontal_4_neon|
; Currently aom only works on iterations 8 at a time. The aom loop filter
; works on 16 iterations at a time.
-; TODO(fgalligan): See about removing the count code as this function is only
-; called with a count of 1.
;
; void aom_lpf_vertical_4_neon(uint8_t *s,
; int p /* pitch */,
; const uint8_t *blimit,
; const uint8_t *limit,
-; const uint8_t *thresh,
-; int count)
+; const uint8_t *thresh)
;
; r0 uint8_t *s,
; r1 int p, /* pitch */
; r2 const uint8_t *blimit,
; r3 const uint8_t *limit,
; sp const uint8_t *thresh,
-; sp+4 int count
|aom_lpf_vertical_4_neon| PROC
push {lr}
vld1.8 {d0[]}, [r2] ; duplicate *blimit
- ldr r12, [sp, #8] ; load count
vld1.8 {d1[]}, [r3] ; duplicate *limit
ldr r3, [sp, #4] ; load thresh
sub r2, r0, #4 ; move s pointer down by 4 columns
- cmp r12, #0
- beq end_aom_lf_v_edge
vld1.8 {d2[]}, [r3] ; duplicate *thresh
-count_lf_v_loop
vld1.u8 {d3}, [r2], r1 ; load s data
vld1.u8 {d4}, [r2], r1
vld1.u8 {d5}, [r2], r1
@@ -152,12 +130,6 @@
vst4.8 {d4[6], d5[6], d6[6], d7[6]}, [r0], r1
vst4.8 {d4[7], d5[7], d6[7], d7[7]}, [r0]
- add r0, r0, r1, lsl #3 ; s += pitch * 8
- subs r12, r12, #1
- subne r2, r0, #4 ; move s pointer down by 4 columns
- bne count_lf_v_loop
-
-end_aom_lf_v_edge
pop {pc}
ENDP ; |aom_lpf_vertical_4_neon|
diff --git a/aom_dsp/arm/loopfilter_4_neon.c b/aom_dsp/arm/loopfilter_4_neon.c
index cc75228..2b1f80b 100644
--- a/aom_dsp/arm/loopfilter_4_neon.c
+++ b/aom_dsp/arm/loopfilter_4_neon.c
@@ -109,22 +109,18 @@
}
void aom_lpf_horizontal_4_neon(uint8_t *src, int pitch, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count) {
+ const uint8_t *limit, const uint8_t *thresh) {
int i;
uint8_t *s, *psrc;
uint8x8_t dblimit, dlimit, dthresh;
uint8x8_t d3u8, d4u8, d5u8, d6u8, d7u8, d16u8, d17u8, d18u8;
- if (count == 0) // end_aom_lf_h_edge
- return;
-
dblimit = vld1_u8(blimit);
dlimit = vld1_u8(limit);
dthresh = vld1_u8(thresh);
psrc = src - (pitch << 2);
- for (i = 0; i < count; i++) {
+ for (i = 0; i < 1; i++) {
s = psrc + i * 8;
d3u8 = vld1_u8(s);
@@ -159,8 +155,7 @@
}
void aom_lpf_vertical_4_neon(uint8_t *src, int pitch, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count) {
+ const uint8_t *limit, const uint8_t *thresh) {
int i, pitch8;
uint8_t *s;
uint8x8_t dblimit, dlimit, dthresh;
@@ -170,15 +165,12 @@
uint8x8x2_t d2tmp8, d2tmp9, d2tmp10, d2tmp11;
uint8x8x4_t d4Result;
- if (count == 0) // end_aom_lf_h_edge
- return;
-
dblimit = vld1_u8(blimit);
dlimit = vld1_u8(limit);
dthresh = vld1_u8(thresh);
pitch8 = pitch * 8;
- for (i = 0; i < count; i++, src += pitch8) {
+ for (i = 0; i < 1; i++, src += pitch8) {
s = src - (i + 1) * 4;
d3u8 = vld1_u8(s);
diff --git a/aom_dsp/arm/loopfilter_8_neon.asm b/aom_dsp/arm/loopfilter_8_neon.asm
index c5d653e..9f3db66 100644
--- a/aom_dsp/arm/loopfilter_8_neon.asm
+++ b/aom_dsp/arm/loopfilter_8_neon.asm
@@ -19,35 +19,26 @@
; Currently aom only works on iterations 8 at a time. The aom loop filter
; works on 16 iterations at a time.
-; TODO(fgalligan): See about removing the count code as this function is only
-; called with a count of 1.
;
; void aom_lpf_horizontal_8_neon(uint8_t *s, int p,
; const uint8_t *blimit,
; const uint8_t *limit,
-; const uint8_t *thresh,
-; int count)
+; const uint8_t *thresh)
; r0 uint8_t *s,
; r1 int p, /* pitch */
; r2 const uint8_t *blimit,
; r3 const uint8_t *limit,
; sp const uint8_t *thresh,
-; sp+4 int count
|aom_lpf_horizontal_8_neon| PROC
push {r4-r5, lr}
vld1.8 {d0[]}, [r2] ; duplicate *blimit
- ldr r12, [sp, #16] ; load count
ldr r2, [sp, #12] ; load thresh
add r1, r1, r1 ; double pitch
- cmp r12, #0
- beq end_aom_mblf_h_edge
-
vld1.8 {d1[]}, [r3] ; duplicate *limit
vld1.8 {d2[]}, [r2] ; duplicate *thresh
-count_mblf_h_loop
sub r3, r0, r1, lsl #1 ; move src pointer down by 4 lines
add r2, r3, r1, lsr #1 ; set to 3 lines down
@@ -72,11 +63,6 @@
vst1.u8 {d4}, [r2@64], r1 ; store oq1
vst1.u8 {d5}, [r3@64], r1 ; store oq2
- add r0, r0, #8
- subs r12, r12, #1
- bne count_mblf_h_loop
-
-end_aom_mblf_h_edge
pop {r4-r5, pc}
ENDP ; |aom_lpf_horizontal_8_neon|
@@ -85,30 +71,24 @@
; int pitch,
; const uint8_t *blimit,
; const uint8_t *limit,
-; const uint8_t *thresh,
-; int count)
+; const uint8_t *thresh)
;
; r0 uint8_t *s,
; r1 int pitch,
; r2 const uint8_t *blimit,
; r3 const uint8_t *limit,
; sp const uint8_t *thresh,
-; sp+4 int count
|aom_lpf_vertical_8_neon| PROC
push {r4-r5, lr}
vld1.8 {d0[]}, [r2] ; duplicate *blimit
- ldr r12, [sp, #16] ; load count
vld1.8 {d1[]}, [r3] ; duplicate *limit
ldr r3, [sp, #12] ; load thresh
sub r2, r0, #4 ; move s pointer down by 4 columns
- cmp r12, #0
- beq end_aom_mblf_v_edge
vld1.8 {d2[]}, [r3] ; duplicate *thresh
-count_mblf_v_loop
vld1.u8 {d3}, [r2], r1 ; load s data
vld1.u8 {d4}, [r2], r1
vld1.u8 {d5}, [r2], r1
@@ -159,12 +139,6 @@
vst2.8 {d4[6], d5[6]}, [r3], r1
vst2.8 {d4[7], d5[7]}, [r3]
- add r0, r0, r1, lsl #3 ; s += pitch * 8
- subs r12, r12, #1
- subne r2, r0, #4 ; move s pointer down by 4 columns
- bne count_mblf_v_loop
-
-end_aom_mblf_v_edge
pop {r4-r5, pc}
ENDP ; |aom_lpf_vertical_8_neon|
diff --git a/aom_dsp/arm/loopfilter_8_neon.c b/aom_dsp/arm/loopfilter_8_neon.c
index 1551adb..c4502fd 100644
--- a/aom_dsp/arm/loopfilter_8_neon.c
+++ b/aom_dsp/arm/loopfilter_8_neon.c
@@ -261,23 +261,19 @@
}
void aom_lpf_horizontal_8_neon(uint8_t *src, int pitch, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count) {
+ const uint8_t *limit, const uint8_t *thresh) {
int i;
uint8_t *s, *psrc;
uint8x8_t dblimit, dlimit, dthresh;
uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8;
uint8x8_t d16u8, d17u8, d18u8;
- if (count == 0) // end_aom_mblf_h_edge
- return;
-
dblimit = vld1_u8(blimit);
dlimit = vld1_u8(limit);
dthresh = vld1_u8(thresh);
psrc = src - (pitch << 2);
- for (i = 0; i < count; i++) {
+ for (i = 0; i < 1; i++) {
s = psrc + i * 8;
d3u8 = vld1_u8(s);
@@ -317,8 +313,7 @@
}
void aom_lpf_vertical_8_neon(uint8_t *src, int pitch, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count) {
+ const uint8_t *limit, const uint8_t *thresh) {
int i;
uint8_t *s;
uint8x8_t dblimit, dlimit, dthresh;
@@ -330,13 +325,11 @@
uint8x8x4_t d4Result;
uint8x8x2_t d2Result;
- if (count == 0) return;
-
dblimit = vld1_u8(blimit);
dlimit = vld1_u8(limit);
dthresh = vld1_u8(thresh);
- for (i = 0; i < count; i++) {
+ for (i = 0; i < 1; i++) {
s = src + (i * (pitch << 3)) - 4;
d3u8 = vld1_u8(s);
diff --git a/aom_dsp/arm/loopfilter_mb_neon.asm b/aom_dsp/arm/loopfilter_mb_neon.asm
index 2611470..6759288 100644
--- a/aom_dsp/arm/loopfilter_mb_neon.asm
+++ b/aom_dsp/arm/loopfilter_mb_neon.asm
@@ -11,27 +11,28 @@
;
- EXPORT |aom_lpf_horizontal_16_neon|
+ EXPORT |aom_lpf_horizontal_edge_8_neon|
+ EXPORT |aom_lpf_horizontal_edge_16_neon|
EXPORT |aom_lpf_vertical_16_neon|
ARM
AREA ||.text||, CODE, READONLY, ALIGN=2
-; void aom_lpf_horizontal_16_neon(uint8_t *s, int p,
-; const uint8_t *blimit,
-; const uint8_t *limit,
-; const uint8_t *thresh
-; int count)
+; void mb_lpf_horizontal_edge(uint8_t *s, int p,
+; const uint8_t *blimit,
+; const uint8_t *limit,
+; const uint8_t *thresh,
+; int count)
; r0 uint8_t *s,
; r1 int p, /* pitch */
; r2 const uint8_t *blimit,
; r3 const uint8_t *limit,
; sp const uint8_t *thresh,
-|aom_lpf_horizontal_16_neon| PROC
+; r12 int count
+|mb_lpf_horizontal_edge| PROC
push {r4-r8, lr}
vpush {d8-d15}
ldr r4, [sp, #88] ; load thresh
- ldr r12, [sp, #92] ; load count
h_count
vld1.8 {d16[]}, [r2] ; load *blimit
@@ -118,7 +119,35 @@
vpop {d8-d15}
pop {r4-r8, pc}
- ENDP ; |aom_lpf_horizontal_16_neon|
+ ENDP ; |mb_lpf_horizontal_edge|
+
+; void aom_lpf_horizontal_edge_8_neon(uint8_t *s, int pitch,
+; const uint8_t *blimit,
+; const uint8_t *limit,
+; const uint8_t *thresh)
+; r0 uint8_t *s,
+; r1 int pitch,
+; r2 const uint8_t *blimit,
+; r3 const uint8_t *limit,
+; sp const uint8_t *thresh
+|aom_lpf_horizontal_edge_8_neon| PROC
+ mov r12, #1
+ b mb_lpf_horizontal_edge
+ ENDP ; |aom_lpf_horizontal_edge_8_neon|
+
+; void aom_lpf_horizontal_edge_16_neon(uint8_t *s, int pitch,
+; const uint8_t *blimit,
+; const uint8_t *limit,
+; const uint8_t *thresh)
+; r0 uint8_t *s,
+; r1 int pitch,
+; r2 const uint8_t *blimit,
+; r3 const uint8_t *limit,
+; sp const uint8_t *thresh
+|aom_lpf_horizontal_edge_16_neon| PROC
+ mov r12, #2
+ b mb_lpf_horizontal_edge
+ ENDP ; |aom_lpf_horizontal_edge_16_neon|
; void aom_lpf_vertical_16_neon(uint8_t *s, int p,
; const uint8_t *blimit,
diff --git a/aom_dsp/arm/loopfilter_neon.c b/aom_dsp/arm/loopfilter_neon.c
index 15b0934..c90d6bf 100644
--- a/aom_dsp/arm/loopfilter_neon.c
+++ b/aom_dsp/arm/loopfilter_neon.c
@@ -19,8 +19,8 @@
const uint8_t *limit0, const uint8_t *thresh0,
const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1) {
- aom_lpf_vertical_4_neon(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_vertical_4_neon(s + 8 * p, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_vertical_4_neon(s, p, blimit0, limit0, thresh0);
+ aom_lpf_vertical_4_neon(s + 8 * p, p, blimit1, limit1, thresh1);
}
#if HAVE_NEON_ASM
@@ -28,16 +28,16 @@
uint8_t *s, int p /* pitch */, const uint8_t *blimit0,
const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1,
const uint8_t *limit1, const uint8_t *thresh1) {
- aom_lpf_horizontal_8_neon(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_horizontal_8_neon(s + 8, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_horizontal_8_neon(s, p, blimit0, limit0, thresh0);
+ aom_lpf_horizontal_8_neon(s + 8, p, blimit1, limit1, thresh1);
}
void aom_lpf_vertical_8_dual_neon(uint8_t *s, int p, const uint8_t *blimit0,
const uint8_t *limit0, const uint8_t *thresh0,
const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1) {
- aom_lpf_vertical_8_neon(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_vertical_8_neon(s + 8 * p, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_vertical_8_neon(s, p, blimit0, limit0, thresh0);
+ aom_lpf_vertical_8_neon(s + 8 * p, p, blimit1, limit1, thresh1);
}
void aom_lpf_vertical_16_dual_neon(uint8_t *s, int p, const uint8_t *blimit,
diff --git a/aom_dsp/loopfilter.c b/aom_dsp/loopfilter.c
index e43ebe8..c5054b5 100644
--- a/aom_dsp/loopfilter.c
+++ b/aom_dsp/loopfilter.c
@@ -12,6 +12,7 @@
#include <stdlib.h>
#include "./aom_config.h"
+#include "./aom_dsp_rtcd.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_ports/mem.h"
@@ -111,12 +112,12 @@
void aom_lpf_horizontal_4_c(uint8_t *s, int p /* pitch */,
const uint8_t *blimit, const uint8_t *limit,
- const uint8_t *thresh, int count) {
+ const uint8_t *thresh) {
int i;
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < 8; ++i) {
const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
const int8_t mask =
@@ -130,18 +131,17 @@
const uint8_t *limit0, const uint8_t *thresh0,
const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1) {
- aom_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0);
+ aom_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1);
}
void aom_lpf_vertical_4_c(uint8_t *s, int pitch, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count) {
+ const uint8_t *limit, const uint8_t *thresh) {
int i;
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < 8; ++i) {
const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
const int8_t mask =
@@ -155,8 +155,8 @@
const uint8_t *limit0, const uint8_t *thresh0,
const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1) {
- aom_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, 1);
- aom_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1);
+ aom_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0);
+ aom_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1);
}
static INLINE void filter8(int8_t mask, uint8_t thresh, uint8_t flat,
@@ -180,13 +180,12 @@
}
void aom_lpf_horizontal_8_c(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count) {
+ const uint8_t *limit, const uint8_t *thresh) {
int i;
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < 8; ++i) {
const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
@@ -203,16 +202,15 @@
const uint8_t *limit0, const uint8_t *thresh0,
const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1) {
- aom_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0);
+ aom_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1);
}
void aom_lpf_vertical_8_c(uint8_t *s, int pitch, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count) {
+ const uint8_t *limit, const uint8_t *thresh) {
int i;
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < 8; ++i) {
const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
const int8_t mask =
@@ -228,8 +226,8 @@
const uint8_t *limit0, const uint8_t *thresh0,
const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1) {
- aom_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, 1);
- aom_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1);
+ aom_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0);
+ aom_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1);
}
static INLINE void filter16(int8_t mask, uint8_t thresh, uint8_t flat,
@@ -286,9 +284,9 @@
}
}
-void aom_lpf_horizontal_16_c(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count) {
