Merge "Added a lossless test"
diff --git a/vp9/common/arm/neon/vp9_loopfilter_neon.asm b/vp9/common/arm/neon/vp9_loopfilter_neon.asm
index 5011315..02cf2f5 100644
--- a/vp9/common/arm/neon/vp9_loopfilter_neon.asm
+++ b/vp9/common/arm/neon/vp9_loopfilter_neon.asm
@@ -10,6 +10,8 @@
EXPORT |vp9_loop_filter_horizontal_edge_neon|
EXPORT |vp9_loop_filter_vertical_edge_neon|
+ EXPORT |vp9_mbloop_filter_horizontal_edge_neon|
+ EXPORT |vp9_mbloop_filter_vertical_edge_neon|
ARM
AREA ||.text||, CODE, READONLY, ALIGN=2
@@ -33,50 +35,47 @@
; sp const uint8_t *thresh,
; sp+4 int count
|vp9_loop_filter_horizontal_edge_neon| PROC
- push {r4-r6, lr}
+ push {lr}
- ldr r12, [sp,#20] ; load count
- ldrb r4, [r2] ; load *blimit
- ldrb r5, [r3] ; load *limit
+ ldr r12, [sp,#8] ; load count
cmp r12, #0
beq end_vp9_lf_h_edge
- ldr r3, [sp, #16] ; load thresh
- vdup.u8 d0, r4 ; duplicate blimit
- ldrb r6, [r3] ; load *thresh
- vdup.u8 d1, r5 ; duplicate limit
- vdup.u8 d2, r6 ; duplicate thresh
+ vld1.8 {d0[]}, [r2] ; duplicate *blimit
+ ldr r2, [sp, #4] ; load thresh
+ vld1.8 {d1[]}, [r3] ; duplicate *limit
+ vld1.8 {d2[]}, [r2] ; duplicate *thresh
count_lf_h_loop
sub r2, r0, r1, lsl #2 ; move src pointer down by 4 lines
- add r6, r2, r1
+ add r3, r2, r1
add r1, r1, r1
vld1.u8 {d3}, [r2@64], r1 ; p3
- vld1.u8 {d4}, [r6@64], r1 ; p2
+ vld1.u8 {d4}, [r3@64], r1 ; p2
vld1.u8 {d5}, [r2@64], r1 ; p1
- vld1.u8 {d6}, [r6@64], r1 ; p0
+ vld1.u8 {d6}, [r3@64], r1 ; p0
vld1.u8 {d7}, [r2@64], r1 ; q0
- vld1.u8 {d16}, [r6@64], r1 ; q1
+ vld1.u8 {d16}, [r3@64], r1 ; q1
vld1.u8 {d17}, [r2@64] ; q2
- vld1.u8 {d18}, [r6@64] ; q3
+ vld1.u8 {d18}, [r3@64] ; q3
sub r2, r2, r1, lsl #1
- sub r6, r6, r1, lsl #1
+ sub r3, r3, r1, lsl #1
bl vp9_loop_filter_neon
vst1.u8 {d4}, [r2@64], r1 ; store op1
- vst1.u8 {d5}, [r6@64], r1 ; store op0
+ vst1.u8 {d5}, [r3@64], r1 ; store op0
vst1.u8 {d6}, [r2@64], r1 ; store oq0
- vst1.u8 {d7}, [r6@64], r1 ; store oq1
+ vst1.u8 {d7}, [r3@64], r1 ; store oq1
add r0, r0, #8
subs r12, r12, #1
bne count_lf_h_loop
end_vp9_lf_h_edge
- pop {r4-r6, pc}
+ pop {pc}
ENDP ; |vp9_loop_filter_horizontal_edge_neon|
; Currently vp9 only works on iterations 8 at a time. The vp8 loop filter
@@ -98,31 +97,28 @@
; sp const uint8_t *thresh,
; sp+4 int count
|vp9_loop_filter_vertical_edge_neon| PROC
- push {r4-r6, lr}
+ push {lr}
- ldr r12, [sp,#20] ; load count
- ldrb r4, [r2] ; load *blimit
- ldrb r5, [r3] ; load *limit
+ ldr r12, [sp,#8] ; load count
cmp r12, #0
beq end_vp9_lf_v_edge
- ldr r3, [sp, #16] ; load thresh
- vdup.u8 d0, r4 ; duplicate blimit
- ldrb r6, [r3] ; load *thresh
- vdup.u8 d1, r5 ; duplicate limit
- vdup.u8 d2, r6 ; duplicate thresh
+ vld1.8 {d0[]}, [r2] ; duplicate *blimit
+ ldr r2, [sp, #4] ; load thresh
+ vld1.8 {d1[]}, [r3] ; duplicate *limit
+ vld1.8 {d2[]}, [r2] ; duplicate *thresh
count_lf_v_loop
- sub r6, r0, #4 ; move s pointer down by 4 columns
+ sub r2, r0, #4 ; move s pointer down by 4 columns
- vld1.u8 {d3}, [r6], r1 ; load s data
- vld1.u8 {d4}, [r6], r1
- vld1.u8 {d5}, [r6], r1
- vld1.u8 {d6}, [r6], r1
- vld1.u8 {d7}, [r6], r1
- vld1.u8 {d16}, [r6], r1
- vld1.u8 {d17}, [r6], r1
- vld1.u8 {d18}, [r6]
+ vld1.u8 {d3}, [r2], r1 ; load s data
+ vld1.u8 {d4}, [r2], r1
+ vld1.u8 {d5}, [r2], r1
+ vld1.u8 {d6}, [r2], r1
+ vld1.u8 {d7}, [r2], r1
+ vld1.u8 {d16}, [r2], r1
+ vld1.u8 {d17}, [r2], r1
+ vld1.u8 {d18}, [r2]
;transpose to 8x16 matrix
vtrn.32 d3, d7
@@ -159,7 +155,7 @@
bne count_lf_v_loop
end_vp9_lf_v_edge
- pop {r4-r6, pc}
+ pop {pc}
ENDP ; |vp9_loop_filter_vertical_edge_neon|
; void vp9_loop_filter_neon();
@@ -224,7 +220,7 @@
vsub.s8 d28, d7, d6 ; ( qs0 - ps0)
- vcge.u8 d17, d0, d17 ; (a > blimit * 2 + limit) * -1
+ vcge.u8 d17, d0, d17 ; a > blimit
vqsub.s8 d27, d5, d16 ; filter = clamp(ps1-qs1)
vorr d22, d21, d22 ; hevmask
@@ -267,4 +263,331 @@
bx lr
ENDP ; |vp9_loop_filter_neon|
+; void vp9_mbloop_filter_horizontal_edge_neon(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,
+; sp+4 int count
+|vp9_mbloop_filter_horizontal_edge_neon| PROC
+ push {lr}
+
+ ldr r12, [sp,#8] ; load count
+ cmp r12, #0
+ beq end_vp9_mblf_h_edge
+
+ vld1.8 {d0[]}, [r2] ; duplicate *blimit
+ ldr r2, [sp, #4] ; load thresh
+ vld1.8 {d1[]}, [r3] ; duplicate *limit
+ vld1.8 {d2[]}, [r2] ; duplicate *thresh
+
+count_mblf_h_loop
+ sub r3, r0, r1, lsl #2 ; move src pointer down by 4 lines
+ add r2, r3, r1
+ add r1, r1, r1
+
+ vld1.u8 {d3}, [r3@64], r1 ; p3
+ vld1.u8 {d4}, [r2@64], r1 ; p2
+ vld1.u8 {d5}, [r3@64], r1 ; p1
+ vld1.u8 {d6}, [r2@64], r1 ; p0
+ vld1.u8 {d7}, [r3@64], r1 ; q0
+ vld1.u8 {d16}, [r2@64], r1 ; q1
+ vld1.u8 {d17}, [r3@64] ; q2
+ vld1.u8 {d18}, [r2@64], r1 ; q3
+
+ sub r3, r3, r1, lsl #1
+ sub r2, r2, r1, lsl #2
+
+ bl vp9_mbloop_filter_neon
+
+ vst1.u8 {d2}, [r2@64], r1 ; store op2
+ vst1.u8 {d3}, [r3@64], r1 ; store op1
+ vst1.u8 {d4}, [r2@64], r1 ; store op0
+ vst1.u8 {d5}, [r3@64], r1 ; store oq0
+ vst1.u8 {d6}, [r2@64], r1 ; store oq1
+ vst1.u8 {d7}, [r3@64], r1 ; store oq2
+
+ add r0, r0, #8
+ subs r12, r12, #1
+ bne count_mblf_h_loop
+
+end_vp9_mblf_h_edge
+ pop {pc}
+
+ ENDP ; |vp9_mbloop_filter_horizontal_edge_neon|
+
+; void vp9_mbloop_filter_vertical_edge_neon(uint8_t *s,
