Files
pcsx2/.github/workflows/scripts/windows/libsvtav1-arm64-msvc.patch
T
2026-09-19 13:07:54 -04:00

606 lines
31 KiB
Diff

diff --git a/CMakeLists.txt b/CMakeLists.txt
index d3f849d0a..dda046372 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -170,9 +170,8 @@ svfloat32_t func(svfloat32_t a) {
endforeach()
add_definitions(-DARCH_AARCH64=1)
- set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fno-lax-vector-conversions")
# Prevents missmatch between arm and x86 bitstreams
- set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -ffp-contract=off")
+ set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /fp:precise")
endif()
include(GNUInstallDirs)
diff --git a/Source/Lib/ASM_NEON/cdef_filter_block_neon.c b/Source/Lib/ASM_NEON/cdef_filter_block_neon.c
index 92eb8a50c..eaf169a51 100644
--- a/Source/Lib/ASM_NEON/cdef_filter_block_neon.c
+++ b/Source/Lib/ASM_NEON/cdef_filter_block_neon.c
@@ -465,33 +465,33 @@ void svt_av1_cdef_filter_block_4xn_8_native_neon(uint8_t* dst, int32_t dstride,
int16x8_t suma = vdupq_n_s16(0), sumb = vdupq_n_s16(0);
int8x16_t c0, c1, c2, c3, csum;
-#define NAT4_TAP(OFF) \
- ({ \
+#define NAT4_TAP(VAR,OFF) \
+ { \
uint32x4_t _t = vdupq_n_u32(0); \
_t = vld1q_lane_u32((const uint32_t*)(r0 + (OFF)), _t, 0); \
_t = vld1q_lane_u32((const uint32_t*)(r1 + (OFF)), _t, 1); \
_t = vld1q_lane_u32((const uint32_t*)(r2 + (OFF)), _t, 2); \
_t = vld1q_lane_u32((const uint32_t*)(r3 + (OFF)), _t, 3); \
- vreinterpretq_u8_u32(_t); \
- })
+ VAR = vreinterpretq_u8_u32(_t); \
+ }
if (pri_strength) {
- tap8 = NAT4_TAP(po1);
+ NAT4_TAP(tap8, po1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c0 = constrain8x16(tap8, row8, prithr8, pridamp8);
- tap8 = NAT4_TAP(-po1);
+ NAT4_TAP(tap8, -po1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c1 = constrain8x16(tap8, row8, prithr8, pridamp8);
csum = vaddq_s8(c0, c1);
suma = vmlal_s8(suma, pri_t0, vget_low_s8(csum));
sumb = vmlal_s8(sumb, pri_t0, vget_high_s8(csum));
- tap8 = NAT4_TAP(po2);
+ NAT4_TAP(tap8, po2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c0 = constrain8x16(tap8, row8, prithr8, pridamp8);
- tap8 = NAT4_TAP(-po2);
+ NAT4_TAP(tap8, -po2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c1 = constrain8x16(tap8, row8, prithr8, pridamp8);
@@ -500,38 +500,38 @@ void svt_av1_cdef_filter_block_4xn_8_native_neon(uint8_t* dst, int32_t dstride,
sumb = vmlal_s8(sumb, pri_t1, vget_high_s8(csum));
}
if (sec_strength) {
- tap8 = NAT4_TAP(s1o1);
+ NAT4_TAP(tap8, s1o1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c0 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT4_TAP(-s1o1);
+ NAT4_TAP(tap8, -s1o1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c1 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT4_TAP(s2o1);
+ NAT4_TAP(tap8, s2o1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c2 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT4_TAP(-s2o1);
+ NAT4_TAP(tap8, -s2o1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c3 = constrain8x16(tap8, row8, secthr8, secdamp8);
csum = vaddq_s8(vaddq_s8(c0, c1), vaddq_s8(c2, c3));
suma = vmlal_s8(suma, sec_t0, vget_low_s8(csum));
sumb = vmlal_s8(sumb, sec_t0, vget_high_s8(csum));