+static void mb_lpf_horizontal_edge_w(uint8_t *s, int p, const uint8_t *blimit,
+ const uint8_t *limit,
+ const uint8_t *thresh, int count) {
int i;
// loop filter designed to work using chars so that we can make maximum use
@@ -311,6 +309,16 @@
}
}
+void aom_lpf_horizontal_edge_8_c(uint8_t *s, int p, const uint8_t *blimit,
+ const uint8_t *limit, const uint8_t *thresh) {
+ mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 1);
+}
+
+void aom_lpf_horizontal_edge_16_c(uint8_t *s, int p, const uint8_t *blimit,
+ const uint8_t *limit, const uint8_t *thresh) {
+ mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 2);
+}
+
static void mb_lpf_vertical_edge_w(uint8_t *s, int p, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh,
int count) {
@@ -435,12 +443,12 @@
void aom_highbd_lpf_horizontal_4_c(uint16_t *s, int p /* pitch */,
const uint8_t *blimit, const uint8_t *limit,
- const uint8_t *thresh, int count, int bd) {
+ const uint8_t *thresh, int bd) {
int i;
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < 8; ++i) {
const uint16_t p3 = s[-4 * p];
const uint16_t p2 = s[-3 * p];
const uint16_t p1 = s[-2 * p];
@@ -460,18 +468,18 @@
uint16_t *s, int p, const uint8_t *blimit0, const uint8_t *limit0,
const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1, int bd) {
- aom_highbd_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, 1, bd);
- aom_highbd_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, 1, bd);
+ aom_highbd_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, bd);
+ aom_highbd_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, bd);
}
void aom_highbd_lpf_vertical_4_c(uint16_t *s, int pitch, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh,
- int count, int bd) {
+ int bd) {
int i;
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < 8; ++i) {
const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
const int8_t mask =
@@ -485,8 +493,8 @@
uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0,
const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1, int bd) {
- aom_highbd_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, 1, bd);
- aom_highbd_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1,
+ aom_highbd_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, bd);
+ aom_highbd_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1,
bd);
}
@@ -512,12 +520,12 @@
void aom_highbd_lpf_horizontal_8_c(uint16_t *s, int p, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh,
- int count, int bd) {
+ int bd) {
int i;
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < 8; ++i) {
const uint16_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
const uint16_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
@@ -535,16 +543,16 @@
uint16_t *s, int p, const uint8_t *blimit0, const uint8_t *limit0,
const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1, int bd) {
- aom_highbd_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, 1, bd);
- aom_highbd_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, 1, bd);
+ aom_highbd_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, bd);
+ aom_highbd_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, bd);
}
void aom_highbd_lpf_vertical_8_c(uint16_t *s, int pitch, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh,
- int count, int bd) {
+ int bd) {
int i;
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < 8; ++i) {
const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
const int8_t mask =
@@ -561,8 +569,8 @@
uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0,
const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1, int bd) {
- aom_highbd_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, 1, bd);
- aom_highbd_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1,
+ aom_highbd_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, bd);
+ aom_highbd_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1,
bd);
}
@@ -632,9 +640,11 @@
}
}
-void aom_highbd_lpf_horizontal_16_c(uint16_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count, int bd) {
+static void highbd_mb_lpf_horizontal_edge_w(uint16_t *s, int p,
+ const uint8_t *blimit,
+ const uint8_t *limit,
+ const uint8_t *thresh, int count,
+ int bd) {
int i;
// loop filter designed to work using chars so that we can make maximum use
@@ -664,6 +674,20 @@
}
}
+void aom_highbd_lpf_horizontal_edge_8_c(uint16_t *s, int p,
+ const uint8_t *blimit,
+ const uint8_t *limit,
+ const uint8_t *thresh, int bd) {
+ highbd_mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 1, bd);
+}
+
+void aom_highbd_lpf_horizontal_edge_16_c(uint16_t *s, int p,
+ const uint8_t *blimit,
+ const uint8_t *limit,
+ const uint8_t *thresh, int bd) {
+ highbd_mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 2, bd);
+}
+
static void highbd_mb_lpf_vertical_edge_w(uint16_t *s, int p,
const uint8_t *blimit,
const uint8_t *limit,
diff --git a/aom_dsp/mips/loopfilter_16_msa.c b/aom_dsp/mips/loopfilter_16_msa.c
index 412eaeb..fc0c32c 100644
--- a/aom_dsp/mips/loopfilter_16_msa.c
+++ b/aom_dsp/mips/loopfilter_16_msa.c
@@ -421,10 +421,10 @@
}
}
-void aom_lpf_horizontal_16_msa(uint8_t *src, int32_t pitch,
- const uint8_t *b_limit_ptr,
- const uint8_t *limit_ptr,
- const uint8_t *thresh_ptr, int32_t count) {
+static void mb_lpf_horizontal_edge(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr, int32_t count) {
if (1 == count) {
uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
uint64_t dword0, dword1;
@@ -644,6 +644,20 @@
}
}
+void aom_lpf_horizontal_edge_8_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
+ mb_lpf_horizontal_edge(src, pitch, b_limit_ptr, limit_ptr, thresh_ptr, 1);
+}
+
+void aom_lpf_horizontal_edge_16_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
+ mb_lpf_horizontal_edge(src, pitch, b_limit_ptr, limit_ptr, thresh_ptr, 2);
+}
+
static void transpose_16x8_to_8x16(uint8_t *input, int32_t in_pitch,
uint8_t *output, int32_t out_pitch) {
v16u8 p7_org, p6_org, p5_org, p4_org, p3_org, p2_org, p1_org, p0_org;
diff --git a/aom_dsp/mips/loopfilter_4_msa.c b/aom_dsp/mips/loopfilter_4_msa.c
index d94936f..dc0a977 100644
--- a/aom_dsp/mips/loopfilter_4_msa.c
+++ b/aom_dsp/mips/loopfilter_4_msa.c
@@ -14,13 +14,11 @@
void aom_lpf_horizontal_4_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr,
const uint8_t *limit_ptr,
- const uint8_t *thresh_ptr, int32_t count) {
+ const uint8_t *thresh_ptr) {
uint64_t p1_d, p0_d, q0_d, q1_d;
v16u8 mask, hev, flat, thresh, b_limit, limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0, p1_out, p0_out, q0_out, q1_out;
- (void)count;
-
/* load vector elements */
LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
@@ -73,14 +71,12 @@
void aom_lpf_vertical_4_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr,
- const uint8_t *limit_ptr, const uint8_t *thresh_ptr,
- int32_t count) {
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
v16u8 mask, hev, flat, limit, thresh, b_limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v8i16 vec0, vec1, vec2, vec3;
- (void)count;
-
LD_UB8((src - 4), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8)__msa_fill_b(*thresh_ptr);
diff --git a/aom_dsp/mips/loopfilter_8_msa.c b/aom_dsp/mips/loopfilter_8_msa.c
index f40eaf5..dc203e7 100644
--- a/aom_dsp/mips/loopfilter_8_msa.c
+++ b/aom_dsp/mips/loopfilter_8_msa.c
@@ -14,7 +14,7 @@
void aom_lpf_horizontal_8_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr,
const uint8_t *limit_ptr,
- const uint8_t *thresh_ptr, int32_t count) {
+ const uint8_t *thresh_ptr) {
uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
v16u8 mask, hev, flat, thresh, b_limit, limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
@@ -23,8 +23,6 @@
v8u16 p3_r, p2_r, p1_r, p0_r, q3_r, q2_r, q1_r, q0_r;
v16i8 zero = { 0 };
- (void)count;
-
/* load vector elements */
LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
@@ -154,8 +152,8 @@
void aom_lpf_vertical_8_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr,
- const uint8_t *limit_ptr, const uint8_t *thresh_ptr,
- int32_t count) {
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 p1_out, p0_out, q0_out, q1_out;
v16u8 flat, mask, hev, thresh, b_limit, limit;
@@ -164,8 +162,6 @@
v16u8 zero = { 0 };
v8i16 vec0, vec1, vec2, vec3, vec4;
- (void)count;
-
/* load vector elements */
LD_UB8(src - 4, pitch, p3, p2, p1, p0, q0, q1, q2, q3);
diff --git a/aom_dsp/mips/loopfilter_filters_dspr2.c b/aom_dsp/mips/loopfilter_filters_dspr2.c
index 08c2ebe..883d052 100644
--- a/aom_dsp/mips/loopfilter_filters_dspr2.c
+++ b/aom_dsp/mips/loopfilter_filters_dspr2.c
@@ -22,7 +22,7 @@
#if HAVE_DSPR2
void aom_lpf_horizontal_4_dspr2(unsigned char *s, int pitch,
const uint8_t *blimit, const uint8_t *limit,
- const uint8_t *thresh, int count) {
+ const uint8_t *thresh) {
uint8_t i;
uint32_t mask;
uint32_t hev;
@@ -107,7 +107,7 @@
void aom_lpf_vertical_4_dspr2(unsigned char *s, int pitch,
const uint8_t *blimit, const uint8_t *limit,
- const uint8_t *thresh, int count) {
+ const uint8_t *thresh) {
uint8_t i;
uint32_t mask, hev;
uint32_t pm1, p0, p1, p2, p3, p4, p5, p6;
@@ -286,16 +286,16 @@
uint8_t *s, int p /* pitch */, const uint8_t *blimit0,
const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1,
const uint8_t *limit1, const uint8_t *thresh1) {
- aom_lpf_horizontal_4_dspr2(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_horizontal_4_dspr2(s + 8, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_horizontal_4_dspr2(s, p, blimit0, limit0, thresh0);
+ aom_lpf_horizontal_4_dspr2(s + 8, p, blimit1, limit1, thresh1);
}
void aom_lpf_horizontal_8_dual_dspr2(
uint8_t *s, int p /* pitch */, const uint8_t *blimit0,
const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1,
const uint8_t *limit1, const uint8_t *thresh1) {
- aom_lpf_horizontal_8_dspr2(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_horizontal_8_dspr2(s + 8, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_horizontal_8_dspr2(s, p, blimit0, limit0, thresh0);
+ aom_lpf_horizontal_8_dspr2(s + 8, p, blimit1, limit1, thresh1);
}
void aom_lpf_vertical_4_dual_dspr2(uint8_t *s, int p, const uint8_t *blimit0,
@@ -304,8 +304,8 @@
const uint8_t *blimit1,
const uint8_t *limit1,
const uint8_t *thresh1) {
- aom_lpf_vertical_4_dspr2(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_vertical_4_dspr2(s + 8 * p, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_vertical_4_dspr2(s, p, blimit0, limit0, thresh0);
+ aom_lpf_vertical_4_dspr2(s + 8 * p, p, blimit1, limit1, thresh1);
}
void aom_lpf_vertical_8_dual_dspr2(uint8_t *s, int p, const uint8_t *blimit0,
@@ -314,8 +314,8 @@
const uint8_t *blimit1,
const uint8_t *limit1,
const uint8_t *thresh1) {
- aom_lpf_vertical_8_dspr2(s, p, blimit0, limit0, thresh0, 1);
- aom_lpf_vertical_8_dspr2(s + 8 * p, p, blimit1, limit1, thresh1, 1);
+ aom_lpf_vertical_8_dspr2(s, p, blimit0, limit0, thresh0);
+ aom_lpf_vertical_8_dspr2(s + 8 * p, p, blimit1, limit1, thresh1);
}
void aom_lpf_vertical_16_dual_dspr2(uint8_t *s, int p, const uint8_t *blimit,
diff --git a/aom_dsp/mips/loopfilter_mb_dspr2.c b/aom_dsp/mips/loopfilter_mb_dspr2.c
index a389ee7..a3b5a9e 100644
--- a/aom_dsp/mips/loopfilter_mb_dspr2.c
+++ b/aom_dsp/mips/loopfilter_mb_dspr2.c
@@ -22,7 +22,7 @@
#if HAVE_DSPR2
void aom_lpf_horizontal_8_dspr2(unsigned char *s, int pitch,
const uint8_t *blimit, const uint8_t *limit,
- const uint8_t *thresh, int count) {
+ const uint8_t *thresh) {
uint32_t mask;
uint32_t hev, flat;
uint8_t i;
@@ -289,7 +289,7 @@
void aom_lpf_vertical_8_dspr2(unsigned char *s, int pitch,
const uint8_t *blimit, const uint8_t *limit,
- const uint8_t *thresh, int count) {
+ const uint8_t *thresh) {
uint8_t i;
uint32_t mask, hev, flat;
uint8_t *s1, *s2, *s3, *s4;
diff --git a/aom_dsp/mips/loopfilter_mb_horiz_dspr2.c b/aom_dsp/mips/loopfilter_mb_horiz_dspr2.c
index a2a6fe4..8d2fd69 100644
--- a/aom_dsp/mips/loopfilter_mb_horiz_dspr2.c
+++ b/aom_dsp/mips/loopfilter_mb_horiz_dspr2.c
@@ -20,9 +20,9 @@
#include "aom_mem/aom_mem.h"
#if HAVE_DSPR2
-void aom_lpf_horizontal_16_dspr2(unsigned char *s, int pitch,
- const uint8_t *blimit, const uint8_t *limit,
- const uint8_t *thresh, int count) {
+static void mb_lpf_horizontal_edge(unsigned char *s, int pitch,
+ const uint8_t *blimit, const uint8_t *limit,
+ const uint8_t *thresh, int count) {
uint32_t mask;
uint32_t hev, flat, flat2;
uint8_t i;
@@ -717,4 +717,18 @@
s = s + 4;
}
}
+
+void aom_lpf_horizontal_edge_8_dspr2(unsigned char *s, int pitch,
+ const uint8_t *blimit,
+ const uint8_t *limit,
+ const uint8_t *thresh) {
+ mb_lpf_horizontal_edge(s, pitch, blimit, limit, thresh, 1);
+}
+
+void aom_lpf_horizontal_edge_16_dspr2(unsigned char *s, int pitch,
+ const uint8_t *blimit,
+ const uint8_t *limit,
+ const uint8_t *thresh) {
+ mb_lpf_horizontal_edge(s, pitch, blimit, limit, thresh, 2);
+}
#endif // #if HAVE_DSPR2
diff --git a/aom_dsp/x86/highbd_loopfilter_sse2.c b/aom_dsp/x86/highbd_loopfilter_sse2.c
index f55e8ce..7636987 100644
--- a/aom_dsp/x86/highbd_loopfilter_sse2.c
+++ b/aom_dsp/x86/highbd_loopfilter_sse2.c
@@ -49,11 +49,10 @@
// TODO(debargha, peter): Break up large functions into smaller ones
// in this file.