+; int pitch,
+; const uint8_t *blimit,
+; const uint8_t *limit,
+; const uint8_t *thresh,
+; int count)
+;
+; 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
+|vp9_mbloop_filter_vertical_edge_neon| PROC
+ push {lr}
+
+ ldr r12, [sp,#8] ; load count
+ cmp r12, #0
+ beq end_vp9_mblf_v_edge
+
+ vld1.8 {d0[]}, [r2] ; duplicate *blimit
+ ldr r2, [sp, #4] ; load thresh
+ vld1.8 {d1[]}, [r3] ; duplicate *limit
+ vld1.8 {d2[]}, [r2] ; duplicate *thresh
+
+count_mblf_v_loop
+ sub r2, r0, #4 ; move s pointer down by 4 columns
+
+ vld1.u8 {d3}, [r2], r1 ; load s data
+ vld1.u8 {d4}, [r2], r1
+ vld1.u8 {d5}, [r2], r1
+ vld1.u8 {d6}, [r2], r1
+ vld1.u8 {d7}, [r2], r1
+ vld1.u8 {d16}, [r2], r1
+ vld1.u8 {d17}, [r2], r1
+ vld1.u8 {d18}, [r2]
+
+ ;transpose to 8x16 matrix
+ vtrn.32 d3, d7
+ vtrn.32 d4, d16
+ vtrn.32 d5, d17
+ vtrn.32 d6, d18
+
+ vtrn.16 d3, d5
+ vtrn.16 d4, d6
+ vtrn.16 d7, d17
+ vtrn.16 d16, d18
+
+ vtrn.8 d3, d4
+ vtrn.8 d5, d6
+ vtrn.8 d7, d16
+ vtrn.8 d17, d18
+
+ sub r2, r0, #3
+ add r3, r0, #1
+
+ bl vp9_mbloop_filter_neon
+
+ ;store op2, op1, op0, oq0
+ vst4.8 {d2[0], d3[0], d4[0], d5[0]}, [r2], r1
+ vst4.8 {d2[1], d3[1], d4[1], d5[1]}, [r2], r1
+ vst4.8 {d2[2], d3[2], d4[2], d5[2]}, [r2], r1
+ vst4.8 {d2[3], d3[3], d4[3], d5[3]}, [r2], r1
+ vst4.8 {d2[4], d3[4], d4[4], d5[4]}, [r2], r1
+ vst4.8 {d2[5], d3[5], d4[5], d5[5]}, [r2], r1
+ vst4.8 {d2[6], d3[6], d4[6], d5[6]}, [r2], r1
+ vst4.8 {d2[7], d3[7], d4[7], d5[7]}, [r2]
+
+ ;store oq1, oq2
+ vst2.8 {d6[0], d7[0]}, [r3], r1
+ vst2.8 {d6[1], d7[1]}, [r3], r1
+ vst2.8 {d6[2], d7[2]}, [r3], r1
+ vst2.8 {d6[3], d7[3]}, [r3], r1
+ vst2.8 {d6[4], d7[4]}, [r3], r1
+ vst2.8 {d6[5], d7[5]}, [r3], r1
+ vst2.8 {d6[6], d7[6]}, [r3], r1
+ vst2.8 {d6[7], d7[7]}, [r3]
+
+ add r0, r0, r1, lsl #3 ; s += pitch * 8
+ subs r12, r12, #1
+ bne count_mblf_v_loop
+
+end_vp9_mblf_v_edge
+ pop {pc}
+ ENDP ; |vp9_mbloop_filter_vertical_edge_neon|
+
+; void vp9_mbloop_filter_neon();
+; This is a helper function for the loopfilters. The invidual functions do the
+; necessary load, transpose (if necessary) and store. The function does not use
+; registers d8-d15.
+;
+; r0-r3 PRESERVE
+; d0 blimit
+; d1 limit
+; d2 thresh
+; d3 p3
+; d4 p2
+; d5 p1
+; d6 p0
+; d7 q0
+; d16 q1
+; d17 q2
+; d18 q3
+|vp9_mbloop_filter_neon| PROC
+ ; filter_mask
+ vabd.u8 d19, d3, d4 ; abs(p3 - p2)
+ vabd.u8 d20, d4, d5 ; abs(p2 - p1)
+ vabd.u8 d21, d5, d6 ; abs(p1 - p0)
+ vabd.u8 d22, d16, d7 ; abs(q1 - q0)
+ vabd.u8 d23, d17, d16 ; abs(q2 - q1)
+ vabd.u8 d24, d18, d17 ; abs(q3 - q2)
+
+ ; only compare the largest value to limit
+ vmax.u8 d19, d19, d20 ; max(abs(p3 - p2), abs(p2 - p1))
+ vmax.u8 d20, d21, d22 ; max(abs(p1 - p0), abs(q1 - q0))
+ vmax.u8 d23, d23, d24 ; max(abs(q2 - q1), abs(q3 - q2))
+ vmax.u8 d19, d19, d20
+
+ vabd.u8 d24, d6, d7 ; abs(p0 - q0)
+
+ vmax.u8 d19, d19, d23
+
+ vabd.u8 d23, d5, d16 ; a = abs(p1 - q1)
+ vqadd.u8 d24, d24, d24 ; b = abs(p0 - q0) * 2
+
+ ; abs () > limit
+ vcge.u8 d19, d1, d19
+
+ ; flatmask4
+ vabd.u8 d25, d6, d4 ; abs(p0 - p2)
+ vabd.u8 d26, d7, d17 ; abs(q0 - q2)
+ vabd.u8 d27, d3, d6 ; abs(p3 - p0)
+ vabd.u8 d28, d18, d7 ; abs(q3 - q0)
+
+ ; only compare the largest value to thresh
+ vmax.u8 d25, d25, d26 ; max(abs(p0 - p2), abs(q0 - q2))
+ vmax.u8 d26, d27, d28 ; max(abs(p3 - p0), abs(q3 - q0))
+ vmax.u8 d25, d25, d26
+ vmax.u8 d20, d20, d25
+
+ vshr.u8 d23, d23, #1 ; a = a / 2
+ vqadd.u8 d24, d24, d23 ; a = b + a
+
+ vmov.u8 d23, #1
+ vcge.u8 d24, d0, d24 ; a > blimit
+
+ vcge.u8 d20, d23, d20 ; flat
+
+ vand d19, d19, d24 ; mask
+
+ ; hevmask
+ vcgt.u8 d21, d21, d2 ; (abs(p1 - p0) > thresh)*-1
+ vcgt.u8 d22, d22, d2 ; (abs(q1 - q0) > thresh)*-1
+ vorr d21, d21, d22 ; hev
+
+ vmov.u8 d22, #0x80
+
+ ; mbfilter() function
+
+ ; filter() function
+ ; convert to signed
+ veor d23, d7, d22 ; qs0
+ veor d24, d6, d22 ; ps0
+ veor d25, d5, d22 ; ps1
+ veor d26, d16, d22 ; qs1
+
+ vmov.u8 d27, #3
+
+ vsub.s8 d28, d23, d24 ; ( qs0 - ps0)
+
+ vqsub.s8 d29, d25, d26 ; filter = clamp(ps1-qs1)
+
+ vmull.s8 q15, d28, d27 ; 3 * ( qs0 - ps0)
+
+ vand d29, d29, d21 ; filter &= hev
+
+ vaddw.s8 q15, q15, d29 ; filter + 3 * (qs0 - ps0)
+
+ vmov.u8 d29, #4
+
+ ; filter = clamp(filter + 3 * ( qs0 - ps0))
+ vqmovn.s16 d28, q15
+
+ vand d28, d28, d19 ; filter &= mask
+
+ vqadd.s8 d30, d28, d27 ; filter2 = clamp(filter+3)
+ vqadd.s8 d29, d28, d29 ; filter1 = clamp(filter+4)
+ vshr.s8 d30, d30, #3 ; filter2 >>= 3
+ vshr.s8 d29, d29, #3 ; filter1 >>= 3
+
+ vand d20, d20, d19 ; flat & mask
+
+ vqadd.s8 d24, d24, d30 ; op0 = clamp(ps0 + filter2)
+ vqsub.s8 d23, d23, d29 ; oq0 = clamp(qs0 - filter1)
+
+ ; outer tap adjustments: ++filter1 >> 1
+ vrshr.s8 d29, d29, #1
+ vbic d29, d29, d21 ; filter &= ~hev
+
+ vqadd.s8 d25, d25, d29 ; op1 = clamp(ps1 + filter)
+ vqsub.s8 d26, d26, d29 ; oq1 = clamp(qs1 - filter)
+
+ veor d24, d24, d22 ; *f_op0 = u^0x80
+ veor d23, d23, d22 ; *f_oq0 = u^0x80
+ veor d25, d25, d22 ; *f_op1 = u^0x80
+ veor d26, d26, d22 ; *f_oq1 = u^0x80
+
+ ; mbfilter flat && mask branch
+ ; TODO(fgalligan): Can I decrease the cycles shifting to consective d's
+ ; and using vibt on the q's?