- tap8 = NAT4_TAP(s1o2);
+ NAT4_TAP(tap8, s1o2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c0 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT4_TAP(-s1o2);
+ NAT4_TAP(tap8, -s1o2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c1 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT4_TAP(s2o2);
+ NAT4_TAP(tap8, s2o2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c2 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT4_TAP(-s2o2);
+ NAT4_TAP(tap8, -s2o2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c3 = constrain8x16(tap8, row8, secthr8, secdamp8);
@@ -561,30 +561,30 @@ void svt_av1_cdef_filter_block_4xn_8_native_neon(uint8_t* dst, int32_t dstride,
uint8x16_t min8 = row8, max8 = row8, tap8;
int16x8_t suma = vdupq_n_s16(0);
int8x16_t c0, c1, c2, c3, csum;
-#define NAT2_TAP(OFF) \
- ({ \
+#define NAT2_TAP(VAR, OFF) \
+ { \
uint32x2_t _t = vdup_n_u32(0); \
_t = vld1_lane_u32((const uint32_t*)(ra + (OFF)), _t, 0); \
_t = vld1_lane_u32((const uint32_t*)(rb + (OFF)), _t, 1); \
const uint8x8_t _l = vreinterpret_u8_u32(_t); \
- vcombine_u8(_l, _l); \
- })
+ VAR = vcombine_u8(_l, _l); \
+ }
if (pri_strength) {
- tap8 = NAT2_TAP(po1);
+ NAT2_TAP(tap8, po1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c0 = constrain8x16(tap8, row8, prithr8, pridamp8);
- tap8 = NAT2_TAP(-po1);
+ NAT2_TAP(tap8, -po1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c1 = constrain8x16(tap8, row8, prithr8, pridamp8);
csum = vaddq_s8(c0, c1);
suma = vmlal_s8(suma, pri_t0, vget_low_s8(csum));
- tap8 = NAT2_TAP(po2);
+ NAT2_TAP(tap8, po2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c0 = constrain8x16(tap8, row8, prithr8, pridamp8);
- tap8 = NAT2_TAP(-po2);
+ NAT2_TAP(tap8, -po2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c1 = constrain8x16(tap8, row8, prithr8, pridamp8);
@@ -592,37 +592,37 @@ void svt_av1_cdef_filter_block_4xn_8_native_neon(uint8_t* dst, int32_t dstride,
suma = vmlal_s8(suma, pri_t1, vget_low_s8(csum));
}
if (sec_strength) {
- tap8 = NAT2_TAP(s1o1);
+ NAT2_TAP(tap8, s1o1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c0 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT2_TAP(-s1o1);
+ NAT2_TAP(tap8, -s1o1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c1 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT2_TAP(s2o1);
+ NAT2_TAP(tap8, s2o1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c2 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT2_TAP(-s2o1);
+ NAT2_TAP(tap8, -s2o1);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c3 = constrain8x16(tap8, row8, secthr8, secdamp8);
csum = vaddq_s8(vaddq_s8(c0, c1), vaddq_s8(c2, c3));
suma = vmlal_s8(suma, sec_t0, vget_low_s8(csum));
- tap8 = NAT2_TAP(s1o2);
+ NAT2_TAP(tap8, s1o2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c0 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT2_TAP(-s1o2);
+ NAT2_TAP(tap8, -s1o2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c1 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT2_TAP(s2o2);
+ NAT2_TAP(tap8, s2o2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c2 = constrain8x16(tap8, row8, secthr8, secdamp8);
- tap8 = NAT2_TAP(-s2o2);
+ NAT2_TAP(tap8, -s2o2);