-static void highbd_mb_lpf_horizontal_edge_w_sse2_8(uint16_t *s, int p,
- const uint8_t *_blimit,
- const uint8_t *_limit,
- const uint8_t *_thresh,
- int bd) {
+void aom_highbd_lpf_horizontal_edge_8_sse2(uint16_t *s, int p,
+ const uint8_t *_blimit,
+ const uint8_t *_limit,
+ const uint8_t *_thresh, int bd) {
const __m128i zero = _mm_set1_epi16(0);
const __m128i one = _mm_set1_epi16(1);
__m128i blimit, limit, thresh;
@@ -477,33 +476,18 @@
_mm_store_si128((__m128i *)(s - 0 * p), q0);
}
-static void highbd_mb_lpf_horizontal_edge_w_sse2_16(uint16_t *s, int p,
- const uint8_t *_blimit,
- const uint8_t *_limit,
- const uint8_t *_thresh,
- int bd) {
- highbd_mb_lpf_horizontal_edge_w_sse2_8(s, p, _blimit, _limit, _thresh, bd);
- highbd_mb_lpf_horizontal_edge_w_sse2_8(s + 8, p, _blimit, _limit, _thresh,
- bd);
-}
-
-// TODO(yunqingwang): remove count and call these 2 functions(8 or 16) directly.
-void aom_highbd_lpf_horizontal_16_sse2(uint16_t *s, int p,
- const uint8_t *_blimit,
- const uint8_t *_limit,
- const uint8_t *_thresh, int count,
- int bd) {
- if (count == 1)
- highbd_mb_lpf_horizontal_edge_w_sse2_8(s, p, _blimit, _limit, _thresh, bd);
- else
- highbd_mb_lpf_horizontal_edge_w_sse2_16(s, p, _blimit, _limit, _thresh, bd);
+void aom_highbd_lpf_horizontal_edge_16_sse2(uint16_t *s, int p,
+ const uint8_t *_blimit,
+ const uint8_t *_limit,
+ const uint8_t *_thresh, int bd) {
+ aom_highbd_lpf_horizontal_edge_8_sse2(s, p, _blimit, _limit, _thresh, bd);
+ aom_highbd_lpf_horizontal_edge_8_sse2(s + 8, p, _blimit, _limit, _thresh, bd);
}
void aom_highbd_lpf_horizontal_8_sse2(uint16_t *s, int p,
const uint8_t *_blimit,
const uint8_t *_limit,
- const uint8_t *_thresh, int count,
- int bd) {
+ const uint8_t *_thresh, int bd) {
DECLARE_ALIGNED(16, uint16_t, flat_op2[16]);
DECLARE_ALIGNED(16, uint16_t, flat_op1[16]);
DECLARE_ALIGNED(16, uint16_t, flat_op0[16]);
@@ -536,8 +520,6 @@
__m128i work_a;
__m128i filter1, filter2;
- (void)count;
-
if (bd == 8) {
blimit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero);
limit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero);
@@ -732,16 +714,14 @@
uint16_t *s, int p, const uint8_t *_blimit0, const uint8_t *_limit0,
const uint8_t *_thresh0, const uint8_t *_blimit1, const uint8_t *_limit1,
const uint8_t *_thresh1, int bd) {
- aom_highbd_lpf_horizontal_8_sse2(s, p, _blimit0, _limit0, _thresh0, 1, bd);
- aom_highbd_lpf_horizontal_8_sse2(s + 8, p, _blimit1, _limit1, _thresh1, 1,
- bd);
+ aom_highbd_lpf_horizontal_8_sse2(s, p, _blimit0, _limit0, _thresh0, bd);
+ aom_highbd_lpf_horizontal_8_sse2(s + 8, p, _blimit1, _limit1, _thresh1, bd);
}
void aom_highbd_lpf_horizontal_4_sse2(uint16_t *s, int p,
const uint8_t *_blimit,
const uint8_t *_limit,
- const uint8_t *_thresh, int count,
- int bd) {
+ const uint8_t *_thresh, int bd) {
const __m128i zero = _mm_set1_epi16(0);
__m128i blimit, limit, thresh;
__m128i mask, hev, flat;
@@ -781,8 +761,6 @@
__m128i work_a;
__m128i filter1, filter2;
- (void)count;
-
if (bd == 8) {
blimit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero);
limit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero);
@@ -906,9 +884,8 @@
uint16_t *s, int p, const uint8_t *_blimit0, const uint8_t *_limit0,
const uint8_t *_thresh0, const uint8_t *_blimit1, const uint8_t *_limit1,
const uint8_t *_thresh1, int bd) {
- aom_highbd_lpf_horizontal_4_sse2(s, p, _blimit0, _limit0, _thresh0, 1, bd);
- aom_highbd_lpf_horizontal_4_sse2(s + 8, p, _blimit1, _limit1, _thresh1, 1,
- bd);
+ aom_highbd_lpf_horizontal_4_sse2(s, p, _blimit0, _limit0, _thresh0, bd);
+ aom_highbd_lpf_horizontal_4_sse2(s + 8, p, _blimit1, _limit1, _thresh1, bd);
}
static INLINE void highbd_transpose(uint16_t *src[], int in_p, uint16_t *dst[],
@@ -1025,11 +1002,10 @@
void aom_highbd_lpf_vertical_4_sse2(uint16_t *s, int p, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh,
- int count, int bd) {
+ int bd) {
DECLARE_ALIGNED(16, uint16_t, t_dst[8 * 8]);
uint16_t *src[1];
uint16_t *dst[1];
- (void)count;
// Transpose 8x8
src[0] = s - 4;
@@ -1038,8 +1014,7 @@
highbd_transpose(src, p, dst, 8, 1);
// Loop filtering
- aom_highbd_lpf_horizontal_4_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, 1,
- bd);
+ aom_highbd_lpf_horizontal_4_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, bd);
src[0] = t_dst;
dst[0] = s - 4;
@@ -1073,11 +1048,10 @@
void aom_highbd_lpf_vertical_8_sse2(uint16_t *s, int p, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh,
- int count, int bd) {
+ int bd) {
DECLARE_ALIGNED(16, uint16_t, t_dst[8 * 8]);
uint16_t *src[1];
uint16_t *dst[1];
- (void)count;
// Transpose 8x8
src[0] = s - 4;
@@ -1086,8 +1060,7 @@
highbd_transpose(src, p, dst, 8, 1);
// Loop filtering
- aom_highbd_lpf_horizontal_8_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, 1,
- bd);
+ aom_highbd_lpf_horizontal_8_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, bd);
src[0] = t_dst;
dst[0] = s - 4;
@@ -1136,8 +1109,8 @@
highbd_transpose(src, p, dst, 8, 2);
// Loop filtering
- highbd_mb_lpf_horizontal_edge_w_sse2_8(t_dst + 8 * 8, 8, blimit, limit,
- thresh, bd);
+ aom_highbd_lpf_horizontal_edge_8_sse2(t_dst + 8 * 8, 8, blimit, limit, thresh,
+ bd);
src[0] = t_dst;
src[1] = t_dst + 8 * 8;
dst[0] = s - 8;
@@ -1158,8 +1131,8 @@
highbd_transpose8x16(s, s + 8 * p, p, t_dst + 8 * 16, 16);
// Loop filtering
- highbd_mb_lpf_horizontal_edge_w_sse2_16(t_dst + 8 * 16, 16, blimit, limit,
- thresh, bd);
+ aom_highbd_lpf_horizontal_edge_16_sse2(t_dst + 8 * 16, 16, blimit, limit,
+ thresh, bd);
// Transpose back
highbd_transpose8x16(t_dst, t_dst + 8 * 16, 16, s - 8, p);
diff --git a/aom_dsp/x86/loopfilter_avx2.c b/aom_dsp/x86/loopfilter_avx2.c
index 53990a8..1098e17 100644
--- a/aom_dsp/x86/loopfilter_avx2.c
+++ b/aom_dsp/x86/loopfilter_avx2.c
@@ -14,10 +14,10 @@
#include "./aom_dsp_rtcd.h"
#include "aom_ports/mem.h"
-static void mb_lpf_horizontal_edge_w_avx2_8(unsigned char *s, int p,
- const unsigned char *_blimit,
- const unsigned char *_limit,
- const unsigned char *_thresh) {
+void aom_lpf_horizontal_edge_8_avx2(unsigned char *s, int p,
+ const unsigned char *_blimit,
+ const unsigned char *_limit,
+ const unsigned char *_thresh) {
__m128i mask, hev, flat, flat2;
const __m128i zero = _mm_set1_epi16(0);
const __m128i one = _mm_set1_epi8(1);
@@ -368,10 +368,10 @@
8, 128, 9, 128, 10, 128, 11, 128, 12, 128, 13, 128, 14, 128, 15, 128
};
-static void mb_lpf_horizontal_edge_w_avx2_16(unsigned char *s, int p,
- const unsigned char *_blimit,
- const unsigned char *_limit,
- const unsigned char *_thresh) {
+void aom_lpf_horizontal_edge_16_avx2(unsigned char *s, int p,
+ const unsigned char *_blimit,
+ const unsigned char *_limit,
+ const unsigned char *_thresh) {
__m128i mask, hev, flat, flat2;
const __m128i zero = _mm_set1_epi16(0);
const __m128i one = _mm_set1_epi8(1);
@@ -912,13 +912,3 @@
_mm_storeu_si128((__m128i *)(s + 6 * p), q6);
}
}
-
-void aom_lpf_horizontal_16_avx2(unsigned char *s, int p,
- const unsigned char *_blimit,
- const unsigned char *_limit,
- const unsigned char *_thresh, int count) {
- if (count == 1)
- mb_lpf_horizontal_edge_w_avx2_8(s, p, _blimit, _limit, _thresh);
- else
- mb_lpf_horizontal_edge_w_avx2_16(s, p, _blimit, _limit, _thresh);
-}
diff --git a/aom_dsp/x86/loopfilter_mmx.asm b/aom_dsp/x86/loopfilter_mmx.asm
deleted file mode 100644
index d97d4e9..0000000
--- a/aom_dsp/x86/loopfilter_mmx.asm
+++ /dev/null
@@ -1,614 +0,0 @@
-;
-; Copyright (c) 2016, Alliance for Open Media. All rights reserved
-;
-; This source code is subject to the terms of the BSD 2 Clause License and
-; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
-; was not distributed with this source code in the LICENSE file, you can
-; obtain it at www.aomedia.org/license/software. If the Alliance for Open
-; Media Patent License 1.0 was not distributed with this source code in the
-; PATENTS file, you can obtain it at www.aomedia.org/license/patent.