+ vmov.u8 d21, #2
+ vaddl.u8 q14, d6, d7 ; op2 = p0 + q0
+ vmlal.u8 q14, d3, d27 ; op2 += p3 * 3
+ vmlal.u8 q14, d4, d21 ; op2 += p2 * 2
+ vaddw.u8 q14, d5 ; op2 += p1
+ vqrshrn.u16 d30, q14, #3 ; r_op2
+
+ vsubw.u8 q14, d3 ; op1 = op2 - p3
+ vsubw.u8 q14, d4 ; op1 -= p2
+ vaddw.u8 q14, d5 ; op1 += p1
+ vaddw.u8 q14, d16 ; op1 += q1
+ vqrshrn.u16 d31, q14, #3 ; r_op1
+
+ vsubw.u8 q14, d3 ; op0 = op1 - p3
+ vsubw.u8 q14, d5 ; op0 -= p1
+ vaddw.u8 q14, d6 ; op0 += p0
+ vaddw.u8 q14, d17 ; op0 += q2
+ vqrshrn.u16 d21, q14, #3 ; r_op0
+
+ vsubw.u8 q14, d3 ; oq0 = op0 - p3
+ vsubw.u8 q14, d6 ; oq0 -= p0
+ vaddw.u8 q14, d7 ; oq0 += q0
+ vaddw.u8 q14, d18 ; oq0 += q3
+ vqrshrn.u16 d22, q14, #3 ; r_oq0
+
+ vsubw.u8 q14, d4 ; oq1 = oq0 - p2
+ vsubw.u8 q14, d7 ; oq1 -= q0
+ vaddw.u8 q14, d16 ; oq1 += q1
+ vaddw.u8 q14, d18 ; oq1 += q3
+ vqrshrn.u16 d0, q14, #3 ; r_oq1
+
+ vsubw.u8 q14, d5 ; oq2 = oq0 - p1
+ vsubw.u8 q14, d16 ; oq2 -= q1
+ vaddw.u8 q14, d17 ; oq2 += q2
+ vaddw.u8 q14, d18 ; oq2 += q3
+ vqrshrn.u16 d1, q14, #3 ; r_oq2
+
+ ; Filter does not set op2 or oq2, so use p2 and q2.
+ vbit d2, d30, d20 ; op2 |= r_op2 & (flat & mask)
+ vbif d2, d4, d20 ; op2 |= op2 & ~(flat & mask)
+
+ vbit d3, d31, d20 ; op1 |= r_op1 & (flat & mask)
+ vbif d3, d25, d20 ; op1 |= f_op1 & ~(flat & mask)
+
+ vbit d4, d21, d20 ; op0 |= r_op0 & (flat & mask)
+ vbif d4, d24, d20 ; op0 |= f_op0 & ~(flat & mask)
+
+ vbit d5, d22, d20 ; oq0 |= r_oq0 & (flat & mask)
+ vbif d5, d23, d20 ; oq0 |= f_oq0 & ~(flat & mask)
+
+ vbit d6, d0, d20 ; oq1 |= r_oq1 & (flat & mask)
+ vbif d6, d26, d20 ; oq1 |= f_oq1 & ~(flat & mask)
+
+ vbit d7, d1, d20 ; oq2 |= r_oq2 & (flat & mask)
+ vbif d7, d17, d20 ; oq2 |= oq2 & ~(flat & mask)
+
+ bx lr
+ ENDP ; |vp9_mbloop_filter_neon|
+
END
diff --git a/vp9/common/vp9_entropymode.c b/vp9/common/vp9_entropymode.c
index 35fb7a4..1cdf138 100644
--- a/vp9/common/vp9_entropymode.c
+++ b/vp9/common/vp9_entropymode.c
@@ -304,14 +304,24 @@
#define MODE_COUNT_SAT 20
#define MODE_MAX_UPDATE_FACTOR 128
+
static int update_mode_ct(vp9_prob pre_prob, vp9_prob prob,
- unsigned int branch_ct[2]) {
- int factor, count = branch_ct[0] + branch_ct[1];
- count = count > MODE_COUNT_SAT ? MODE_COUNT_SAT : count;
- factor = (MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT);
+ unsigned int ct[2]) {
+ const int count = MIN(ct[0] + ct[1], MODE_COUNT_SAT);
+ const int factor = MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT;
return weighted_prob(pre_prob, prob, factor);
}
+static int update_mode_ct2(vp9_prob pre_prob, unsigned int ct[2]) {
+ return update_mode_ct(pre_prob, get_binary_prob(ct[0], ct[1]), ct);
+}
+
+static vp9_prob update_tx_ct(vp9_prob prob, unsigned int ct[2]) {
+ const int count = MIN(ct[0] + ct[1], MODE_COUNT_SAT);
+ const int factor = count * MODE_MAX_UPDATE_FACTOR / MODE_COUNT_SAT;
+ return weighted_prob(prob, get_binary_prob(ct[0], ct[1]), factor);
+}
+
static void update_mode_probs(int n_modes,
const vp9_tree_index *tree, unsigned int *cnt,
vp9_prob *pre_probs, vp9_prob *dst_probs,
@@ -327,11 +337,6 @@
dst_probs[t] = update_mode_ct(pre_probs[t], probs[t], branch_ct[t]);
}
-static int update_mode_ct2(vp9_prob pre_prob, unsigned int branch_ct[2]) {
- return update_mode_ct(pre_prob, get_binary_prob(branch_ct[0],
- branch_ct[1]), branch_ct);
-}
-
// #define MODE_COUNT_TESTING
void vp9_adapt_mode_probs(VP9_COMMON *cm) {
int i, j;
@@ -400,61 +405,39 @@
fc->partition_prob[INTER_FRAME][i], 0);
if (cm->mcomp_filter_type == SWITCHABLE) {
- for (i = 0; i <= VP9_SWITCHABLE_FILTERS; i++) {
+ for (i = 0; i <= VP9_SWITCHABLE_FILTERS; i++)
update_mode_probs(VP9_SWITCHABLE_FILTERS, vp9_switchable_interp_tree,
fc->switchable_interp_count[i],
fc->pre_switchable_interp_prob[i],
fc->switchable_interp_prob[i], 0);
- }
}
+
if (cm->txfm_mode == TX_MODE_SELECT) {
int j;
unsigned int branch_ct_8x8p[TX_SIZE_MAX_SB - 3][2];
unsigned int branch_ct_16x16p[TX_SIZE_MAX_SB - 2][2];
unsigned int branch_ct_32x32p[TX_SIZE_MAX_SB - 1][2];
+
for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
- tx_counts_to_branch_counts_8x8(cm->fc.tx_count_8x8p[i],
- branch_ct_8x8p);
- for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j) {
- int factor;
- int count = branch_ct_8x8p[j][0] + branch_ct_8x8p[j][1];
- vp9_prob prob = get_binary_prob(branch_ct_8x8p[j][0],
- branch_ct_8x8p[j][1]);
- count = count > MODE_COUNT_SAT ? MODE_COUNT_SAT : count;
- factor = (MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT);
- cm->fc.tx_probs_8x8p[i][j] = weighted_prob(
- cm->fc.pre_tx_probs_8x8p[i][j], prob, factor);
- }
- }
- for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
- tx_counts_to_branch_counts_16x16(cm->fc.tx_count_16x16p[i],
+ tx_counts_to_branch_counts_8x8(fc->tx_count_8x8p[i], branch_ct_8x8p);
+ for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j)
+ fc->tx_probs_8x8p[i][j] = update_tx_ct(fc->pre_tx_probs_8x8p[i][j],
+ branch_ct_8x8p[j]);
+
+ tx_counts_to_branch_counts_16x16(fc->tx_count_16x16p[i],
branch_ct_16x16p);
- for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j) {
- int factor;
- int count = branch_ct_16x16p[j][0] + branch_ct_16x16p[j][1];
- vp9_prob prob = get_binary_prob(branch_ct_16x16p[j][0],
- branch_ct_16x16p[j][1]);
- count = count > MODE_COUNT_SAT ? MODE_COUNT_SAT : count;
- factor = (MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT);
- cm->fc.tx_probs_16x16p[i][j] = weighted_prob(
- cm->fc.pre_tx_probs_16x16p[i][j], prob, factor);
- }
- }
- for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
- tx_counts_to_branch_counts_32x32(cm->fc.tx_count_32x32p[i],
+ for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j)
+ fc->tx_probs_16x16p[i][j] = update_tx_ct(fc->pre_tx_probs_16x16p[i][j],
+ branch_ct_16x16p[j]);
+
+ tx_counts_to_branch_counts_32x32(fc->tx_count_32x32p[i],
branch_ct_32x32p);
- for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j) {
- int factor;
- int count = branch_ct_32x32p[j][0] + branch_ct_32x32p[j][1];
- vp9_prob prob = get_binary_prob(branch_ct_32x32p[j][0],
- branch_ct_32x32p[j][1]);
- count = count > MODE_COUNT_SAT ? MODE_COUNT_SAT : count;
- factor = (MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT);
- cm->fc.tx_probs_32x32p[i][j] = weighted_prob(
- cm->fc.pre_tx_probs_32x32p[i][j], prob, factor);
- }