min8 = vminq_u8(min8, tap8);
max8 = vmaxq_u8(max8, tap8);
c3 = constrain8x16(tap8, row8, secthr8, secdamp8);
@@ -1285,20 +1285,20 @@ static void cdef_filter_block_4xn_8_bounded_neon(uint8_t* dst, int32_t dstride,
uint8x16_t min8 = row8, max8 = row8, tap8, av;
int16x8_t suma = vdupq_n_s16(0), sumb = vdupq_n_s16(0);
int8x16_t c0, c1, c2, c3, csum;
-#define NAT4_TAP(OFF) \
- ({ \
+#define NAT4_TAP(VAR, OFF) \
+ { \
uint32x4_t _t = vdupq_n_u32(0); \
_t = vld1q_lane_u32((const uint32_t*)(r0 + (OFF)), _t, 0); \
_t = vld1q_lane_u32((const uint32_t*)(r1 + (OFF)), _t, 1); \
_t = vld1q_lane_u32((const uint32_t*)(r2 + (OFF)), _t, 2); \
_t = vld1q_lane_u32((const uint32_t*)(r3 + (OFF)), _t, 3); \
- vreinterpretq_u8_u32(_t); \
- })
+ VAR = vreinterpretq_u8_u32(_t); \
+ }
#define B4(B, SGN, THR, DAMP, CACC) \
do { \
const int _need = (SGN) > 0 ? needp[B] : needn[B]; \
const int _rs = (SGN) > 0 ? rsensp[B] : rsensn[B]; \
- tap8 = NAT4_TAP((SGN) * off_b[B]); \
+ NAT4_TAP(tap8, (SGN) * off_b[B]); \
if (!_need) { \
max8 = vmaxq_u8(max8, tap8); \
min8 = vminq_u8(min8, tap8); \
@@ -1361,19 +1361,19 @@ static void cdef_filter_block_4xn_8_bounded_neon(uint8_t* dst, int32_t dstride,
uint8x16_t min8 = row8, max8 = row8, tap8, av;
int16x8_t suma = vdupq_n_s16(0);
int8x16_t c0, c1, c2, c3, csum;
-#define NAT2_TAP(OFF) \
- ({ \
+#define NAT2_TAP(VAR, OFF) \
+ { \
uint32x2_t _t = vdup_n_u32(0); \
_t = vld1_lane_u32((const uint32_t*)(ra + (OFF)), _t, 0); \
_t = vld1_lane_u32((const uint32_t*)(rb + (OFF)), _t, 1); \
const uint8x8_t _l = vreinterpret_u8_u32(_t); \
- vcombine_u8(_l, _l); \
- })
+ VAR = vcombine_u8(_l, _l); \
+ }
#define B2(B, SGN, THR, DAMP, CACC) \
do { \
const int _need = (SGN) > 0 ? needp[B] : needn[B]; \
const int _rs = (SGN) > 0 ? rsensp[B] : rsensn[B]; \
- tap8 = NAT2_TAP((SGN) * off_b[B]); \
+ NAT2_TAP(tap8, (SGN) * off_b[B]); \
if (!_need) { \
max8 = vmaxq_u8(max8, tap8); \
min8 = vminq_u8(min8, tap8); \
diff --git a/Source/Lib/ASM_NEON/cdef_neon.c b/Source/Lib/ASM_NEON/cdef_neon.c
index 1d0a8a506..f0b96a2c7 100644
--- a/Source/Lib/ASM_NEON/cdef_neon.c
+++ b/Source/Lib/ASM_NEON/cdef_neon.c
@@ -15,6 +15,7 @@
#include "cdef.h"
#include "definitions.h"
#include "mem_neon.h"
+#include <malloc.h>
/* 8-bit MSE (no FEAT_DotProd): abs-diff fits u8 and d*d fits u16 (255^2=65025), so use
* vabdq_u8 (16-lane abs in one op) + vmull_u8/vmull2 (square 8 lanes each into u16) +
@@ -234,7 +235,7 @@ uint64_t svt_search_one_dual_neon(int* lev0, int* lev1, int nb_strengths, uint64
}
size_t stride = ((end_gi + 3) & ~3);
- uint64_t tot_mse[end_gi * stride];
+ uint64_t* tot_mse = (uint64_t*)_malloca(sizeof(uint64_t) * (end_gi * stride));
size_t start_idx = start_gi;
size_t end_idx = end_gi;
@@ -362,5 +363,6 @@ uint64_t svt_search_one_dual_neon(int* lev0, int* lev1, int nb_strengths, uint64
lev0[nb_strengths] = best_id0; // Add the identified luma strength to the list of selected luma strengths
lev1[nb_strengths] = best_id1; // Add the identified chroma strength to the list of selected chroma strengths
+ _freea(tot_mse);
return best_tot_mse;
}
diff --git a/Source/Lib/ASM_NEON/compound_convolve_neon.c b/Source/Lib/ASM_NEON/compound_convolve_neon.c