-;
-
-;
-
-
-%include "aom_ports/x86_abi_support.asm"
-
-
-;void aom_lpf_horizontal_4_mmx
-;(
-; unsigned char *src_ptr,
-; int src_pixel_step,
-; const char *blimit,
-; const char *limit,
-; const char *thresh,
-; int count
-;)
-global sym(aom_lpf_horizontal_4_mmx) PRIVATE
-sym(aom_lpf_horizontal_4_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 32 ; reserve 32 bytes
- %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8];
- %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8];
-
- mov rsi, arg(0) ;src_ptr
- movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
-
- movsxd rcx, dword ptr arg(5) ;count
-.next8_h:
- mov rdx, arg(3) ;limit
- movq mm7, [rdx]
- mov rdi, rsi ; rdi points to row +1 for indirect addressing
- add rdi, rax
-
- ; calculate breakout conditions
- movq mm2, [rdi+2*rax] ; q3
- movq mm1, [rsi+2*rax] ; q2
- movq mm6, mm1 ; q2
- psubusb mm1, mm2 ; q2-=q3
- psubusb mm2, mm6 ; q3-=q2
- por mm1, mm2 ; abs(q3-q2)
- psubusb mm1, mm7 ;
-
-
- movq mm4, [rsi+rax] ; q1
- movq mm3, mm4 ; q1
- psubusb mm4, mm6 ; q1-=q2
- psubusb mm6, mm3 ; q2-=q1
- por mm4, mm6 ; abs(q2-q1)
-
- psubusb mm4, mm7
- por mm1, mm4
-
- movq mm4, [rsi] ; q0
- movq mm0, mm4 ; q0
- psubusb mm4, mm3 ; q0-=q1
- psubusb mm3, mm0 ; q1-=q0
- por mm4, mm3 ; abs(q0-q1)
- movq t0, mm4 ; save to t0
- psubusb mm4, mm7
- por mm1, mm4
-
-
- neg rax ; negate pitch to deal with above border
-
- movq mm2, [rsi+4*rax] ; p3
- movq mm4, [rdi+4*rax] ; p2
- movq mm5, mm4 ; p2
- psubusb mm4, mm2 ; p2-=p3
- psubusb mm2, mm5 ; p3-=p2
- por mm4, mm2 ; abs(p3 - p2)
- psubusb mm4, mm7
- por mm1, mm4
-
-
- movq mm4, [rsi+2*rax] ; p1
- movq mm3, mm4 ; p1
- psubusb mm4, mm5 ; p1-=p2
- psubusb mm5, mm3 ; p2-=p1
- por mm4, mm5 ; abs(p2 - p1)
- psubusb mm4, mm7
- por mm1, mm4
-
- movq mm2, mm3 ; p1
-
- movq mm4, [rsi+rax] ; p0
- movq mm5, mm4 ; p0
- psubusb mm4, mm3 ; p0-=p1
- psubusb mm3, mm5 ; p1-=p0
- por mm4, mm3 ; abs(p1 - p0)
- movq t1, mm4 ; save to t1
- psubusb mm4, mm7
- por mm1, mm4
-
- movq mm3, [rdi] ; q1
- movq mm4, mm3 ; q1
- psubusb mm3, mm2 ; q1-=p1
- psubusb mm2, mm4 ; p1-=q1
- por mm2, mm3 ; abs(p1-q1)
- pand mm2, [GLOBAL(tfe)] ; set lsb of each byte to zero
- psrlw mm2, 1 ; abs(p1-q1)/2
-
- movq mm6, mm5 ; p0
- movq mm3, [rsi] ; q0
- psubusb mm5, mm3 ; p0-=q0
- psubusb mm3, mm6 ; q0-=p0
- por mm5, mm3 ; abs(p0 - q0)
- paddusb mm5, mm5 ; abs(p0-q0)*2
- paddusb mm5, mm2 ; abs (p0 - q0) *2 + abs(p1-q1)/2
-
- mov rdx, arg(2) ;blimit ; get blimit
- movq mm7, [rdx] ; blimit
-
- psubusb mm5, mm7 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit
- por mm1, mm5
- pxor mm5, mm5
- pcmpeqb mm1, mm5 ; mask mm1
-
- ; calculate high edge variance
- mov rdx, arg(4) ;thresh ; get thresh
- movq mm7, [rdx] ;
- movq mm4, t0 ; get abs (q1 - q0)
- psubusb mm4, mm7
- movq mm3, t1 ; get abs (p1 - p0)
- psubusb mm3, mm7
- paddb mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
-
- pcmpeqb mm4, mm5
-
- pcmpeqb mm5, mm5
- pxor mm4, mm5
-
-
- ; start work on filters
- movq mm2, [rsi+2*rax] ; p1
- movq mm7, [rdi] ; q1
- pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
- pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
- psubsb mm2, mm7 ; p1 - q1
- pand mm2, mm4 ; high var mask (hvm)(p1 - q1)
- pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
- pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values
- movq mm3, mm0 ; q0
- psubsb mm0, mm6 ; q0 - p0
- paddsb mm2, mm0 ; 1 * (q0 - p0) + hvm(p1 - q1)
- paddsb mm2, mm0 ; 2 * (q0 - p0) + hvm(p1 - q1)
- paddsb mm2, mm0 ; 3 * (q0 - p0) + hvm(p1 - q1)
- pand mm1, mm2 ; mask filter values we don't care about
- movq mm2, mm1
- paddsb mm1, [GLOBAL(t4)] ; 3* (q0 - p0) + hvm(p1 - q1) + 4
- paddsb mm2, [GLOBAL(t3)] ; 3* (q0 - p0) + hvm(p1 - q1) + 3
-
- pxor mm0, mm0 ;
- pxor mm5, mm5
- punpcklbw mm0, mm2 ;
- punpckhbw mm5, mm2 ;
- psraw mm0, 11 ;
- psraw mm5, 11
- packsswb mm0, mm5
- movq mm2, mm0 ; (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
-
- pxor mm0, mm0 ; 0
- movq mm5, mm1 ; abcdefgh
- punpcklbw mm0, mm1 ; e0f0g0h0
- psraw mm0, 11 ; sign extended shift right by 3
- pxor mm1, mm1 ; 0
- punpckhbw mm1, mm5 ; a0b0c0d0
- psraw mm1, 11 ; sign extended shift right by 3
- movq mm5, mm0 ; save results
-
- packsswb mm0, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
- paddsw mm5, [GLOBAL(ones)]
- paddsw mm1, [GLOBAL(ones)]
- psraw mm5, 1 ; partial shifted one more time for 2nd tap
- psraw mm1, 1 ; partial shifted one more time for 2nd tap
- packsswb mm5, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
- pandn mm4, mm5 ; high edge variance additive
-
- paddsb mm6, mm2 ; p0+= p0 add
- pxor mm6, [GLOBAL(t80)] ; unoffset
- movq [rsi+rax], mm6 ; write back
-
- movq mm6, [rsi+2*rax] ; p1
- pxor mm6, [GLOBAL(t80)] ; reoffset
- paddsb mm6, mm4 ; p1+= p1 add
- pxor mm6, [GLOBAL(t80)] ; unoffset
- movq [rsi+2*rax], mm6 ; write back
-
- psubsb mm3, mm0 ; q0-= q0 add
- pxor mm3, [GLOBAL(t80)] ; unoffset
- movq [rsi], mm3 ; write back
-
- psubsb mm7, mm4 ; q1-= q1 add
- pxor mm7, [GLOBAL(t80)] ; unoffset
- movq [rdi], mm7 ; write back
-
- add rsi,8
- neg rax
- dec rcx
- jnz .next8_h
-
- add rsp, 32
- pop rsp
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-;void aom_lpf_vertical_4_mmx
-;(
-; unsigned char *src_ptr,
-; int src_pixel_step,
-; const char *blimit,
-; const char *limit,
-; const char *thresh,
-; int count
-;)
-global sym(aom_lpf_vertical_4_mmx) PRIVATE
-sym(aom_lpf_vertical_4_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 64 ; reserve 64 bytes
- %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8];
- %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8];
- %define srct [rsp + 32] ;__declspec(align(16)) char srct[32];
-
- mov rsi, arg(0) ;src_ptr
- movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
-
- lea rsi, [rsi + rax*4 - 4]
-
- movsxd rcx, dword ptr arg(5) ;count
-.next8_v:
- mov rdi, rsi ; rdi points to row +1 for indirect addressing
- add rdi, rax
-
-
- ;transpose
- movq mm6, [rsi+2*rax] ; 67 66 65 64 63 62 61 60
- movq mm7, mm6 ; 77 76 75 74 73 72 71 70
-
- punpckhbw mm7, [rdi+2*rax] ; 77 67 76 66 75 65 74 64
- punpcklbw mm6, [rdi+2*rax] ; 73 63 72 62 71 61 70 60
-
- movq mm4, [rsi] ; 47 46 45 44 43 42 41 40
- movq mm5, mm4 ; 47 46 45 44 43 42 41 40
-
- punpckhbw mm5, [rsi+rax] ; 57 47 56 46 55 45 54 44
- punpcklbw mm4, [rsi+rax] ; 53 43 52 42 51 41 50 40
-
- movq mm3, mm5 ; 57 47 56 46 55 45 54 44
- punpckhwd mm5, mm7 ; 77 67 57 47 76 66 56 46
-
- punpcklwd mm3, mm7 ; 75 65 55 45 74 64 54 44
- movq mm2, mm4 ; 53 43 52 42 51 41 50 40
-
- punpckhwd mm4, mm6 ; 73 63 53 43 72 62 52 42
- punpcklwd mm2, mm6 ; 71 61 51 41 70 60 50 40
-
- neg rax
- movq mm6, [rsi+rax*2] ; 27 26 25 24 23 22 21 20
-
- movq mm1, mm6 ; 27 26 25 24 23 22 21 20
- punpckhbw mm6, [rsi+rax] ; 37 27 36 36 35 25 34 24
-
- punpcklbw mm1, [rsi+rax] ; 33 23 32 22 31 21 30 20
- movq mm7, [rsi+rax*4]; ; 07 06 05 04 03 02 01 00
-
- punpckhbw mm7, [rdi+rax*4] ; 17 07 16 06 15 05 14 04
- movq mm0, mm7 ; 17 07 16 06 15 05 14 04
-
- punpckhwd mm7, mm6 ; 37 27 17 07 36 26 16 06
- punpcklwd mm0, mm6 ; 35 25 15 05 34 24 14 04
-
- movq mm6, mm7 ; 37 27 17 07 36 26 16 06
- punpckhdq mm7, mm5 ; 77 67 57 47 37 27 17 07 = q3
-
- punpckldq mm6, mm5 ; 76 66 56 46 36 26 16 06 = q2
-
- movq mm5, mm6 ; 76 66 56 46 36 26 16 06
- psubusb mm5, mm7 ; q2-q3
-
- psubusb mm7, mm6 ; q3-q2
- por mm7, mm5; ; mm7=abs (q3-q2)
-
- movq mm5, mm0 ; 35 25 15 05 34 24 14 04
- punpckhdq mm5, mm3 ; 75 65 55 45 35 25 15 05 = q1
-
- punpckldq mm0, mm3 ; 74 64 54 44 34 24 15 04 = q0
- movq mm3, mm5 ; 75 65 55 45 35 25 15 05 = q1
-
- psubusb mm3, mm6 ; q1-q2
- psubusb mm6, mm5 ; q2-q1
-
- por mm6, mm3 ; mm6=abs(q2-q1)
- lea rdx, srct
-
- movq [rdx+24], mm5 ; save q1
- movq [rdx+16], mm0 ; save q0
-
- movq mm3, [rsi+rax*4] ; 07 06 05 04 03 02 01 00
- punpcklbw mm3, [rdi+rax*4] ; 13 03 12 02 11 01 10 00
-
- movq mm0, mm3 ; 13 03 12 02 11 01 10 00
- punpcklwd mm0, mm1 ; 31 21 11 01 30 20 10 00
-
- punpckhwd mm3, mm1 ; 33 23 13 03 32 22 12 02
- movq mm1, mm0 ; 31 21 11 01 30 20 10 00
-
- punpckldq mm0, mm2 ; 70 60 50 40 30 20 10 00 =p3
- punpckhdq mm1, mm2 ; 71 61 51 41 31 21 11 01 =p2
-
- movq mm2, mm1 ; 71 61 51 41 31 21 11 01 =p2
- psubusb mm2, mm0 ; p2-p3
-
- psubusb mm0, mm1 ; p3-p2
- por mm0, mm2 ; mm0=abs(p3-p2)
-
- movq mm2, mm3 ; 33 23 13 03 32 22 12 02
- punpckldq mm2, mm4 ; 72 62 52 42 32 22 12 02 = p1
-
- punpckhdq mm3, mm4 ; 73 63 53 43 33 23 13 03 = p0
- movq [rdx+8], mm3 ; save p0
-
- movq [rdx], mm2 ; save p1
- movq mm5, mm2 ; mm5 = p1
-
- psubusb mm2, mm1 ; p1-p2
- psubusb mm1, mm5 ; p2-p1
-
- por mm1, mm2 ; mm1=abs(p2-p1)
- mov rdx, arg(3) ;limit
-
- movq mm4, [rdx] ; mm4 = limit
- psubusb mm7, mm4
-
- psubusb mm0, mm4
- psubusb mm1, mm4
-
- psubusb mm6, mm4
- por mm7, mm6
-
- por mm0, mm1
- por mm0, mm7 ; abs(q3-q2) > limit || abs(p3-p2) > limit ||abs(p2-p1) > limit || abs(q2-q1) > limit
-
- movq mm1, mm5 ; p1
-
- movq mm7, mm3 ; mm3=mm7=p0
- psubusb mm7, mm5 ; p0 - p1
-
- psubusb mm5, mm3 ; p1 - p0
- por mm5, mm7 ; abs(p1-p0)
-
- movq t0, mm5 ; save abs(p1-p0)
- lea rdx, srct
-
- psubusb mm5, mm4
- por mm0, mm5 ; mm0=mask
-
- movq mm5, [rdx+16] ; mm5=q0
- movq mm7, [rdx+24] ; mm7=q1
-
- movq mm6, mm5 ; mm6=q0
- movq mm2, mm7 ; q1
- psubusb mm5, mm7 ; q0-q1
-
- psubusb mm7, mm6 ; q1-q0
- por mm7, mm5 ; abs(q1-q0)
-
- movq t1, mm7 ; save abs(q1-q0)
- psubusb mm7, mm4
-
- por mm0, mm7 ; mask
-
- movq mm5, mm2 ; q1