+ for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j)
+ fc->tx_probs_32x32p[i][j] = update_tx_ct(fc->pre_tx_probs_32x32p[i][j],
+ branch_ct_32x32p[j]);
}
}
+
for (i = 0; i < MBSKIP_CONTEXTS; ++i)
fc->mbskip_probs[i] = update_mode_ct2(fc->pre_mbskip_probs[i],
fc->mbskip_count[i]);
diff --git a/vp9/common/vp9_reconintra.c b/vp9/common/vp9_reconintra.c
index 77d8981..cf1b175 100644
--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -34,47 +34,45 @@
};
-static INLINE void d27_predictor(uint8_t *ypred_ptr, int y_stride,
- int bw, int bh,
- uint8_t *yabove_row, uint8_t *yleft_col) {
+void vp9_d27_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
// first column
- for (r = 0; r < bh - 1; ++r) {
+ for (r = 0; r < bs - 1; ++r) {
ypred_ptr[r * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r] +
yleft_col[r + 1], 1);
}
- ypred_ptr[(bh - 1) * y_stride] = yleft_col[bh-1];
+ ypred_ptr[(bs - 1) * y_stride] = yleft_col[bs - 1];
ypred_ptr++;
// second column
- for (r = 0; r < bh - 2; ++r) {
+ for (r = 0; r < bs - 2; ++r) {
ypred_ptr[r * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r] +
yleft_col[r + 1] * 2 +
yleft_col[r + 2], 2);
}
- ypred_ptr[(bh - 2) * y_stride] = ROUND_POWER_OF_TWO(yleft_col[bh - 2] +
- yleft_col[bh - 1] * 3,
+ ypred_ptr[(bs - 2) * y_stride] = ROUND_POWER_OF_TWO(yleft_col[bs - 2] +
+ yleft_col[bs - 1] * 3,
2);
- ypred_ptr[(bh - 1) * y_stride] = yleft_col[bh-1];
+ ypred_ptr[(bs - 1) * y_stride] = yleft_col[bs - 1];
ypred_ptr++;
// rest of last row
- for (c = 0; c < bw - 2; ++c) {
- ypred_ptr[(bh - 1) * y_stride + c] = yleft_col[bh-1];
+ for (c = 0; c < bs - 2; ++c) {
+ ypred_ptr[(bs - 1) * y_stride + c] = yleft_col[bs - 1];
}
- for (r = bh - 2; r >= 0; --r) {
- for (c = 0; c < bw - 2; ++c) {
+ for (r = bs - 2; r >= 0; --r) {
+ for (c = 0; c < bs - 2; ++c) {
ypred_ptr[r * y_stride + c] = ypred_ptr[(r + 1) * y_stride + c - 2];
}
}
}
-static INLINE void d63_predictor(uint8_t *ypred_ptr, int y_stride,
- int bw, int bh,
- uint8_t *yabove_row, uint8_t *yleft_col) {
+void vp9_d63_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
- for (r = 0; r < bh; ++r) {
- for (c = 0; c < bw; ++c) {
+ for (r = 0; r < bs; ++r) {
+ for (c = 0; c < bs; ++c) {
if (r & 1) {
ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[r/2 + c] +
yabove_row[r/2 + c + 1] * 2 +
@@ -88,29 +86,27 @@
}
}
-static INLINE void d45_predictor(uint8_t *ypred_ptr, int y_stride,
- int bw, int bh,
- uint8_t *yabove_row, uint8_t *yleft_col) {
+void vp9_d45_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
- for (r = 0; r < bh; ++r) {
- for (c = 0; c < bw; ++c) {
- if (r + c + 2 < bw * 2)
+ for (r = 0; r < bs; ++r) {
+ for (c = 0; c < bs; ++c) {
+ if (r + c + 2 < bs * 2)
ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[r + c] +
yabove_row[r + c + 1] * 2 +
yabove_row[r + c + 2], 2);
else
- ypred_ptr[c] = yabove_row[bw * 2 - 1];
+ ypred_ptr[c] = yabove_row[bs * 2 - 1];
}
ypred_ptr += y_stride;
}
}
-static INLINE void d117_predictor(uint8_t *ypred_ptr, int y_stride,
- int bw, int bh,
- uint8_t *yabove_row, uint8_t *yleft_col) {
+void vp9_d117_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
// first row
- for (c = 0; c < bw; c++)
+ for (c = 0; c < bs; c++)
ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[c - 1] + yabove_row[c], 1);
ypred_ptr += y_stride;
@@ -118,7 +114,7 @@
ypred_ptr[0] = ROUND_POWER_OF_TWO(yleft_col[0] +
yabove_row[-1] * 2 +
yabove_row[0], 2);
- for (c = 1; c < bw; c++)
+ for (c = 1; c < bs; c++)
ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[c - 2] +
yabove_row[c - 1] * 2 +
yabove_row[c], 2);
@@ -128,27 +124,26 @@
ypred_ptr[0] = ROUND_POWER_OF_TWO(yabove_row[-1] +
yleft_col[0] * 2 +
yleft_col[1], 2);
- for (r = 3; r < bh; ++r)
+ for (r = 3; r < bs; ++r)
ypred_ptr[(r-2) * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r - 3] +
yleft_col[r - 2] * 2 +
yleft_col[r - 1], 2);
// the rest of the block
- for (r = 2; r < bh; ++r) {
- for (c = 1; c < bw; c++)
+ for (r = 2; r < bs; ++r) {
+ for (c = 1; c < bs; c++)
ypred_ptr[c] = ypred_ptr[-2 * y_stride + c - 1];
ypred_ptr += y_stride;
}
}
-static INLINE void d135_predictor(uint8_t *ypred_ptr, int y_stride,
- int bw, int bh,
- uint8_t *yabove_row, uint8_t *yleft_col) {
+void vp9_d135_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
ypred_ptr[0] = ROUND_POWER_OF_TWO(yleft_col[0] +
yabove_row[-1] * 2 +
yabove_row[0], 2);
- for (c = 1; c < bw; c++)
+ for (c = 1; c < bs; c++)
ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[c - 2] +
yabove_row[c - 1] * 2 +
yabove_row[c], 2);
@@ -156,27 +151,24 @@
ypred_ptr[y_stride] = ROUND_POWER_OF_TWO(yabove_row[-1] +
yleft_col[0] * 2 +
yleft_col[1], 2);
- for (r = 2; r < bh; ++r)
+ for (r = 2; r < bs; ++r)
ypred_ptr[r * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r - 2] +
yleft_col[r - 1] * 2 +
yleft_col[r], 2);
ypred_ptr += y_stride;
- for (r = 1; r < bh; ++r) {
- for (c = 1; c < bw; c++)
+ for (r = 1; r < bs; ++r) {
+ for (c = 1; c < bs; c++)
ypred_ptr[c] = ypred_ptr[-y_stride + c - 1];
ypred_ptr += y_stride;
}
}
-static INLINE void d153_predictor(uint8_t *ypred_ptr,
- int y_stride,
- int bw, int bh,
- uint8_t *yabove_row,
- uint8_t *yleft_col) {
+void vp9_d153_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
ypred_ptr[0] = ROUND_POWER_OF_TWO(yabove_row[-1] + yleft_col[0], 1);
- for (r = 1; r < bh; r++)
+ for (r = 1; r < bs; r++)
ypred_ptr[r * y_stride] =
ROUND_POWER_OF_TWO(yleft_col[r - 1] + yleft_col[r], 1);
ypred_ptr++;
@@ -187,33 +179,136 @@
ypred_ptr[y_stride] = ROUND_POWER_OF_TWO(yabove_row[-1] +
yleft_col[0] * 2 +
yleft_col[1], 2);
- for (r = 2; r < bh; r++)
+ for (r = 2; r < bs; r++)
ypred_ptr[r * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r - 2] +
yleft_col[r - 1] * 2 +
yleft_col[r], 2);
ypred_ptr++;
- for (c = 0; c < bw - 2; c++)
+ for (c = 0; c < bs - 2; c++)
ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[c - 1] +
yabove_row[c] * 2 +
yabove_row[c + 1], 2);
ypred_ptr += y_stride;
- for (r = 1; r < bh; ++r) {
- for (c = 0; c < bw - 2; c++)
+ for (r = 1; r < bs; ++r) {
+ for (c = 0; c < bs - 2; c++)
ypred_ptr[c] = ypred_ptr[-y_stride + c - 2];
ypred_ptr += y_stride;
}
}
-void vp9_build_intra_predictors(uint8_t *src, int src_stride,
- uint8_t *ypred_ptr,
- int y_stride, int mode,
- int bw, int bh,
- int up_available, int left_available,
- int right_available) {