index 3814b697b..7bfed2305 100644
--- a/Source/Lib/ASM_NEON/compound_convolve_neon.c
+++ b/Source/Lib/ASM_NEON/compound_convolve_neon.c
@@ -250,7 +250,7 @@ void svt_av1_jnt_convolve_2d_neon(const uint8_t* src, int32_t src_stride, uint8_
if (clamped_y_taps == 6) {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_2d_vert_6tap_dist_wtd_avg_neon(
im_block, im_stride, dst8, dst8_stride, conv_params, y_filter, h, w);
} else {
@@ -262,7 +262,7 @@ void svt_av1_jnt_convolve_2d_neon(const uint8_t* src, int32_t src_stride, uint8_
}
} else {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_2d_vert_8tap_dist_wtd_avg_neon(
im_block, im_stride, dst8, dst8_stride, conv_params, y_filter, h, w);
} else {
@@ -557,7 +557,7 @@ void svt_av1_jnt_convolve_2d_copy_neon(const uint8_t* src, int src_stride, uint8
}
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_2d_copy_dist_wtd_avg_neon(src, src_stride, dst8, dst8_stride, w, h, conv_params);
} else {
dist_wtd_convolve_2d_copy_avg_neon(src, src_stride, dst8, dst8_stride, w, h, conv_params);
@@ -1202,7 +1202,7 @@ void svt_av1_jnt_convolve_x_neon(const uint8_t* src, int32_t src_stride, uint8_t
}
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_x_dist_wtd_avg_neon(
src, src_stride, dst8, dst8_stride, w, h, filter_params_x, subpel_x_qn, conv_params);
} else {
@@ -2360,7 +2360,7 @@ void svt_av1_jnt_convolve_y_neon(const uint8_t* src, int32_t src_stride, uint8_t
if (get_filter_tap(filter_params_y, subpel_y_qn) <= 6) {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_y_6tap_dist_wtd_avg_neon(
src_ptr + src_stride, src_stride, dst8, dst8_stride, w, h, y_filter, conv_params);
} else {
@@ -2372,7 +2372,7 @@ void svt_av1_jnt_convolve_y_neon(const uint8_t* src, int32_t src_stride, uint8_t
}
} else {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_y_8tap_dist_wtd_avg_neon(
src_ptr, src_stride, dst8, dst8_stride, w, h, y_filter, conv_params);
} else {
diff --git a/Source/Lib/ASM_NEON/compound_convolve_neon.h b/Source/Lib/ASM_NEON/compound_convolve_neon.h
index a2d26dc9d..b89a5c480 100644
--- a/Source/Lib/ASM_NEON/compound_convolve_neon.h
+++ b/Source/Lib/ASM_NEON/compound_convolve_neon.h
@@ -12,13 +12,12 @@
#ifndef AOM_AV1_COMMON_ARM_COMPOUND_CONVOLVE_NEON_H_
#define AOM_AV1_COMMON_ARM_COMPOUND_CONVOLVE_NEON_H_
+#include "definitions.h"
#include "mem_neon.h"
#include "convolve.h"
#include "filter.h"
#define __builtin_prefetch(x)
-#define LIKELY(v) __builtin_expect(v, 1)
-#define UNLIKELY(v) __builtin_expect(v, 0)
static inline void compute_dist_wtd_avg_4x4(uint16x4_t dd0, uint16x4_t dd1, uint16x4_t dd2, uint16x4_t dd3,
uint16x4_t d0, uint16x4_t d1, uint16x4_t d2, uint16x4_t d3,
diff --git a/Source/Lib/ASM_NEON/convolve_neon.c b/Source/Lib/ASM_NEON/convolve_neon.c
index b8b47e808..8efe03ee8 100644
--- a/Source/Lib/ASM_NEON/convolve_neon.c
+++ b/Source/Lib/ASM_NEON/convolve_neon.c
@@ -20,6 +20,10 @@
#include "transpose_neon.h"
#include "utility.h"
+#if _MSC_VER
+#define __builtin_prefetch(x) __prefetch(x)
+#endif
+
// Sign mask for is_fast_path: +1 = tap must be >= 0, -1 = tap must be <= 0,
// 0 = tap must be zero/unused. This is the EIGHTTAP_REGULAR 6-tap pattern.