- psubusb mm5, mm1 ; q1-=p1
- psubusb mm1, mm2 ; p1-=q1
- por mm5, mm1 ; abs(p1-q1)
- pand mm5, [GLOBAL(tfe)] ; set lsb of each byte to zero
- psrlw mm5, 1 ; abs(p1-q1)/2
-
- mov rdx, arg(2) ;blimit ;
-
- movq mm4, [rdx] ;blimit
- movq mm1, mm3 ; mm1=mm3=p0
-
- movq mm7, mm6 ; mm7=mm6=q0
- psubusb mm1, mm7 ; p0-q0
-
- psubusb mm7, mm3 ; q0-p0
- por mm1, mm7 ; abs(q0-p0)
- paddusb mm1, mm1 ; abs(q0-p0)*2
- paddusb mm1, mm5 ; abs (p0 - q0) *2 + abs(p1-q1)/2
-
- psubusb mm1, mm4 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit
- por mm1, mm0; ; mask
-
- pxor mm0, mm0
- pcmpeqb mm1, mm0
-
- ; calculate high edge variance
- mov rdx, arg(4) ;thresh ; get thresh
- movq mm7, [rdx]
- ;
- movq mm4, t0 ; get abs (q1 - q0)
- psubusb mm4, mm7
-
- movq mm3, t1 ; get abs (p1 - p0)
- psubusb mm3, mm7
-
- por mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
- pcmpeqb mm4, mm0
-
- pcmpeqb mm0, mm0
- pxor mm4, mm0
-
-
-
- ; start work on filters
- lea rdx, srct
-
- movq mm2, [rdx] ; p1
- movq mm7, [rdx+24] ; q1
-
- movq mm6, [rdx+8] ; p0
- movq mm0, [rdx+16] ; q0
-
- pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
- pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
-
- psubsb mm2, mm7 ; p1 - q1
- pand mm2, mm4 ; high var mask (hvm)(p1 - q1)
-
- pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
- pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values
-
- movq mm3, mm0 ; q0
- psubsb mm0, mm6 ; q0 - p0
-
- paddsb mm2, mm0 ; 1 * (q0 - p0) + hvm(p1 - q1)
- paddsb mm2, mm0 ; 2 * (q0 - p0) + hvm(p1 - q1)
-
- paddsb mm2, mm0 ; 3 * (q0 - p0) + hvm(p1 - q1)
- pand mm1, mm2 ; mask filter values we don't care about
-
- movq mm2, mm1
- paddsb mm1, [GLOBAL(t4)] ; 3* (q0 - p0) + hvm(p1 - q1) + 4
-
- paddsb mm2, [GLOBAL(t3)] ; 3* (q0 - p0) + hvm(p1 - q1) + 3
- pxor mm0, mm0 ;
-
- pxor mm5, mm5
- punpcklbw mm0, mm2 ;
-
- punpckhbw mm5, mm2 ;
- psraw mm0, 11 ;
-
- psraw mm5, 11
- packsswb mm0, mm5
-
- movq mm2, mm0 ; (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
-
- pxor mm0, mm0 ; 0
- movq mm5, mm1 ; abcdefgh
-
- punpcklbw mm0, mm1 ; e0f0g0h0
- psraw mm0, 11 ; sign extended shift right by 3
-
- pxor mm1, mm1 ; 0
- punpckhbw mm1, mm5 ; a0b0c0d0
-
- psraw mm1, 11 ; sign extended shift right by 3
- movq mm5, mm0 ; save results
-
- packsswb mm0, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
- paddsw mm5, [GLOBAL(ones)]
-
- paddsw mm1, [GLOBAL(ones)]
- psraw mm5, 1 ; partial shifted one more time for 2nd tap
-
- psraw mm1, 1 ; partial shifted one more time for 2nd tap
- packsswb mm5, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
-
- pandn mm4, mm5 ; high edge variance additive
-
- paddsb mm6, mm2 ; p0+= p0 add
- pxor mm6, [GLOBAL(t80)] ; unoffset
-
- ; mm6=p0 ;
- movq mm1, [rdx] ; p1
- pxor mm1, [GLOBAL(t80)] ; reoffset
-
- paddsb mm1, mm4 ; p1+= p1 add
- pxor mm1, [GLOBAL(t80)] ; unoffset
- ; mm6 = p0 mm1 = p1
-
- psubsb mm3, mm0 ; q0-= q0 add
- pxor mm3, [GLOBAL(t80)] ; unoffset
-
- ; mm3 = q0
- psubsb mm7, mm4 ; q1-= q1 add
- pxor mm7, [GLOBAL(t80)] ; unoffset
- ; mm7 = q1
-
- ; transpose and write back
- ; mm1 = 72 62 52 42 32 22 12 02
- ; mm6 = 73 63 53 43 33 23 13 03
- ; mm3 = 74 64 54 44 34 24 14 04
- ; mm7 = 75 65 55 45 35 25 15 05
-
- movq mm2, mm1 ; 72 62 52 42 32 22 12 02
- punpcklbw mm2, mm6 ; 33 32 23 22 13 12 03 02
-
- movq mm4, mm3 ; 74 64 54 44 34 24 14 04
- punpckhbw mm1, mm6 ; 73 72 63 62 53 52 43 42
-
- punpcklbw mm4, mm7 ; 35 34 25 24 15 14 05 04
- punpckhbw mm3, mm7 ; 75 74 65 64 55 54 45 44
-
- movq mm6, mm2 ; 33 32 23 22 13 12 03 02
- punpcklwd mm2, mm4 ; 15 14 13 12 05 04 03 02
-
- punpckhwd mm6, mm4 ; 35 34 33 32 25 24 23 22
- movq mm5, mm1 ; 73 72 63 62 53 52 43 42
-
- punpcklwd mm1, mm3 ; 55 54 53 52 45 44 43 42
- punpckhwd mm5, mm3 ; 75 74 73 72 65 64 63 62
-
-
- ; mm2 = 15 14 13 12 05 04 03 02
- ; mm6 = 35 34 33 32 25 24 23 22
- ; mm5 = 55 54 53 52 45 44 43 42
- ; mm1 = 75 74 73 72 65 64 63 62
-
-
-
- movd [rsi+rax*4+2], mm2
- psrlq mm2, 32
-
- movd [rdi+rax*4+2], mm2
- movd [rsi+rax*2+2], mm6
-
- psrlq mm6, 32
- movd [rsi+rax+2],mm6
-
- movd [rsi+2], mm1
- psrlq mm1, 32
-
- movd [rdi+2], mm1
- neg rax
-
- movd [rdi+rax+2],mm5
- psrlq mm5, 32
-
- movd [rdi+rax*2+2], mm5
-
- lea rsi, [rsi+rax*8]
- dec rcx
- jnz .next8_v
-
- add rsp, 64
- pop rsp
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-SECTION_RODATA
-align 16
-tfe:
- times 8 db 0xfe
-align 16
-t80:
- times 8 db 0x80
-align 16
-t3:
- times 8 db 0x03
-align 16
-t4:
- times 8 db 0x04
-align 16
-ones:
- times 4 dw 0x0001
diff --git a/aom_dsp/x86/loopfilter_sse2.c b/aom_dsp/x86/loopfilter_sse2.c
index d77077b..2630b51 100644
--- a/aom_dsp/x86/loopfilter_sse2.c
+++ b/aom_dsp/x86/loopfilter_sse2.c
@@ -19,10 +19,221 @@
return _mm_or_si128(_mm_subs_epu8(a, b), _mm_subs_epu8(b, a));
}
-static void mb_lpf_horizontal_edge_w_sse2_8(unsigned char *s, int p,
- const unsigned char *_blimit,
- const unsigned char *_limit,
- const unsigned char *_thresh) {
+// filter_mask and hev_mask
+#define FILTER_HEV_MASK \
+ do { \
+ /* (abs(q1 - q0), abs(p1 - p0) */ \
+ __m128i flat = abs_diff(q1p1, q0p0); \
+ /* abs(p1 - q1), abs(p0 - q0) */ \
+ const __m128i abs_p1q1p0q0 = abs_diff(p1p0, q1q0); \
+ __m128i abs_p0q0, abs_p1q1, work; \
+ \
+ /* const uint8_t hev = hev_mask(thresh, *op1, *op0, *oq0, *oq1); */ \
+ hev = \
+ _mm_unpacklo_epi8(_mm_max_epu8(flat, _mm_srli_si128(flat, 8)), zero); \
+ hev = _mm_cmpgt_epi16(hev, thresh); \
+ hev = _mm_packs_epi16(hev, hev); \
+ \
+ /* const int8_t mask = filter_mask(*limit, *blimit, */ \
+ /* p3, p2, p1, p0, q0, q1, q2, q3); */ \
+ abs_p0q0 = \
+ _mm_adds_epu8(abs_p1q1p0q0, abs_p1q1p0q0); /* abs(p0 - q0) * 2 */ \
+ abs_p1q1 = \
+ _mm_unpackhi_epi8(abs_p1q1p0q0, abs_p1q1p0q0); /* abs(p1 - q1) */ \
+ abs_p1q1 = _mm_srli_epi16(abs_p1q1, 9); \
+ abs_p1q1 = _mm_packs_epi16(abs_p1q1, abs_p1q1); /* abs(p1 - q1) / 2 */ \
+ /* abs(p0 - q0) * 2 + abs(p1 - q1) / 2 */ \
+ mask = _mm_adds_epu8(abs_p0q0, abs_p1q1); \
+ /* abs(p3 - p2), abs(p2 - p1) */ \
+ work = abs_diff(p3p2, p2p1); \
+ flat = _mm_max_epu8(work, flat); \
+ /* abs(q3 - q2), abs(q2 - q1) */ \
+ work = abs_diff(q3q2, q2q1); \
+ flat = _mm_max_epu8(work, flat); \
+ flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8)); \
+ mask = _mm_unpacklo_epi64(mask, flat); \
+ mask = _mm_subs_epu8(mask, limit); \
+ mask = _mm_cmpeq_epi8(mask, zero); \
+ mask = _mm_and_si128(mask, _mm_srli_si128(mask, 8)); \
+ } while (0)
+
+#define FILTER4 \
+ do { \
+ const __m128i t3t4 = \
+ _mm_set_epi8(3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4); \
+ const __m128i t80 = _mm_set1_epi8(0x80); \
+ __m128i filter, filter2filter1, work; \
+ \
+ ps1ps0 = _mm_xor_si128(p1p0, t80); /* ^ 0x80 */ \
+ qs1qs0 = _mm_xor_si128(q1q0, t80); \
+ \
+ /* int8_t filter = signed_char_clamp(ps1 - qs1) & hev; */ \
+ work = _mm_subs_epi8(ps1ps0, qs1qs0); \
+ filter = _mm_and_si128(_mm_srli_si128(work, 8), hev); \
+ /* filter = signed_char_clamp(filter + 3 * (qs0 - ps0)) & mask; */ \
+ filter = _mm_subs_epi8(filter, work); \
+ filter = _mm_subs_epi8(filter, work); \
+ filter = _mm_subs_epi8(filter, work); /* + 3 * (qs0 - ps0) */ \
+ filter = _mm_and_si128(filter, mask); /* & mask */ \
+ filter = _mm_unpacklo_epi64(filter, filter); \
+ \
+ /* filter1 = signed_char_clamp(filter + 4) >> 3; */ \
+ /* filter2 = signed_char_clamp(filter + 3) >> 3; */ \
+ filter2filter1 = _mm_adds_epi8(filter, t3t4); /* signed_char_clamp */ \
+ filter = _mm_unpackhi_epi8(filter2filter1, filter2filter1); \
+ filter2filter1 = _mm_unpacklo_epi8(filter2filter1, filter2filter1); \
+ filter2filter1 = _mm_srai_epi16(filter2filter1, 11); /* >> 3 */ \
+ filter = _mm_srai_epi16(filter, 11); /* >> 3 */ \
+ filter2filter1 = _mm_packs_epi16(filter2filter1, filter); \
+ \
+ /* filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev; */ \
+ filter = _mm_subs_epi8(filter2filter1, ff); /* + 1 */ \
+ filter = _mm_unpacklo_epi8(filter, filter); \
+ filter = _mm_srai_epi16(filter, 9); /* round */ \
+ filter = _mm_packs_epi16(filter, filter); \
+ filter = _mm_andnot_si128(hev, filter); \
+ \
+ hev = _mm_unpackhi_epi64(filter2filter1, filter); \
+ filter2filter1 = _mm_unpacklo_epi64(filter2filter1, filter); \
+ \
+ /* signed_char_clamp(qs1 - filter), signed_char_clamp(qs0 - filter1) */ \
+ qs1qs0 = _mm_subs_epi8(qs1qs0, filter2filter1); \
+ /* signed_char_clamp(ps1 + filter), signed_char_clamp(ps0 + filter2) */ \
+ ps1ps0 = _mm_adds_epi8(ps1ps0, hev); \
+ qs1qs0 = _mm_xor_si128(qs1qs0, t80); /* ^ 0x80 */ \
+ ps1ps0 = _mm_xor_si128(ps1ps0, t80); /* ^ 0x80 */ \
+ } while (0)
+
+void aom_lpf_horizontal_4_sse2(uint8_t *s, int p /* pitch */,
+ const uint8_t *_blimit, const uint8_t *_limit,
+ const uint8_t *_thresh) {
+ const __m128i zero = _mm_set1_epi16(0);
+ const __m128i limit =
+ _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)_blimit),
+ _mm_loadl_epi64((const __m128i *)_limit));
+ const __m128i thresh =
+ _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)_thresh), zero);
+ const __m128i ff = _mm_cmpeq_epi8(zero, zero);
+ __m128i q1p1, q0p0, p3p2, p2p1, p1p0, q3q2, q2q1, q1q0, ps1ps0, qs1qs0;
+ __m128i mask, hev;
+
+ p3p2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 3 * p)),
+ _mm_loadl_epi64((__m128i *)(s - 4 * p)));
+ q1p1 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 2 * p)),
+ _mm_loadl_epi64((__m128i *)(s + 1 * p)));
+ q0p0 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 1 * p)),