- int r, c, i;
- uint8_t yleft_col[64], yabove_data[129], ytop_left;
- uint8_t *yabove_row = yabove_data + 1;
+void vp9_v_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
+ int r;
+
+ for (r = 0; r < bs; r++) {
+ vpx_memcpy(ypred_ptr, yabove_row, bs);
+ ypred_ptr += y_stride;
+ }
+}
+
+void vp9_h_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
+ int r;
+
+ for (r = 0; r < bs; r++) {
+ vpx_memset(ypred_ptr, yleft_col[r], bs);
+ ypred_ptr += y_stride;
+ }
+}
+
+void vp9_tm_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
+ int r, c;
+ int ytop_left = yabove_row[-1];
+
+ for (r = 0; r < bs; r++) {
+ for (c = 0; c < bs; c++)
+ ypred_ptr[c] = clip_pixel(yleft_col[r] + yabove_row[c] - ytop_left);
+ ypred_ptr += y_stride;
+ }
+}
+
+void vp9_dc_128_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
+ int r;
+
+ for (r = 0; r < bs; r++) {
+ vpx_memset(ypred_ptr, 128, bs);
+ ypred_ptr += y_stride;
+ }
+}
+
+void vp9_dc_left_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
+ int i, r;
+ int expected_dc = 128;
+ int average = 0;
+ const int count = bs;
+
+ for (i = 0; i < bs; i++)
+ average += yleft_col[i];
+ expected_dc = (average + (count >> 1)) / count;
+
+ for (r = 0; r < bs; r++) {
+ vpx_memset(ypred_ptr, expected_dc, bs);
+ ypred_ptr += y_stride;
+ }
+}
+
+void vp9_dc_top_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
+ int i, r;
+ int expected_dc = 128;
+ int average = 0;
+ const int count = bs;
+
+ for (i = 0; i < bs; i++)
+ average += yabove_row[i];
+ expected_dc = (average + (count >> 1)) / count;
+
+ for (r = 0; r < bs; r++) {
+ vpx_memset(ypred_ptr, expected_dc, bs);
+ ypred_ptr += y_stride;
+ }
+}
+
+void vp9_dc_predictor_c(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col) {
+ int i, r;
+ int expected_dc = 128;
+ int average = 0;
+ const int count = 2 * bs;
+
+ for (i = 0; i < bs; i++)
+ average += yabove_row[i];
+ for (i = 0; i < bs; i++)
+ average += yleft_col[i];
+ expected_dc = (average + (count >> 1)) / count;
+
+ for (r = 0; r < bs; r++) {
+ vpx_memset(ypred_ptr, expected_dc, bs);
+ ypred_ptr += y_stride;
+ }
+}
+
+typedef void (*intra_pred_fn)(uint8_t *ypred_ptr, int y_stride, int bs,
+ uint8_t *yabove_row, uint8_t *yleft_col);
+
+static void build_intra_predictors(uint8_t *src, int src_stride,
+ uint8_t *ypred_ptr, int y_stride,
+ MB_PREDICTION_MODE mode, int bs,
+ int up_available, int left_available,
+ int right_available) {
+ int i;
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, yleft_col, 64);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, yabove_data, 128 + 16);
+ uint8_t *yabove_row = yabove_data + 16;
+ static const intra_pred_fn pred[VP9_INTRA_MODES] = {
+ NULL, vp9_v_predictor, vp9_h_predictor, vp9_d45_predictor,
+ vp9_d135_predictor, vp9_d117_predictor, vp9_d153_predictor,
+ vp9_d27_predictor, vp9_d63_predictor, vp9_tm_predictor
+ };
// 127 127 127 .. 127 127 127 127 127 127
// 129 A B .. Y Z
@@ -222,94 +317,44 @@
// 129 G H .. S T T T T T
// ..
- assert(bw == bh);
-
if (left_available) {
- for (i = 0; i < bh; i++)
+ for (i = 0; i < bs; i++)
yleft_col[i] = src[i * src_stride - 1];
} else {
- vpx_memset(yleft_col, 129, bh);
+ vpx_memset(yleft_col, 129, bs);
}
if (up_available) {
uint8_t *yabove_ptr = src - src_stride;
- vpx_memcpy(yabove_row, yabove_ptr, bw);
- if (bw == 4 && right_available)
- vpx_memcpy(yabove_row + bw, yabove_ptr + bw, bw);
- else
- vpx_memset(yabove_row + bw, yabove_row[bw -1], bw);
- ytop_left = left_available ? yabove_ptr[-1] : 129;
- } else {
- vpx_memset(yabove_row, 127, bw * 2);
- ytop_left = 127;
- }
- yabove_row[-1] = ytop_left;
-
- switch (mode) {
- case DC_PRED: {
- int i;
- int expected_dc = 128;
- int average = 0;
- int count = 0;
-
- if (up_available || left_available) {
- if (up_available) {
- for (i = 0; i < bw; i++)
- average += yabove_row[i];
- count += bw;
- }
- if (left_available) {
- for (i = 0; i < bh; i++)
- average += yleft_col[i];
- count += bh;
- }
- expected_dc = (average + (count >> 1)) / count;
- }
- for (r = 0; r < bh; r++) {
- vpx_memset(ypred_ptr, expected_dc, bw);
- ypred_ptr += y_stride;
- }
+ if (bs == 4 && right_available && left_available) {
+ yabove_row = yabove_ptr;
+ } else {
+ vpx_memcpy(yabove_row, yabove_ptr, bs);
+ if (bs == 4 && right_available)
+ vpx_memcpy(yabove_row + bs, yabove_ptr + bs, bs);
+ else
+ vpx_memset(yabove_row + bs, yabove_row[bs - 1], bs);
+ yabove_row[-1] = left_available ? yabove_ptr[-1] : 129;
}
- break;
- case V_PRED:
- for (r = 0; r < bh; r++) {
- vpx_memcpy(ypred_ptr, yabove_row, bw);
- ypred_ptr += y_stride;
+ } else {
+ vpx_memset(yabove_row, 127, bs * 2);
+ yabove_row[-1] = 127;
+ }
+
+ if (mode == DC_PRED) {
+ if (left_available) {
+ if (up_available) {
+ vp9_dc_predictor(ypred_ptr, y_stride, bs, yabove_row, yleft_col);
+ } else {
+ vp9_dc_left_predictor(ypred_ptr, y_stride, bs, yabove_row, yleft_col);
}
- break;
- case H_PRED:
- for (r = 0; r < bh; r++) {
- vpx_memset(ypred_ptr, yleft_col[r], bw);
- ypred_ptr += y_stride;
- }
- break;
- case TM_PRED:
- for (r = 0; r < bh; r++) {
- for (c = 0; c < bw; c++)
- ypred_ptr[c] = clip_pixel(yleft_col[r] + yabove_row[c] - ytop_left);
- ypred_ptr += y_stride;
- }
- break;
- case D45_PRED:
- d45_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
- break;
- case D135_PRED:
- d135_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
- break;
- case D117_PRED:
- d117_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
- break;
- case D153_PRED:
- d153_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
- break;
- case D27_PRED:
- d27_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
- break;
- case D63_PRED:
- d63_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
- break;
- default:
- break;
+ } else if (up_available) {
+ vp9_dc_top_predictor(ypred_ptr, y_stride, bs, yabove_row, yleft_col);
+ } else {
+ vp9_dc_128_predictor(ypred_ptr, y_stride, bs, yabove_row, yleft_col);
+ }
+ } else {
+ pred[mode](ypred_ptr, y_stride, bs, yabove_row, yleft_col);
}
}
@@ -328,11 +373,10 @@
const int txfm_block_size = 4 << tx_size;
assert(bwl >= 0);
- vp9_build_intra_predictors(reference, ref_stride,
- predictor, pre_stride,
- mode,
- txfm_block_size,
- txfm_block_size,
- have_top, have_left,
- have_right);
+ build_intra_predictors(reference, ref_stride,
+ predictor, pre_stride,