static const int8_t kRegularSignMask6Tap[8] = {0, 1, -1, 1, 1, -1, 1, 0};
diff --git a/Source/Lib/ASM_NEON/convolve_scale_neon.c b/Source/Lib/ASM_NEON/convolve_scale_neon.c
index 5c56294fe..53672a4c0 100644
--- a/Source/Lib/ASM_NEON/convolve_scale_neon.c
+++ b/Source/Lib/ASM_NEON/convolve_scale_neon.c
@@ -16,6 +16,7 @@
#include "common_dsp_rtcd.h"
#include "compound_convolve_neon.h"
#include "convolve_scale_neon.h"
+#include "definitions.h"
#include "filter.h"
#include "mem_neon.h"
#include "transpose_neon.h"
@@ -668,7 +669,7 @@ void svt_av1_convolve_2d_scale_neon(const uint8_t* src, int src_stride, uint8_t*
// Vertical filter
if (filter_params_y->interp_filter == MULTITAP_SHARP) {
- if (LIKELY(!conv_params->is_compound)) {
+ if (EB_LIKELY(!conv_params->is_compound)) {
convolve_vert_scale_8tap_neon(
im_block, im_stride, dst, dst_stride, w, h, filter_params_y->filter_ptr, subpel_y_qn, y_step_qn);
} else if (!conv_params->do_average) {
@@ -701,7 +702,7 @@ void svt_av1_convolve_2d_scale_neon(const uint8_t* src, int src_stride, uint8_t*
y_step_qn);
}
} else {
- if (LIKELY(!conv_params->is_compound)) {
+ if (EB_LIKELY(!conv_params->is_compound)) {
convolve_vert_scale_6tap_neon(im_block + im_stride,
im_stride,
dst,
diff --git a/Source/Lib/ASM_NEON/sad_neon.c b/Source/Lib/ASM_NEON/sad_neon.c
index 5abf5e9ab..150ac5946 100644
--- a/Source/Lib/ASM_NEON/sad_neon.c
+++ b/Source/Lib/ASM_NEON/sad_neon.c
@@ -15,6 +15,19 @@
#include "mem_neon.h"
#include "sum_neon.h"
+#if _MSC_VER
+#include <intrin.h>
+// rw -> read 0, write 1, shared read 2
+// PLD 0, PLI 1, PST 2
+// L1 0, l2 1, L3 2
+// locality -> no temporal 0, max temporal 3
+// KEEP 0, STRM 1
+#define __builtin_prefetch(addr,rw,locality) __prefetch2(addr, \
+ (((rw & 1) * 2) << 3) | \
+ ((rw & 2) << 1) | \
+ (locality > 0 ? 0 : 1))
+#endif
+
int svt_aom_satd_neon(const TranLow* coeff, int length) {
const int32x4_t zero = vdupq_n_s32(0);
int32x4_t accum = zero;
diff --git a/Source/Lib/ASM_NEON/sad_x4d_neon.c b/Source/Lib/ASM_NEON/sad_x4d_neon.c
index 36bc4ada3..38a10bd9c 100644
--- a/Source/Lib/ASM_NEON/sad_x4d_neon.c
+++ b/Source/Lib/ASM_NEON/sad_x4d_neon.c
@@ -14,6 +14,7 @@
#include <arm_neon.h>
#include "aom_dsp_rtcd.h"
+#include "definitions.h"
#include "mem_neon.h"
// svt_aom_sad<W>x<H>x4d() computes the SAD of one source block against four
@@ -99,7 +100,7 @@ static inline void sadwxhx4d_neon(const uint8_t* src, int src_stride, const uint
// cost is a single tail-branch, which microbenchmarks at parity with the fully
// specialized form.