+ _mm_loadl_epi64((__m128i *)(s + 0 * p)));
+ q3q2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s + 2 * p)),
+ _mm_loadl_epi64((__m128i *)(s + 3 * p)));
+ p1p0 = _mm_unpacklo_epi64(q0p0, q1p1);
+ p2p1 = _mm_unpacklo_epi64(q1p1, p3p2);
+ q1q0 = _mm_unpackhi_epi64(q0p0, q1p1);
+ q2q1 = _mm_unpacklo_epi64(_mm_srli_si128(q1p1, 8), q3q2);
+
+ FILTER_HEV_MASK;
+ FILTER4;
+
+ _mm_storeh_pi((__m64 *)(s - 2 * p), _mm_castsi128_ps(ps1ps0)); // *op1
+ _mm_storel_epi64((__m128i *)(s - 1 * p), ps1ps0); // *op0
+ _mm_storel_epi64((__m128i *)(s + 0 * p), qs1qs0); // *oq0
+ _mm_storeh_pi((__m64 *)(s + 1 * p), _mm_castsi128_ps(qs1qs0)); // *oq1
+}
+
+void aom_lpf_vertical_4_sse2(uint8_t *s, int p /* pitch */,
+ const uint8_t *_blimit, const uint8_t *_limit,
+ const uint8_t *_thresh) {
+ const __m128i zero = _mm_set1_epi16(0);
+ const __m128i limit =
+ _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)_blimit),
+ _mm_loadl_epi64((const __m128i *)_limit));
+ const __m128i thresh =
+ _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)_thresh), zero);
+ const __m128i ff = _mm_cmpeq_epi8(zero, zero);
+ __m128i x0, x1, x2, x3;
+ __m128i q1p1, q0p0, p3p2, p2p1, p1p0, q3q2, q2q1, q1q0, ps1ps0, qs1qs0;
+ __m128i mask, hev;
+
+ // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
+ q1q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(s + 0 * p - 4)),
+ _mm_loadl_epi64((__m128i *)(s + 1 * p - 4)));
+
+ // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
+ x1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(s + 2 * p - 4)),
+ _mm_loadl_epi64((__m128i *)(s + 3 * p - 4)));
+
+ // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
+ x2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(s + 4 * p - 4)),
+ _mm_loadl_epi64((__m128i *)(s + 5 * p - 4)));
+
+ // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
+ x3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(s + 6 * p - 4)),
+ _mm_loadl_epi64((__m128i *)(s + 7 * p - 4)));
+
+ // Transpose 8x8
+ // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ p1p0 = _mm_unpacklo_epi16(q1q0, x1);
+ // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
+ x0 = _mm_unpacklo_epi16(x2, x3);
+ // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
+ p3p2 = _mm_unpacklo_epi32(p1p0, x0);
+ // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
+ p1p0 = _mm_unpackhi_epi32(p1p0, x0);
+ p3p2 = _mm_unpackhi_epi64(p3p2, _mm_slli_si128(p3p2, 8)); // swap lo and high
+ p1p0 = _mm_unpackhi_epi64(p1p0, _mm_slli_si128(p1p0, 8)); // swap lo and high
+
+ // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ q1q0 = _mm_unpackhi_epi16(q1q0, x1);
+ // 44 54 64 74 45 55 65 75 46 56 66 76 47 57 67 77
+ x2 = _mm_unpackhi_epi16(x2, x3);
+ // 06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77
+ q3q2 = _mm_unpackhi_epi32(q1q0, x2);
+ // 04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75
+ q1q0 = _mm_unpacklo_epi32(q1q0, x2);
+
+ q0p0 = _mm_unpacklo_epi64(p1p0, q1q0);
+ q1p1 = _mm_unpackhi_epi64(p1p0, q1q0);
+ p1p0 = _mm_unpacklo_epi64(q0p0, q1p1);
+ p2p1 = _mm_unpacklo_epi64(q1p1, p3p2);
+ q2q1 = _mm_unpacklo_epi64(_mm_srli_si128(q1p1, 8), q3q2);
+
+ FILTER_HEV_MASK;
+ FILTER4;
+
+ // Transpose 8x4 to 4x8
+ // qs1qs0: 20 21 22 23 24 25 26 27 30 31 32 33 34 34 36 37
+ // ps1ps0: 10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07
+ // 00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17
+ ps1ps0 = _mm_unpackhi_epi64(ps1ps0, _mm_slli_si128(ps1ps0, 8));
+ // 10 30 11 31 12 32 13 33 14 34 15 35 16 36 17 37
+ x0 = _mm_unpackhi_epi8(ps1ps0, qs1qs0);
+ // 00 20 01 21 02 22 03 23 04 24 05 25 06 26 07 27
+ ps1ps0 = _mm_unpacklo_epi8(ps1ps0, qs1qs0);
+ // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+ qs1qs0 = _mm_unpackhi_epi8(ps1ps0, x0);
+ // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+ ps1ps0 = _mm_unpacklo_epi8(ps1ps0, x0);
+
+ *(int *)(s + 0 * p - 2) = _mm_cvtsi128_si32(ps1ps0);
+ ps1ps0 = _mm_srli_si128(ps1ps0, 4);
+ *(int *)(s + 1 * p - 2) = _mm_cvtsi128_si32(ps1ps0);
+ ps1ps0 = _mm_srli_si128(ps1ps0, 4);
+ *(int *)(s + 2 * p - 2) = _mm_cvtsi128_si32(ps1ps0);
+ ps1ps0 = _mm_srli_si128(ps1ps0, 4);
+ *(int *)(s + 3 * p - 2) = _mm_cvtsi128_si32(ps1ps0);
+
+ *(int *)(s + 4 * p - 2) = _mm_cvtsi128_si32(qs1qs0);
+ qs1qs0 = _mm_srli_si128(qs1qs0, 4);
+ *(int *)(s + 5 * p - 2) = _mm_cvtsi128_si32(qs1qs0);
+ qs1qs0 = _mm_srli_si128(qs1qs0, 4);
+ *(int *)(s + 6 * p - 2) = _mm_cvtsi128_si32(qs1qs0);
+ qs1qs0 = _mm_srli_si128(qs1qs0, 4);
+ *(int *)(s + 7 * p - 2) = _mm_cvtsi128_si32(qs1qs0);
+}
+
+void aom_lpf_horizontal_edge_8_sse2(unsigned char *s, int p,
+ const unsigned char *_blimit,
+ const unsigned char *_limit,
+ const unsigned char *_thresh) {
const __m128i zero = _mm_set1_epi16(0);
const __m128i one = _mm_set1_epi8(1);
const __m128i blimit = _mm_load_si128((const __m128i *)_blimit);
@@ -381,10 +592,10 @@
return _mm_or_si128(_mm_andnot_si128(*flat, *other_filt), result);
}
-static void mb_lpf_horizontal_edge_w_sse2_16(unsigned char *s, int p,
- const unsigned char *_blimit,
- const unsigned char *_limit,
- const unsigned char *_thresh) {
+void aom_lpf_horizontal_edge_16_sse2(unsigned char *s, int p,
+ const unsigned char *_blimit,
+ const unsigned char *_limit,
+ const unsigned char *_thresh) {
const __m128i zero = _mm_set1_epi16(0);
const __m128i one = _mm_set1_epi8(1);
const __m128i blimit = _mm_load_si128((const __m128i *)_blimit);
@@ -713,21 +924,10 @@
}
}
-// TODO(yunqingwang): remove count and call these 2 functions(8 or 16) directly.
-void aom_lpf_horizontal_16_sse2(unsigned char *s, int p,
- const unsigned char *_blimit,
- const unsigned char *_limit,
- const unsigned char *_thresh, int count) {
- if (count == 1)
- mb_lpf_horizontal_edge_w_sse2_8(s, p, _blimit, _limit, _thresh);
- else
- mb_lpf_horizontal_edge_w_sse2_16(s, p, _blimit, _limit, _thresh);
-}
-
void aom_lpf_horizontal_8_sse2(unsigned char *s, int p,
const unsigned char *_blimit,
const unsigned char *_limit,
- const unsigned char *_thresh, int count) {
+ const unsigned char *_thresh) {
DECLARE_ALIGNED(16, unsigned char, flat_op2[16]);
DECLARE_ALIGNED(16, unsigned char, flat_op1[16]);
DECLARE_ALIGNED(16, unsigned char, flat_op0[16]);
@@ -742,8 +942,6 @@
__m128i p3, p2, p1, p0, q0, q1, q2, q3;
__m128i q3p3, q2p2, q1p1, q0p0, p1q1, p0q0;
- (void)count;
-
q3p3 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 4 * p)),
_mm_loadl_epi64((__m128i *)(s + 3 * p)));
q2p2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 3 * p)),
@@ -1486,11 +1684,10 @@
void aom_lpf_vertical_8_sse2(unsigned char *s, int p,
const unsigned char *blimit,
const unsigned char *limit,
- const unsigned char *thresh, int count) {
+ const unsigned char *thresh) {
DECLARE_ALIGNED(8, unsigned char, t_dst[8 * 8]);
unsigned char *src[1];
unsigned char *dst[1];
- (void)count;
// Transpose 8x8
src[0] = s - 4;
@@ -1499,7 +1696,7 @@
transpose(src, p, dst, 8, 1);
// Loop filtering
- aom_lpf_horizontal_8_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, 1);
+ aom_lpf_horizontal_8_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh);
src[0] = t_dst;
dst[0] = s - 4;
@@ -1549,7 +1746,7 @@
transpose(src, p, dst, 8, 2);
// Loop filtering
- mb_lpf_horizontal_edge_w_sse2_8(t_dst + 8 * 8, 8, blimit, limit, thresh);
+ aom_lpf_horizontal_edge_8_sse2(t_dst + 8 * 8, 8, blimit, limit, thresh);
src[0] = t_dst;
src[1] = t_dst + 8 * 8;
@@ -1570,7 +1767,7 @@
transpose8x16(s, s + 8 * p, p, t_dst + 8 * 16, 16);
// Loop filtering
- mb_lpf_horizontal_edge_w_sse2_16(t_dst + 8 * 16, 16, blimit, limit, thresh);
+ aom_lpf_horizontal_edge_16_sse2(t_dst + 8 * 16, 16, blimit, limit, thresh);
// Transpose back
transpose8x16(t_dst, t_dst + 8 * 16, 16, s - 8, p);
diff --git a/av1/common/loopfilter.c b/av1/common/loopfilter.c
index 09afaf1..5d6b87e 100644
--- a/av1/common/loopfilter.c
+++ b/av1/common/loopfilter.c
@@ -335,7 +335,6 @@
const loop_filter_thresh *lfi0 = lfi_n->lfthr + *lfl;
const loop_filter_thresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward);
- // TODO(yunqingwang): count in loopfilter functions should be removed.
if (mask & 1) {
if ((mask_16x16_0 | mask_16x16_1) & 1) {
if ((mask_16x16_0 & mask_16x16_1) & 1) {
@@ -355,11 +354,10 @@
lfi0->hev_thr, lfi1->mblim, lfi1->lim,
lfi1->hev_thr);
} else if (mask_8x8_0 & 1) {
- aom_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr,
- 1);
+ aom_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
} else {
aom_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
- lfi1->hev_thr, 1);
+ lfi1->hev_thr);
}
}
@@ -369,11 +367,10 @@
lfi0->hev_thr, lfi1->mblim, lfi1->lim,
lfi1->hev_thr);
} else if (mask_4x4_0 & 1) {
- aom_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr,
- 1);
+ aom_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
} else {
aom_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
- lfi1->hev_thr, 1);
+ lfi1->hev_thr);
}
}
@@ -384,10 +381,10 @@
lfi1->hev_thr);
} else if (mask_4x4_int_0 & 1) {
aom_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
- lfi0->hev_thr, 1);
+ lfi0->hev_thr);
} else {
aom_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim, lfi1->lim,
- lfi1->hev_thr, 1);
+ lfi1->hev_thr);
}
}
}
@@ -431,7 +428,6 @@
const loop_filter_thresh *lfi0 = lfi_n->lfthr + *lfl;
const loop_filter_thresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward);
- // TODO(yunqingwang): count in loopfilter functions should be removed.