+ mode,
+ txfm_block_size,
+ have_top, have_left,
+ have_right);
}
diff --git a/vp9/common/vp9_rtcd_defs.sh b/vp9/common/vp9_rtcd_defs.sh
index 330c60f..c6167d2 100644
--- a/vp9/common/vp9_rtcd_defs.sh
+++ b/vp9/common/vp9_rtcd_defs.sh
@@ -35,8 +35,6 @@
prototype void vp9_idct_add "int16_t *input, uint8_t *dest, int stride, int eob"
specialize vp9_idct_add
-
-
prototype void vp9_idct_add_32x32 "int16_t *q, uint8_t *dst, int stride, int eob"
specialize vp9_idct_add_32x32
@@ -54,8 +52,44 @@
prototype void vp9_copy_mem8x4 "const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch"
specialize vp9_copy_mem8x4 mmx
-prototype void vp9_build_intra_predictors "uint8_t *src, int src_stride, uint8_t *pred, int y_stride, int mode, int bw, int bh, int up_available, int left_available, int right_available"
-specialize void vp9_build_intra_predictors
+prototype void vp9_d27_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_d27_predictor
+
+prototype void vp9_d45_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_d45_predictor
+
+prototype void vp9_d63_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_d63_predictor
+
+prototype void vp9_h_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_h_predictor
+
+prototype void vp9_d117_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_d117_predictor
+
+prototype void vp9_d135_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_d135_predictor
+
+prototype void vp9_d153_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_d153_predictor
+
+prototype void vp9_v_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_v_predictor
+
+prototype void vp9_tm_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_tm_predictor
+
+prototype void vp9_dc_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_dc_predictor
+
+prototype void vp9_dc_top_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_dc_top_predictor
+
+prototype void vp9_dc_left_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_dc_left_predictor
+
+prototype void vp9_dc_128_predictor "uint8_t *ypred_ptr, int y_stride, int bs, uint8_t *yabove_row, uint8_t *yleft_col"
+specialize vp9_dc_128_predictor
if [ "$CONFIG_VP9_DECODER" = "yes" ]; then
prototype void vp9_add_constant_residual_8x8 "const int16_t diff, uint8_t *dest, int stride"
@@ -75,7 +109,7 @@
specialize vp9_mb_lpf_vertical_edge_w sse2
prototype void vp9_mbloop_filter_vertical_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
-specialize vp9_mbloop_filter_vertical_edge sse2
+specialize vp9_mbloop_filter_vertical_edge sse2 neon
prototype void vp9_loop_filter_vertical_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
specialize vp9_loop_filter_vertical_edge mmx neon
@@ -84,7 +118,7 @@
specialize vp9_mb_lpf_horizontal_edge_w sse2
prototype void vp9_mbloop_filter_horizontal_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
-specialize vp9_mbloop_filter_horizontal_edge sse2
+specialize vp9_mbloop_filter_horizontal_edge sse2 neon
prototype void vp9_loop_filter_horizontal_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
specialize vp9_loop_filter_horizontal_edge mmx neon
diff --git a/vp9/common/x86/vp9_recon_sse2.asm b/vp9/common/x86/vp9_recon_sse2.asm
index 9ee3043..f7cc611 100644
--- a/vp9/common/x86/vp9_recon_sse2.asm
+++ b/vp9/common/x86/vp9_recon_sse2.asm
@@ -113,460 +113,3 @@
UNSHADOW_ARGS
pop rbp
ret
-
-
-;void vp9_intra_pred_uv_dc_mmx2(
-; unsigned char *dst,
-; int dst_stride
-; unsigned char *src,
-; int src_stride,
-; )
-global sym(vp9_intra_pred_uv_dc_mmx2) PRIVATE
-sym(vp9_intra_pred_uv_dc_mmx2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- push rsi
- push rdi
- ; end prolog
-
- ; from top
- mov rsi, arg(2) ;src;
- movsxd rax, dword ptr arg(3) ;src_stride;
- sub rsi, rax
- pxor mm0, mm0
- movq mm1, [rsi]
- psadbw mm1, mm0
-
- ; from left
- dec rsi
- lea rdi, [rax*3]
- movzx ecx, byte [rsi+rax]
- movzx edx, byte [rsi+rax*2]
- add ecx, edx
- movzx edx, byte [rsi+rdi]
- add ecx, edx
- lea rsi, [rsi+rax*4]
- movzx edx, byte [rsi]
- add ecx, edx
- movzx edx, byte [rsi+rax]
- add ecx, edx
- movzx edx, byte [rsi+rax*2]
- add ecx, edx
- movzx edx, byte [rsi+rdi]
- add ecx, edx
- movzx edx, byte [rsi+rax*4]
- add ecx, edx
-
- ; add up
- pextrw edx, mm1, 0x0
- lea edx, [edx+ecx+8]
- sar edx, 4
- movd mm1, edx
- pshufw mm1, mm1, 0x0
- packuswb mm1, mm1
-
- ; write out
- mov rdi, arg(0) ;dst;
- movsxd rcx, dword ptr arg(1) ;dst_stride
- lea rax, [rcx*3]
-
- movq [rdi ], mm1
- movq [rdi+rcx ], mm1
- movq [rdi+rcx*2], mm1
- movq [rdi+rax ], mm1
- lea rdi, [rdi+rcx*4]
- movq [rdi ], mm1
- movq [rdi+rcx ], mm1
- movq [rdi+rcx*2], mm1
- movq [rdi+rax ], mm1
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_intra_pred_uv_dctop_mmx2(
-; unsigned char *dst,
-; int dst_stride
-; unsigned char *src,
-; int src_stride,
-; )
-global sym(vp9_intra_pred_uv_dctop_mmx2) PRIVATE
-sym(vp9_intra_pred_uv_dctop_mmx2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ; from top
- mov rsi, arg(2) ;src;
- movsxd rax, dword ptr arg(3) ;src_stride;
- sub rsi, rax
- pxor mm0, mm0
- movq mm1, [rsi]
- psadbw mm1, mm0
-
- ; add up
- paddw mm1, [GLOBAL(dc_4)]
- psraw mm1, 3
- pshufw mm1, mm1, 0x0
- packuswb mm1, mm1
-
- ; write out
- mov rdi, arg(0) ;dst;
- movsxd rcx, dword ptr arg(1) ;dst_stride
- lea rax, [rcx*3]
-
- movq [rdi ], mm1
- movq [rdi+rcx ], mm1
- movq [rdi+rcx*2], mm1
- movq [rdi+rax ], mm1
- lea rdi, [rdi+rcx*4]
- movq [rdi ], mm1
- movq [rdi+rcx ], mm1
- movq [rdi+rcx*2], mm1
- movq [rdi+rax ], mm1
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_intra_pred_uv_dcleft_mmx2(
-; unsigned char *dst,
-; int dst_stride
-; unsigned char *src,
-; int src_stride,
-; )
-global sym(vp9_intra_pred_uv_dcleft_mmx2) PRIVATE
-sym(vp9_intra_pred_uv_dcleft_mmx2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- push rsi
- push rdi
- ; end prolog
-
- ; from left
- mov rsi, arg(2) ;src;
- movsxd rax, dword ptr arg(3) ;src_stride;
- dec rsi
- lea rdi, [rax*3]
- movzx ecx, byte [rsi]
- movzx edx, byte [rsi+rax]
- add ecx, edx
- movzx edx, byte [rsi+rax*2]
- add ecx, edx
- movzx edx, byte [rsi+rdi]
- add ecx, edx
- lea rsi, [rsi+rax*4]
- movzx edx, byte [rsi]
- add ecx, edx
- movzx edx, byte [rsi+rax]
- add ecx, edx
- movzx edx, byte [rsi+rax*2]
- add ecx, edx
- movzx edx, byte [rsi+rdi]
- lea edx, [ecx+edx+4]