#define SAD_WXHX4D_NEON(w) \
- static void __attribute__((noinline)) sad##w##xhx4d_neon(const uint8_t* src, \
+ static void NOINLINE sad##w##xhx4d_neon(const uint8_t* src, \
int src_stride, \
const uint8_t* const ref_array[], \
int ref_stride, \
diff --git a/Source/Lib/ASM_NEON_DOTPROD/convolve_scale_neon_dotprod.c b/Source/Lib/ASM_NEON_DOTPROD/convolve_scale_neon_dotprod.c
index 0f83a116d..f64828a39 100644
--- a/Source/Lib/ASM_NEON_DOTPROD/convolve_scale_neon_dotprod.c
+++ b/Source/Lib/ASM_NEON_DOTPROD/convolve_scale_neon_dotprod.c
@@ -16,6 +16,7 @@
#include "common_dsp_rtcd.h"
#include "compound_convolve_neon.h"
#include "convolve_scale_neon.h"
+#include "definitions.h"
#include "filter.h"
#include "mem_neon.h"
#include "transpose_neon.h"
@@ -349,7 +350,7 @@ void svt_av1_convolve_2d_scale_neon_dotprod(const uint8_t* src, int src_stride,
// Vertical filter
if (filter_params_y->interp_filter == MULTITAP_SHARP) {
- if (LIKELY(!conv_params->is_compound)) {
+ if (EB_LIKELY(!conv_params->is_compound)) {
convolve_vert_scale_8tap_neon(
im_block, im_stride, dst, dst_stride, w, h, filter_params_y->filter_ptr, subpel_y_qn, y_step_qn);
} else if (!conv_params->do_average) {
@@ -382,7 +383,7 @@ void svt_av1_convolve_2d_scale_neon_dotprod(const uint8_t* src, int src_stride,
y_step_qn);
}
} else {
- if (LIKELY(!conv_params->is_compound)) {
+ if (EB_LIKELY(!conv_params->is_compound)) {
convolve_vert_scale_6tap_neon(im_block + im_stride,
im_stride,
dst,
diff --git a/Source/Lib/ASM_NEON_DOTPROD/jnt_convolve_neon_dotprod.c b/Source/Lib/ASM_NEON_DOTPROD/jnt_convolve_neon_dotprod.c
index 4154280e5..ca84f1f5d 100644
--- a/Source/Lib/ASM_NEON_DOTPROD/jnt_convolve_neon_dotprod.c
+++ b/Source/Lib/ASM_NEON_DOTPROD/jnt_convolve_neon_dotprod.c
@@ -601,7 +601,7 @@ void svt_av1_jnt_convolve_2d_neon_dotprod(const uint8_t* src, int src_stride, ui
if (clamped_y_taps == 6) {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_2d_vert_6tap_dist_wtd_avg_neon(
im_block, im_stride, dst8, dst8_stride, conv_params, y_filter, h, w);
} else {
@@ -613,7 +613,7 @@ void svt_av1_jnt_convolve_2d_neon_dotprod(const uint8_t* src, int src_stride, ui
}
} else {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_2d_vert_8tap_dist_wtd_avg_neon(
im_block, im_stride, dst8, dst8_stride, conv_params, y_filter, h, w);
} else {
@@ -1737,7 +1737,7 @@ void svt_av1_jnt_convolve_y_neon_dotprod(const uint8_t* src, int32_t src_stride,
if (get_filter_tap(filter_params_y, subpel_y_qn) <= 4) {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_y_4tap_dist_wtd_avg_neon_dotprod(
src - src_stride, src_stride, dst8, dst8_stride, w, h, y_filter_ptr, conv_params);
} else {
@@ -1749,7 +1749,7 @@ void svt_av1_jnt_convolve_y_neon_dotprod(const uint8_t* src, int32_t src_stride,
}
} else {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_y_8tap_dist_wtd_avg_neon_dotprod(
src - 3 * src_stride, src_stride, dst8, dst8_stride, w, h, y_filter_ptr, conv_params);
} else {
diff --git a/Source/Lib/ASM_NEON_DOTPROD/ssim_neon_dotprod.c b/Source/Lib/ASM_NEON_DOTPROD/ssim_neon_dotprod.c
index 6b6ad948c..89af35cea 100644
--- a/Source/Lib/ASM_NEON_DOTPROD/ssim_neon_dotprod.c
+++ b/Source/Lib/ASM_NEON_DOTPROD/ssim_neon_dotprod.c
@@ -46,8 +46,10 @@ double svt_ssim_4x4_neon_dotprod(const uint8_t* s, uint32_t sp, const uint8_t* r
memcpy(&r1, r + rp, 4);
memcpy(&r2, r + 2 * rp, 4);
memcpy(&r3, r + 3 * rp, 4);
- const uint8x16_t vs = vreinterpretq_u8_u32((uint32x4_t){s0, s1, s2, s3});