if (mask & 1) {
if ((mask_16x16_0 | mask_16x16_1) & 1) {
if ((mask_16x16_0 & mask_16x16_1) & 1) {
@@ -453,10 +449,10 @@
lfi1->hev_thr, bd);
} else if (mask_8x8_0 & 1) {
aom_highbd_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim,
- lfi0->hev_thr, 1, bd);
+ lfi0->hev_thr, bd);
} else {
aom_highbd_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim,
- lfi1->lim, lfi1->hev_thr, 1, bd);
+ lfi1->lim, lfi1->hev_thr, bd);
}
}
@@ -467,10 +463,10 @@
lfi1->hev_thr, bd);
} else if (mask_4x4_0 & 1) {
aom_highbd_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim,
- lfi0->hev_thr, 1, bd);
+ lfi0->hev_thr, bd);
} else {
aom_highbd_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim,
- lfi1->lim, lfi1->hev_thr, 1, bd);
+ lfi1->lim, lfi1->hev_thr, bd);
}
}
@@ -481,10 +477,10 @@
lfi1->hev_thr, bd);
} else if (mask_4x4_int_0 & 1) {
aom_highbd_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
- lfi0->hev_thr, 1, bd);
+ lfi0->hev_thr, bd);
} else {
aom_highbd_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim,
- lfi1->lim, lfi1->hev_thr, 1, bd);
+ lfi1->lim, lfi1->hev_thr, bd);
}
}
}
@@ -518,12 +514,12 @@
if (mask & 1) {
if (mask_16x16 & 1) {
if ((mask_16x16 & 3) == 3) {
- aom_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
- 2);
+ aom_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim,
+ lfi->hev_thr);
count = 2;
} else {
- aom_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
- 1);
+ aom_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim,
+ lfi->hev_thr);
}
} else if (mask_8x8 & 1) {
if ((mask_8x8 & 3) == 3) {
@@ -541,18 +537,18 @@
} else {
if (mask_4x4_int & 1)
aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1);
+ lfi->hev_thr);
else if (mask_4x4_int & 2)
aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
- lfin->lim, lfin->hev_thr, 1);
+ lfin->lim, lfin->hev_thr);
}
count = 2;
} else {
- aom_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+ aom_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
if (mask_4x4_int & 1)
aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1);
+ lfi->hev_thr);
}
} else if (mask_4x4 & 1) {
if ((mask_4x4 & 3) == 3) {
@@ -569,22 +565,22 @@
} else {
if (mask_4x4_int & 1)
aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1);
+ lfi->hev_thr);
else if (mask_4x4_int & 2)
aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
- lfin->lim, lfin->hev_thr, 1);
+ lfin->lim, lfin->hev_thr);
}
count = 2;
} else {
- aom_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+ aom_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
if (mask_4x4_int & 1)
aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1);
+ lfi->hev_thr);
}
} else if (mask_4x4_int & 1) {
aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1);
+ lfi->hev_thr);
}
}
s += 8 * count;
@@ -612,12 +608,12 @@
if (mask & 1) {
if (mask_16x16 & 1) {
if ((mask_16x16 & 3) == 3) {
- aom_highbd_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 2, bd);
+ aom_highbd_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim,
+ lfi->hev_thr, bd);
count = 2;
} else {
- aom_highbd_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1, bd);
+ aom_highbd_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim,
+ lfi->hev_thr, bd);
}
} else if (mask_8x8 & 1) {
if ((mask_8x8 & 3) == 3) {
@@ -635,20 +631,20 @@
} else {
if (mask_4x4_int & 1) {
aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
- lfi->lim, lfi->hev_thr, 1, bd);
+ lfi->lim, lfi->hev_thr, bd);
} else if (mask_4x4_int & 2) {
aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
- lfin->lim, lfin->hev_thr, 1, bd);
+ lfin->lim, lfin->hev_thr, bd);
}
}
count = 2;
} else {
aom_highbd_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1, bd);
+ lfi->hev_thr, bd);
if (mask_4x4_int & 1) {
aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
- lfi->lim, lfi->hev_thr, 1, bd);
+ lfi->lim, lfi->hev_thr, bd);
}
}
} else if (mask_4x4 & 1) {
@@ -666,25 +662,25 @@
} else {
if (mask_4x4_int & 1) {
aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
- lfi->lim, lfi->hev_thr, 1, bd);
+ lfi->lim, lfi->hev_thr, bd);
} else if (mask_4x4_int & 2) {
aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
- lfin->lim, lfin->hev_thr, 1, bd);
+ lfin->lim, lfin->hev_thr, bd);
}
}
count = 2;
} else {
aom_highbd_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1, bd);
+ lfi->hev_thr, bd);
if (mask_4x4_int & 1) {
aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
- lfi->lim, lfi->hev_thr, 1, bd);
+ lfi->lim, lfi->hev_thr, bd);
}
}
} else if (mask_4x4_int & 1) {
aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1, bd);
+ lfi->hev_thr, bd);
}
}
s += 8 * count;
@@ -1088,13 +1084,13 @@
if (mask_16x16 & 1) {
aom_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
} else if (mask_8x8 & 1) {
- aom_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+ aom_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
} else if (mask_4x4 & 1) {
- aom_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+ aom_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
}
}
if (mask_4x4_int & 1)
- aom_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+ aom_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
s += 8;
lfl += 1;
mask_16x16 >>= 1;
@@ -1121,15 +1117,15 @@
bd);
} else if (mask_8x8 & 1) {
aom_highbd_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
- 1, bd);
+ bd);
} else if (mask_4x4 & 1) {
aom_highbd_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
- 1, bd);
+ bd);
}
}
if (mask_4x4_int & 1)
aom_highbd_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim,
- lfi->hev_thr, 1, bd);
+ lfi->hev_thr, bd);
s += 8;
lfl += 1;
mask_16x16 >>= 1;
diff --git a/test/lpf_8_test.cc b/test/lpf_8_test.cc
index 34ed20b..5d814f4 100644
--- a/test/lpf_8_test.cc
+++ b/test/lpf_8_test.cc
@@ -37,121 +37,23 @@
#if CONFIG_AOM_HIGHBITDEPTH
typedef void (*loop_op_t)(uint16_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count, int bd);
+ const uint8_t *limit, const uint8_t *thresh, int bd);
typedef void (*dual_loop_op_t)(uint16_t *s, int p, const uint8_t *blimit0,
const uint8_t *limit0, const uint8_t *thresh0,
const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1, int bd);
#else
typedef void (*loop_op_t)(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int count);
+ const uint8_t *limit, const uint8_t *thresh);
typedef void (*dual_loop_op_t)(uint8_t *s, int p, const uint8_t *blimit0,
const uint8_t *limit0, const uint8_t *thresh0,
const uint8_t *blimit1, const uint8_t *limit1,
const uint8_t *thresh1);
#endif // CONFIG_AOM_HIGHBITDEPTH
-typedef std::tr1::tuple<loop_op_t, loop_op_t, int, int> loop8_param_t;
+typedef std::tr1::tuple<loop_op_t, loop_op_t, int> loop8_param_t;
typedef std::tr1::tuple<dual_loop_op_t, dual_loop_op_t, int> dualloop8_param_t;
-#if HAVE_SSE2
-#if CONFIG_AOM_HIGHBITDEPTH
-void wrapper_vertical_16_sse2(uint16_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/, int bd) {
- aom_highbd_lpf_vertical_16_sse2(s, p, blimit, limit, thresh, bd);
-}
-
-void wrapper_vertical_16_c(uint16_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/, int bd) {
- aom_highbd_lpf_vertical_16_c(s, p, blimit, limit, thresh, bd);
-}
-
-void wrapper_vertical_16_dual_sse2(uint16_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/, int bd) {
- aom_highbd_lpf_vertical_16_dual_sse2(s, p, blimit, limit, thresh, bd);
-}
-
-void wrapper_vertical_16_dual_c(uint16_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/, int bd) {
- aom_highbd_lpf_vertical_16_dual_c(s, p, blimit, limit, thresh, bd);
-}
-#else
-void wrapper_vertical_16_sse2(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_sse2(s, p, blimit, limit, thresh);
-}
-
-void wrapper_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_c(s, p, blimit, limit, thresh);
-}
-
-void wrapper_vertical_16_dual_sse2(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_dual_sse2(s, p, blimit, limit, thresh);
-}
-
-void wrapper_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_dual_c(s, p, blimit, limit, thresh);
-}
-#endif // CONFIG_AOM_HIGHBITDEPTH
-#endif // HAVE_SSE2
-
-#if HAVE_NEON_ASM
-#if CONFIG_AOM_HIGHBITDEPTH
-// No neon high bitdepth functions.
-#else
-void wrapper_vertical_16_neon(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_neon(s, p, blimit, limit, thresh);
-}
-
-void wrapper_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_c(s, p, blimit, limit, thresh);
-}
-
-void wrapper_vertical_16_dual_neon(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_dual_neon(s, p, blimit, limit, thresh);
-}
-
-void wrapper_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_dual_c(s, p, blimit, limit, thresh);
-}
-#endif // CONFIG_AOM_HIGHBITDEPTH
-#endif // HAVE_NEON_ASM
-
-#if HAVE_MSA && (!CONFIG_AOM_HIGHBITDEPTH)
-void wrapper_vertical_16_msa(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_msa(s, p, blimit, limit, thresh);
-}
-
-void wrapper_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit,
- const uint8_t *limit, const uint8_t *thresh,
- int /*count*/) {
- aom_lpf_vertical_16_c(s, p, blimit, limit, thresh);
-}
-#endif // HAVE_MSA && (!CONFIG_AOM_HIGHBITDEPTH)
-
class Loop8Test6Param : public ::testing::TestWithParam<loop8_param_t> {
public:
virtual ~Loop8Test6Param() {}
@@ -159,7 +61,6 @@
loopfilter_op_ = GET_PARAM(0);
ref_loopfilter_op_ = GET_PARAM(1);
bit_depth_ = GET_PARAM(2);
- count_ = GET_PARAM(3);
mask_ = (1 << bit_depth_) - 1;
}
@@ -167,7 +68,6 @@
protected:
int bit_depth_;
- int count_;
int mask_;
loop_op_t loopfilter_op_;
loop_op_t ref_loopfilter_op_;
@@ -251,16 +151,16 @@
ref_s[j] = s[j];
}
#if CONFIG_AOM_HIGHBITDEPTH
- ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh, count_, bd);
+ ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh, bd);
ASM_REGISTER_STATE_CHECK(
- loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, count_, bd));
+ loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, bd));
#else
- ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh, count_);
+ ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh);
ASM_REGISTER_STATE_CHECK(
- loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, count_));
+ loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh));
#endif // CONFIG_AOM_HIGHBITDEPTH
- for (j = 0; j < kNumCoeffs; ++j) {
+ for (int j = 0; j < kNumCoeffs; ++j) {
err_count += ref_s[j] != s[j];
}
if (err_count && !err_count_total) {
@@ -320,13 +220,13 @@
ref_s[j] = s[j];
}
#if CONFIG_AOM_HIGHBITDEPTH
- ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh, count_, bd);
+ ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh, bd);