-
- ; add up
- shr edx, 3
- movd mm1, edx
- pshufw mm1, mm1, 0x0
- packuswb mm1, mm1
-
- ; write out
- mov rdi, arg(0) ;dst;
- movsxd rcx, dword ptr arg(1) ;dst_stride
- lea rax, [rcx*3]
-
- movq [rdi ], mm1
- movq [rdi+rcx ], mm1
- movq [rdi+rcx*2], mm1
- movq [rdi+rax ], mm1
- lea rdi, [rdi+rcx*4]
- movq [rdi ], mm1
- movq [rdi+rcx ], mm1
- movq [rdi+rcx*2], mm1
- movq [rdi+rax ], mm1
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_intra_pred_uv_dc128_mmx(
-; unsigned char *dst,
-; int dst_stride
-; unsigned char *src,
-; int src_stride,
-; )
-global sym(vp9_intra_pred_uv_dc128_mmx) PRIVATE
-sym(vp9_intra_pred_uv_dc128_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- GET_GOT rbx
- ; end prolog
-
- ; write out
- movq mm1, [GLOBAL(dc_128)]
- mov rax, arg(0) ;dst;
- movsxd rdx, dword ptr arg(1) ;dst_stride
- lea rcx, [rdx*3]
-
- movq [rax ], mm1
- movq [rax+rdx ], mm1
- movq [rax+rdx*2], mm1
- movq [rax+rcx ], mm1
- lea rax, [rax+rdx*4]
- movq [rax ], mm1
- movq [rax+rdx ], mm1
- movq [rax+rdx*2], mm1
- movq [rax+rcx ], mm1
-
- ; begin epilog
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_intra_pred_uv_tm_sse2(
-; unsigned char *dst,
-; int dst_stride
-; unsigned char *src,
-; int src_stride,
-; )
-%macro vp9_intra_pred_uv_tm 1
-global sym(vp9_intra_pred_uv_tm_%1) PRIVATE
-sym(vp9_intra_pred_uv_tm_%1):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ; read top row
- mov edx, 4
- mov rsi, arg(2) ;src;
- movsxd rax, dword ptr arg(3) ;src_stride;
- sub rsi, rax
- pxor xmm0, xmm0
-%ifidn %1, ssse3
- movdqa xmm2, [GLOBAL(dc_1024)]
-%endif
- movq xmm1, [rsi]
- punpcklbw xmm1, xmm0
-
- ; set up left ptrs ans subtract topleft
- movd xmm3, [rsi-1]
- lea rsi, [rsi+rax-1]
-%ifidn %1, sse2
- punpcklbw xmm3, xmm0
- pshuflw xmm3, xmm3, 0x0
- punpcklqdq xmm3, xmm3
-%else
- pshufb xmm3, xmm2
-%endif
- psubw xmm1, xmm3
-
- ; set up dest ptrs
- mov rdi, arg(0) ;dst;
- movsxd rcx, dword ptr arg(1) ;dst_stride
-
-.vp9_intra_pred_uv_tm_%1_loop:
- movd xmm3, [rsi]
- movd xmm5, [rsi+rax]
-%ifidn %1, sse2
- punpcklbw xmm3, xmm0
- punpcklbw xmm5, xmm0
- pshuflw xmm3, xmm3, 0x0
- pshuflw xmm5, xmm5, 0x0
- punpcklqdq xmm3, xmm3
- punpcklqdq xmm5, xmm5
-%else
- pshufb xmm3, xmm2
- pshufb xmm5, xmm2
-%endif
- paddw xmm3, xmm1
- paddw xmm5, xmm1
- packuswb xmm3, xmm5
- movq [rdi ], xmm3
- movhps[rdi+rcx], xmm3
- lea rsi, [rsi+rax*2]
- lea rdi, [rdi+rcx*2]
- dec edx
- jnz .vp9_intra_pred_uv_tm_%1_loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-%endmacro
-
-vp9_intra_pred_uv_tm sse2
-vp9_intra_pred_uv_tm ssse3
-
-;void vp9_intra_pred_uv_ve_mmx(
-; unsigned char *dst,
-; int dst_stride
-; unsigned char *src,
-; int src_stride,
-; )
-global sym(vp9_intra_pred_uv_ve_mmx) PRIVATE
-sym(vp9_intra_pred_uv_ve_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- ; end prolog
-
- ; read from top
- mov rax, arg(2) ;src;
- movsxd rdx, dword ptr arg(3) ;src_stride;
- sub rax, rdx
- movq mm1, [rax]
-
- ; write out
- mov rax, arg(0) ;dst;
- movsxd rdx, dword ptr arg(1) ;dst_stride
- lea rcx, [rdx*3]
-
- movq [rax ], mm1
- movq [rax+rdx ], mm1
- movq [rax+rdx*2], mm1
- movq [rax+rcx ], mm1
- lea rax, [rax+rdx*4]
- movq [rax ], mm1
- movq [rax+rdx ], mm1
- movq [rax+rdx*2], mm1
- movq [rax+rcx ], mm1
-
- ; begin epilog
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_intra_pred_uv_ho_mmx2(
-; unsigned char *dst,
-; int dst_stride
-; unsigned char *src,
-; int src_stride,
-; )
-%macro vp9_intra_pred_uv_ho 1
-global sym(vp9_intra_pred_uv_ho_%1) PRIVATE
-sym(vp9_intra_pred_uv_ho_%1):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- push rsi
- push rdi
-%ifidn %1, ssse3
-%ifndef GET_GOT_SAVE_ARG
- push rbx
-%endif
- GET_GOT rbx
-%endif
- ; end prolog
-
- ; read from left and write out
-%ifidn %1, mmx2
- mov edx, 4
-%endif
- mov rsi, arg(2) ;src;
- movsxd rax, dword ptr arg(3) ;src_stride;
- mov rdi, arg(0) ;dst;
- movsxd rcx, dword ptr arg(1) ;dst_stride
-%ifidn %1, ssse3
- lea rdx, [rcx*3]
- movdqa xmm2, [GLOBAL(dc_00001111)]
- lea rbx, [rax*3]
-%endif
- dec rsi
-%ifidn %1, mmx2
-.vp9_intra_pred_uv_ho_%1_loop:
- movd mm0, [rsi]
- movd mm1, [rsi+rax]
- punpcklbw mm0, mm0
- punpcklbw mm1, mm1
- pshufw mm0, mm0, 0x0
- pshufw mm1, mm1, 0x0
- movq [rdi ], mm0
- movq [rdi+rcx], mm1
- lea rsi, [rsi+rax*2]
- lea rdi, [rdi+rcx*2]
- dec edx
- jnz .vp9_intra_pred_uv_ho_%1_loop
-%else
- movd xmm0, [rsi]
- movd xmm3, [rsi+rax]
- movd xmm1, [rsi+rax*2]
- movd xmm4, [rsi+rbx]
- punpcklbw xmm0, xmm3
- punpcklbw xmm1, xmm4
- pshufb xmm0, xmm2
- pshufb xmm1, xmm2
- movq [rdi ], xmm0
- movhps [rdi+rcx], xmm0
- movq [rdi+rcx*2], xmm1
- movhps [rdi+rdx], xmm1
- lea rsi, [rsi+rax*4]
- lea rdi, [rdi+rcx*4]
- movd xmm0, [rsi]
- movd xmm3, [rsi+rax]
- movd xmm1, [rsi+rax*2]
- movd xmm4, [rsi+rbx]
- punpcklbw xmm0, xmm3
- punpcklbw xmm1, xmm4
- pshufb xmm0, xmm2
- pshufb xmm1, xmm2
- movq [rdi ], xmm0
- movhps [rdi+rcx], xmm0
- movq [rdi+rcx*2], xmm1
- movhps [rdi+rdx], xmm1
-%endif
-
- ; begin epilog
-%ifidn %1, ssse3
- RESTORE_GOT
-%ifndef GET_GOT_SAVE_ARG
- pop rbx
-%endif
-%endif
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-%endmacro
-
-vp9_intra_pred_uv_ho mmx2
-vp9_intra_pred_uv_ho ssse3
-
-SECTION_RODATA
-dc_128:
- times 8 db 128
-dc_4:
- times 4 dw 4
-align 16
-dc_1024:
- times 8 dw 0x400
-align 16
-dc_00001111:
- times 8 db 0
- times 8 db 1
diff --git a/vp9/common/x86/vp9_recon_wrapper_sse2.c b/vp9/common/x86/vp9_recon_wrapper_sse2.c
deleted file mode 100644
index 97148fb..0000000
--- a/vp9/common/x86/vp9_recon_wrapper_sse2.c
+++ /dev/null
@@ -1,101 +0,0 @@
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vpx_config.h"
-#include "vpx_mem/vpx_mem.h"
-#include "vp9/common/vp9_blockd.h"
-
-#define build_intra_predictors_mbuv_prototype(sym) \
- void sym(unsigned char *dst, int dst_stride, \
- const unsigned char *src, int src_stride)
-typedef build_intra_predictors_mbuv_prototype((*build_intra_pred_mbuv_fn_t));
-
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_dc_mmx2);
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_dctop_mmx2);
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_dcleft_mmx2);
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_dc128_mmx);
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_ho_mmx2);
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_ho_ssse3);