- const uint8x16_t vr = vreinterpretq_u8_u32((uint32x4_t){r0, r1, r2, r3});
+ const uint32x4_t ts = (uint32x4_t){.n128_u32 = {s0, s1, s2, s3}};
+ const uint8x16_t vs = vreinterpretq_u8_u32(ts);
+ const uint32x4_t tr = (uint32x4_t){.n128_u32 = {r0, r1, r2, r3}};
+ const uint8x16_t vr = vreinterpretq_u8_u32(tr);
const uint8x16_t ones = vdupq_n_u8(1);
return svt_aom_similarity(vaddvq_u32(vdotq_u32(vdupq_n_u32(0), vs, ones)),
vaddvq_u32(vdotq_u32(vdupq_n_u32(0), vr, ones)),
diff --git a/Source/Lib/ASM_NEON_I8MM/convolve_scale_neon_i8mm.c b/Source/Lib/ASM_NEON_I8MM/convolve_scale_neon_i8mm.c
index bf298b5fd..09b244d0d 100644
--- a/Source/Lib/ASM_NEON_I8MM/convolve_scale_neon_i8mm.c
+++ b/Source/Lib/ASM_NEON_I8MM/convolve_scale_neon_i8mm.c
@@ -17,6 +17,7 @@
#include "compound_convolve_neon.h"
#include "convolve_scale_neon.h"
#include "convolve_scale_neon_dotprod.h"
+#include "definitions.h"
#include "filter.h"
#include "mem_neon.h"
#include "transpose_neon.h"
@@ -316,7 +317,7 @@ void svt_av1_convolve_2d_scale_neon_i8mm(const uint8_t* src, int src_stride, uin
// Vertical filter
if (filter_params_y->interp_filter == MULTITAP_SHARP) {
- if (LIKELY(!conv_params->is_compound)) {
+ if (EB_LIKELY(!conv_params->is_compound)) {
convolve_vert_scale_8tap_neon(
im_block, im_stride, dst, dst_stride, w, h, filter_params_y->filter_ptr, subpel_y_qn, y_step_qn);
} else if (!conv_params->do_average) {
@@ -349,7 +350,7 @@ void svt_av1_convolve_2d_scale_neon_i8mm(const uint8_t* src, int src_stride, uin
y_step_qn);
}
} else {
- if (LIKELY(!conv_params->is_compound)) {
+ if (EB_LIKELY(!conv_params->is_compound)) {
convolve_vert_scale_6tap_neon(im_block + im_stride,
im_stride,
dst,
diff --git a/Source/Lib/ASM_NEON_I8MM/jnt_convolve_neon_i8mm.c b/Source/Lib/ASM_NEON_I8MM/jnt_convolve_neon_i8mm.c
index d1a0f1781..d78b95063 100644
--- a/Source/Lib/ASM_NEON_I8MM/jnt_convolve_neon_i8mm.c
+++ b/Source/Lib/ASM_NEON_I8MM/jnt_convolve_neon_i8mm.c
@@ -575,7 +575,7 @@ void svt_av1_jnt_convolve_x_neon_i8mm(const uint8_t* src, int src_stride, uint8_
src -= (SUBPEL_TAPS / 2 - 1);
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
if (filter_taps < 8) {
dist_wtd_convolve_x_dist_wtd_avg_6tap_neon_i8mm(src + 1,
src_stride,
@@ -898,7 +898,7 @@ void svt_av1_jnt_convolve_2d_neon_i8mm(const uint8_t* src, int src_stride, uint8
if (clamped_y_taps == 6) {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_2d_vert_6tap_dist_wtd_avg_neon(
im_block, im_stride, dst8, dst8_stride, conv_params, y_filter, h, w);
} else {
@@ -910,7 +910,7 @@ void svt_av1_jnt_convolve_2d_neon_i8mm(const uint8_t* src, int src_stride, uint8
}
} else {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_2d_vert_8tap_dist_wtd_avg_neon(
im_block, im_stride, dst8, dst8_stride, conv_params, y_filter, h, w);
} else {
@@ -1874,7 +1874,7 @@ void svt_av1_jnt_convolve_y_neon_i8mm(const uint8_t* src, int32_t src_stride, ui
if (get_filter_tap(filter_params_y, subpel_y_qn) <= 4) {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_y_4tap_dist_wtd_avg_neon_i8mm(
src - src_stride, src_stride, dst8, dst8_stride, w, h, y_filter_ptr, conv_params);
} else {
@@ -1886,7 +1886,7 @@ void svt_av1_jnt_convolve_y_neon_i8mm(const uint8_t* src, int32_t src_stride, ui
}
} else {
if (conv_params->do_average) {
- if (UNLIKELY(conv_params->use_jnt_comp_avg)) {
+ if (EB_UNLIKELY(conv_params->use_jnt_comp_avg)) {
dist_wtd_convolve_y_8tap_dist_wtd_avg_neon_i8mm(
src - 3 * src_stride, src_stride, dst8, dst8_stride, w, h, y_filter_ptr, conv_params);
} else {