ASM_REGISTER_STATE_CHECK(
- loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, count_, bd));
+ loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, bd));
#else
- ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh, count_);
+ ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh);
ASM_REGISTER_STATE_CHECK(
- loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, count_));
+ loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh));
#endif // CONFIG_AOM_HIGHBITDEPTH
for (int j = 0; j < kNumCoeffs; ++j) {
err_count += ref_s[j] != s[j];
@@ -424,7 +324,7 @@
ASM_REGISTER_STATE_CHECK(loopfilter_op_(s + 8 + p * 8, p, blimit0, limit0,
thresh0, blimit1, limit1, thresh1));
#endif // CONFIG_AOM_HIGHBITDEPTH
- for (j = 0; j < kNumCoeffs; ++j) {
+ for (int j = 0; j < kNumCoeffs; ++j) {
err_count += ref_s[j] != s[j];
}
if (err_count && !err_count_total) {
@@ -514,73 +414,79 @@
#if CONFIG_AOM_HIGHBITDEPTH
INSTANTIATE_TEST_CASE_P(
SSE2, Loop8Test6Param,
- ::testing::Values(
- make_tuple(&aom_highbd_lpf_horizontal_4_sse2,
- &aom_highbd_lpf_horizontal_4_c, 8, 1),
- make_tuple(&aom_highbd_lpf_vertical_4_sse2,
- &aom_highbd_lpf_vertical_4_c, 8, 1),
- make_tuple(&aom_highbd_lpf_horizontal_8_sse2,
- &aom_highbd_lpf_horizontal_8_c, 8, 1),
- make_tuple(&aom_highbd_lpf_horizontal_16_sse2,
- &aom_highbd_lpf_horizontal_16_c, 8, 1),
- make_tuple(&aom_highbd_lpf_horizontal_16_sse2,
- &aom_highbd_lpf_horizontal_16_c, 8, 2),
- make_tuple(&aom_highbd_lpf_vertical_8_sse2,
- &aom_highbd_lpf_vertical_8_c, 8, 1),
- make_tuple(&wrapper_vertical_16_sse2, &wrapper_vertical_16_c, 8, 1),
- make_tuple(&aom_highbd_lpf_horizontal_4_sse2,
- &aom_highbd_lpf_horizontal_4_c, 10, 1),
- make_tuple(&aom_highbd_lpf_vertical_4_sse2,
- &aom_highbd_lpf_vertical_4_c, 10, 1),
- make_tuple(&aom_highbd_lpf_horizontal_8_sse2,
- &aom_highbd_lpf_horizontal_8_c, 10, 1),
- make_tuple(&aom_highbd_lpf_horizontal_16_sse2,
- &aom_highbd_lpf_horizontal_16_c, 10, 1),
- make_tuple(&aom_highbd_lpf_horizontal_16_sse2,
- &aom_highbd_lpf_horizontal_16_c, 10, 2),
- make_tuple(&aom_highbd_lpf_vertical_8_sse2,
- &aom_highbd_lpf_vertical_8_c, 10, 1),
- make_tuple(&wrapper_vertical_16_sse2, &wrapper_vertical_16_c, 10, 1),
- make_tuple(&aom_highbd_lpf_horizontal_4_sse2,
- &aom_highbd_lpf_horizontal_4_c, 12, 1),
- make_tuple(&aom_highbd_lpf_vertical_4_sse2,
- &aom_highbd_lpf_vertical_4_c, 12, 1),
- make_tuple(&aom_highbd_lpf_horizontal_8_sse2,
- &aom_highbd_lpf_horizontal_8_c, 12, 1),
- make_tuple(&aom_highbd_lpf_horizontal_16_sse2,
- &aom_highbd_lpf_horizontal_16_c, 12, 1),
- make_tuple(&aom_highbd_lpf_horizontal_16_sse2,
- &aom_highbd_lpf_horizontal_16_c, 12, 2),
- make_tuple(&aom_highbd_lpf_vertical_8_sse2,
- &aom_highbd_lpf_vertical_8_c, 12, 1),
- make_tuple(&wrapper_vertical_16_sse2, &wrapper_vertical_16_c, 12, 1),
- make_tuple(&wrapper_vertical_16_dual_sse2, &wrapper_vertical_16_dual_c,
- 8, 1),
- make_tuple(&wrapper_vertical_16_dual_sse2, &wrapper_vertical_16_dual_c,
- 10, 1),
- make_tuple(&wrapper_vertical_16_dual_sse2, &wrapper_vertical_16_dual_c,
- 12, 1)));
+ ::testing::Values(make_tuple(&aom_highbd_lpf_horizontal_4_sse2,
+ &aom_highbd_lpf_horizontal_4_c, 8),
+ make_tuple(&aom_highbd_lpf_vertical_4_sse2,
+ &aom_highbd_lpf_vertical_4_c, 8),
+ make_tuple(&aom_highbd_lpf_horizontal_8_sse2,
+ &aom_highbd_lpf_horizontal_8_c, 8),
+ make_tuple(&aom_highbd_lpf_horizontal_edge_8_sse2,
+ &aom_highbd_lpf_horizontal_edge_8_c, 8),
+ make_tuple(&aom_highbd_lpf_horizontal_edge_16_sse2,
+ &aom_highbd_lpf_horizontal_edge_16_c, 8),
+ make_tuple(&aom_highbd_lpf_vertical_8_sse2,
+ &aom_highbd_lpf_vertical_8_c, 8),
+ make_tuple(&aom_highbd_lpf_vertical_16_sse2,
+ &aom_highbd_lpf_vertical_16_c, 8),
+ make_tuple(&aom_highbd_lpf_horizontal_4_sse2,
+ &aom_highbd_lpf_horizontal_4_c, 10),
+ make_tuple(&aom_highbd_lpf_vertical_4_sse2,
+ &aom_highbd_lpf_vertical_4_c, 10),
+ make_tuple(&aom_highbd_lpf_horizontal_8_sse2,
+ &aom_highbd_lpf_horizontal_8_c, 10),
+ make_tuple(&aom_highbd_lpf_horizontal_edge_8_sse2,
+ &aom_highbd_lpf_horizontal_edge_8_c, 10),
+ make_tuple(&aom_highbd_lpf_horizontal_edge_16_sse2,
+ &aom_highbd_lpf_horizontal_edge_16_c, 10),
+ make_tuple(&aom_highbd_lpf_vertical_8_sse2,
+ &aom_highbd_lpf_vertical_8_c, 10),
+ make_tuple(&aom_highbd_lpf_vertical_16_sse2,
+ &aom_highbd_lpf_vertical_16_c, 10),
+ make_tuple(&aom_highbd_lpf_horizontal_4_sse2,
+ &aom_highbd_lpf_horizontal_4_c, 12),
+ make_tuple(&aom_highbd_lpf_vertical_4_sse2,
+ &aom_highbd_lpf_vertical_4_c, 12),
+ make_tuple(&aom_highbd_lpf_horizontal_8_sse2,
+ &aom_highbd_lpf_horizontal_8_c, 12),
+ make_tuple(&aom_highbd_lpf_horizontal_edge_8_sse2,
+ &aom_highbd_lpf_horizontal_edge_8_c, 12),
+ make_tuple(&aom_highbd_lpf_horizontal_edge_16_sse2,
+ &aom_highbd_lpf_horizontal_edge_16_c, 12),
+ make_tuple(&aom_highbd_lpf_vertical_8_sse2,
+ &aom_highbd_lpf_vertical_8_c, 12),
+ make_tuple(&aom_highbd_lpf_vertical_16_sse2,
+ &aom_highbd_lpf_vertical_16_c, 12),
+ make_tuple(&aom_highbd_lpf_vertical_16_dual_sse2,
+ &aom_highbd_lpf_vertical_16_dual_c, 8),
+ make_tuple(&aom_highbd_lpf_vertical_16_dual_sse2,
+ &aom_highbd_lpf_vertical_16_dual_c, 10),
+ make_tuple(&aom_highbd_lpf_vertical_16_dual_sse2,
+ &aom_highbd_lpf_vertical_16_dual_c, 12)));
#else
INSTANTIATE_TEST_CASE_P(
SSE2, Loop8Test6Param,
::testing::Values(
- make_tuple(&aom_lpf_horizontal_8_sse2, &aom_lpf_horizontal_8_c, 8, 1),
- make_tuple(&aom_lpf_horizontal_16_sse2, &aom_lpf_horizontal_16_c, 8, 1),
- make_tuple(&aom_lpf_horizontal_16_sse2, &aom_lpf_horizontal_16_c, 8, 2),
- make_tuple(&aom_lpf_vertical_8_sse2, &aom_lpf_vertical_8_c, 8, 1),
- make_tuple(&wrapper_vertical_16_sse2, &wrapper_vertical_16_c, 8, 1),
- make_tuple(&wrapper_vertical_16_dual_sse2, &wrapper_vertical_16_dual_c,
- 8, 1)));
+ make_tuple(&aom_lpf_horizontal_4_sse2, &aom_lpf_horizontal_4_c, 8),
+ make_tuple(&aom_lpf_horizontal_8_sse2, &aom_lpf_horizontal_8_c, 8),
+ make_tuple(&aom_lpf_horizontal_edge_8_sse2,
+ &aom_lpf_horizontal_edge_8_c, 8),
+ make_tuple(&aom_lpf_horizontal_edge_16_sse2,
+ &aom_lpf_horizontal_edge_16_c, 8),
+ make_tuple(&aom_lpf_vertical_4_sse2, &aom_lpf_vertical_4_c, 8),
+ make_tuple(&aom_lpf_vertical_8_sse2, &aom_lpf_vertical_8_c, 8),
+ make_tuple(&aom_lpf_vertical_16_sse2, &aom_lpf_vertical_16_c, 8),
+ make_tuple(&aom_lpf_vertical_16_dual_sse2, &aom_lpf_vertical_16_dual_c,
+ 8)));
#endif // CONFIG_AOM_HIGHBITDEPTH
#endif
#if HAVE_AVX2 && (!CONFIG_AOM_HIGHBITDEPTH)
INSTANTIATE_TEST_CASE_P(
AVX2, Loop8Test6Param,
- ::testing::Values(make_tuple(&aom_lpf_horizontal_16_avx2,
- &aom_lpf_horizontal_16_c, 8, 1),
- make_tuple(&aom_lpf_horizontal_16_avx2,
- &aom_lpf_horizontal_16_c, 8, 2)));
+ ::testing::Values(make_tuple(&aom_lpf_horizontal_edge_8_avx2,
+ &aom_lpf_horizontal_edge_8_c, 8),
+ make_tuple(&aom_lpf_horizontal_edge_16_avx2,
+ &aom_lpf_horizontal_edge_16_c, 8)));
#endif
#if HAVE_SSE2
@@ -636,16 +542,18 @@
// Using #if inside the macro is unsupported on MSVS but the tests are
// not
// currently built for MSVS with ARM and NEON.
- make_tuple(&aom_lpf_horizontal_16_neon, &aom_lpf_horizontal_16_c, 8, 1),
- make_tuple(&aom_lpf_horizontal_16_neon, &aom_lpf_horizontal_16_c, 8, 2),
- make_tuple(&wrapper_vertical_16_neon, &wrapper_vertical_16_c, 8, 1),
- make_tuple(&wrapper_vertical_16_dual_neon, &wrapper_vertical_16_dual_c,
- 8, 1),
+ make_tuple(&aom_lpf_horizontal_edge_8_neon,
+ &aom_lpf_horizontal_edge_8_c, 8),
+ make_tuple(&aom_lpf_horizontal_edge_16_neon,
+ &aom_lpf_horizontal_edge_16_c, 8),
+ make_tuple(&aom_lpf_vertical_16_neon, &aom_lpf_vertical_16_c, 8),
+ make_tuple(&aom_lpf_vertical_16_dual_neon, &aom_lpf_vertical_16_dual_c,
+ 8),
#endif // HAVE_NEON_ASM
- make_tuple(&aom_lpf_horizontal_8_neon, &aom_lpf_horizontal_8_c, 8, 1),
- make_tuple(&aom_lpf_vertical_8_neon, &aom_lpf_vertical_8_c, 8, 1),
- make_tuple(&aom_lpf_horizontal_4_neon, &aom_lpf_horizontal_4_c, 8, 1),
- make_tuple(&aom_lpf_vertical_4_neon, &aom_lpf_vertical_4_c, 8, 1)));
+ make_tuple(&aom_lpf_horizontal_8_neon, &aom_lpf_horizontal_8_c, 8),
+ make_tuple(&aom_lpf_vertical_8_neon, &aom_lpf_vertical_8_c, 8),
+ make_tuple(&aom_lpf_horizontal_4_neon, &aom_lpf_horizontal_4_c, 8),
+ make_tuple(&aom_lpf_vertical_4_neon, &aom_lpf_vertical_4_c, 8)));
INSTANTIATE_TEST_CASE_P(NEON, Loop8Test9Param,
::testing::Values(
#if HAVE_NEON_ASM
@@ -661,15 +569,45 @@
#endif // CONFIG_AOM_HIGHBITDEPTH
#endif // HAVE_NEON
+#if HAVE_DSPR2 && !CONFIG_AOM_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+ DSPR2, Loop8Test6Param,
+ ::testing::Values(
+ make_tuple(&aom_lpf_horizontal_4_dspr2, &aom_lpf_horizontal_4_c, 8),
+ make_tuple(&aom_lpf_horizontal_8_dspr2, &aom_lpf_horizontal_8_c, 8),
+ make_tuple(&aom_lpf_horizontal_edge_8, &aom_lpf_horizontal_edge_8, 8),
+ make_tuple(&aom_lpf_horizontal_edge_16, &aom_lpf_horizontal_edge_16, 8),
+ make_tuple(&aom_lpf_vertical_4_dspr2, &aom_lpf_vertical_4_c, 8),
+ make_tuple(&aom_lpf_vertical_8_dspr2, &aom_lpf_vertical_8_c, 8),
+ make_tuple(&aom_lpf_vertical_16_dspr2, &aom_lpf_vertical_16_c, 8),
+ make_tuple(&aom_lpf_vertical_16_dual_dspr2, &aom_lpf_vertical_16_dual_c,
+ 8)));
+
+INSTANTIATE_TEST_CASE_P(
+ DSPR2, Loop8Test9Param,
+ ::testing::Values(make_tuple(&aom_lpf_horizontal_4_dual_dspr2,
+ &aom_lpf_horizontal_4_dual_c, 8),
+ make_tuple(&aom_lpf_horizontal_8_dual_dspr2,
+ &aom_lpf_horizontal_8_dual_c, 8),
+ make_tuple(&aom_lpf_vertical_4_dual_dspr2,
+ &aom_lpf_vertical_4_dual_c, 8),
+ make_tuple(&aom_lpf_vertical_8_dual_dspr2,
+ &aom_lpf_vertical_8_dual_c, 8)));
+#endif // HAVE_DSPR2 && !CONFIG_AOM_HIGHBITDEPTH
+
#if HAVE_MSA && (!CONFIG_AOM_HIGHBITDEPTH)
INSTANTIATE_TEST_CASE_P(
MSA, Loop8Test6Param,
::testing::Values(
- make_tuple(&aom_lpf_horizontal_8_msa, &aom_lpf_horizontal_8_c, 8, 1),
- make_tuple(&aom_lpf_horizontal_16_msa, &aom_lpf_horizontal_16_c, 8, 1),
- make_tuple(&aom_lpf_horizontal_16_msa, &aom_lpf_horizontal_16_c, 8, 2),
- make_tuple(&aom_lpf_vertical_8_msa, &aom_lpf_vertical_8_c, 8, 1),
- make_tuple(&wrapper_vertical_16_msa, &wrapper_vertical_16_c, 8, 1)));
+ make_tuple(&aom_lpf_horizontal_4_msa, &aom_lpf_horizontal_4_c, 8),
+ make_tuple(&aom_lpf_horizontal_8_msa, &aom_lpf_horizontal_8_c, 8),
+ make_tuple(&aom_lpf_horizontal_edge_8_msa, &aom_lpf_horizontal_edge_8_c,
+ 8),
+ make_tuple(&aom_lpf_horizontal_edge_16_msa,
+ &aom_lpf_horizontal_edge_16_c, 8),
+ make_tuple(&aom_lpf_vertical_4_msa, &aom_lpf_vertical_4_c, 8),
+ make_tuple(&aom_lpf_vertical_8_msa, &aom_lpf_vertical_8_c, 8),
+ make_tuple(&aom_lpf_vertical_16_msa, &aom_lpf_vertical_16_c, 8)));
INSTANTIATE_TEST_CASE_P(
MSA, Loop8Test9Param,