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_ve_mmx);
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_tm_sse2);
-extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_tm_ssse3);
-
-static void build_intra_predictors_mbuv_x86(MACROBLOCKD *xd,
- unsigned char *dst_u,
- unsigned char *dst_v,
- int dst_stride,
- build_intra_pred_mbuv_fn_t tm_fn,
- build_intra_pred_mbuv_fn_t ho_fn) {
- int mode = xd->mode_info_context->mbmi.uv_mode;
- build_intra_pred_mbuv_fn_t fn;
- int src_stride = xd->plane[1].dst.stride;
-
- switch (mode) {
- case V_PRED:
- fn = vp9_intra_pred_uv_ve_mmx;
- break;
- case H_PRED:
- fn = ho_fn;
- break;
- case TM_PRED:
- fn = tm_fn;
- break;
- case DC_PRED:
- if (xd->up_available) {
- if (xd->left_available) {
- fn = vp9_intra_pred_uv_dc_mmx2;
- break;
- } else {
- fn = vp9_intra_pred_uv_dctop_mmx2;
- break;
- }
- } else if (xd->left_available) {
- fn = vp9_intra_pred_uv_dcleft_mmx2;
- break;
- } else {
- fn = vp9_intra_pred_uv_dc128_mmx;
- break;
- }
- break;
- default:
- return;
- }
-
- fn(dst_u, dst_stride, xd->plane[1].dst.buf, src_stride);
- fn(dst_v, dst_stride, xd->plane[2].dst.buf, src_stride);
-}
-
-void vp9_build_intra_predictors_mbuv_sse2(MACROBLOCKD *xd) {
- build_intra_predictors_mbuv_x86(xd, xd->plane[1].dst.buf,
- xd->plane[2].dst.buf, xd->plane[1].dst.stride,
- vp9_intra_pred_uv_tm_sse2,
- vp9_intra_pred_uv_ho_mmx2);
-}
-
-void vp9_build_intra_predictors_mbuv_ssse3(MACROBLOCKD *xd) {
- build_intra_predictors_mbuv_x86(xd, xd->plane[1].dst.buf,
- xd->plane[2].dst.buf, xd->plane[1].dst.stride,
- vp9_intra_pred_uv_tm_ssse3,
- vp9_intra_pred_uv_ho_ssse3);
-}
-
-void vp9_build_intra_predictors_mbuv_s_sse2(MACROBLOCKD *xd) {
- build_intra_predictors_mbuv_x86(xd, xd->plane[1].dst.buf,
- xd->plane[2].dst.buf, xd->plane[1].dst.stride,
- vp9_intra_pred_uv_tm_sse2,
- vp9_intra_pred_uv_ho_mmx2);
-}
-
-void vp9_build_intra_predictors_mbuv_s_ssse3(MACROBLOCKD *xd) {
- build_intra_predictors_mbuv_x86(xd, xd->plane[1].dst.buf,
- xd->plane[2].dst.buf, xd->plane[1].dst.stride,
- vp9_intra_pred_uv_tm_ssse3,
- vp9_intra_pred_uv_ho_ssse3);
-}
diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index 875561b..9a547c9 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -367,13 +367,14 @@
VP9_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &pbi->mb;
const BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
- const int pred_segment_id = vp9_get_segment_id(cm, cm->last_frame_seg_map,
- bsize, mi_row, mi_col);
+ int pred_segment_id;
int segment_id;
if (!xd->segmentation_enabled)
return 0; // Default for disabled segmentation
+ pred_segment_id = vp9_get_segment_id(cm, cm->last_frame_seg_map,
+ bsize, mi_row, mi_col);
if (!xd->update_mb_segmentation_map)
return pred_segment_id;
@@ -481,10 +482,6 @@
int idx, idy;
- // Make sure the MACROBLOCKD mode info pointer is pointed at the
- // correct entry for the current macroblock.
- xd->mode_info_context = mi;
-
// Distance of Mb to the various image edges.
// These specified to 8th pel as they are always compared to MV values
// that are in 1/8th pel units
@@ -748,21 +745,18 @@
MACROBLOCKD *const xd = &pbi->mb;
MODE_INFO *mi = xd->mode_info_context;
const BLOCK_SIZE_TYPE bsize = mi->mbmi.sb_type;
+ const int bw = 1 << mi_width_log2(bsize);
+ const int bh = 1 << mi_height_log2(bsize);
+ const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+ const int x_mis = MIN(bw, cm->mi_cols - mi_col);
+ int x, y;
if (cm->frame_type == KEY_FRAME || cm->intra_only)
read_intra_mode_info(pbi, mi, mi_row, mi_col, r);
else
read_inter_mode_info(pbi, mi, mi_row, mi_col, r);
- if (1) {
- const int bw = 1 << mi_width_log2(bsize);
- const int bh = 1 << mi_height_log2(bsize);
- const int y_mis = MIN(bh, cm->mi_rows - mi_row);
- const int x_mis = MIN(bw, cm->mi_cols - mi_col);
- int x, y;
-
- for (y = 0; y < y_mis; y++)
- for (x = !y; x < x_mis; x++)
- mi[y * cm->mode_info_stride + x] = *mi;
- }
+ for (y = 0; y < y_mis; y++)
+ for (x = !y; x < x_mis; x++)
+ mi[y * cm->mode_info_stride + x] = *mi;
}
diff --git a/vp9/encoder/vp9_mbgraph.c b/vp9/encoder/vp9_mbgraph.c
index 93c2605..9637393 100644
--- a/vp9/encoder/vp9_mbgraph.c
+++ b/vp9/encoder/vp9_mbgraph.c
@@ -15,6 +15,7 @@
#include <vp9/encoder/vp9_rdopt.h>
#include <vp9/common/vp9_blockd.h>
#include <vp9/common/vp9_reconinter.h>
+#include <vp9/common/vp9_reconintra.h>
#include <vp9/common/vp9_systemdependent.h>
#include <vp9/encoder/vp9_segmentation.h>
@@ -146,16 +147,11 @@
// we're intentionally not doing 4x4, we just want a rough estimate
for (mode = DC_PRED; mode <= TM_PRED; mode++) {
unsigned int err;
- const int bwl = b_width_log2(BLOCK_SIZE_MB16X16), bw = 4 << bwl;
- const int bhl = b_height_log2(BLOCK_SIZE_MB16X16), bh = 4 << bhl;
xd->mode_info_context->mbmi.mode = mode;
- vp9_build_intra_predictors(x->plane[0].src.buf, x->plane[0].src.stride,
- xd->plane[0].dst.buf, xd->plane[0].dst.stride,
- xd->mode_info_context->mbmi.mode,
- bw, bh,
- xd->up_available, xd->left_available,
- xd->right_available);
+ vp9_predict_intra_block(xd, 0, 2, TX_16X16, mode,
+ x->plane[0].src.buf, x->plane[0].src.stride,
+ xd->plane[0].dst.buf, xd->plane[0].dst.stride);
err = vp9_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
xd->plane[0].dst.buf, xd->plane[0].dst.stride, best_err);
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 7d7c9de..215d354 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -3506,7 +3506,7 @@
}
}
} else if (mb_skip_allowed && ref_frame != INTRA_FRAME &&
- this_mode != SPLITMV) {
+ this_mode != SPLITMV && !xd->lossless) {
if (RDCOST(x->rdmult, x->rddiv, rate_y + rate_uv, distortion2) <
RDCOST(x->rdmult, x->rddiv, 0, total_sse)) {
// Add in the cost of the no skip flag.
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index bdcdb99..587b1fc 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -80,7 +80,6 @@
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_iwalsh_sse2.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_loopfilter_sse2.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_recon_sse2.asm
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_recon_wrapper_sse2.c
VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_ssse3.asm
ifeq ($(CONFIG_POSTPROC),yes)
VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_postproc_mmx.asm
