Merge "Fix int64_t to unsigned int conversion warnings"
diff --git a/test/invalid_file_test.cc b/test/invalid_file_test.cc
index 8b18df6..cf86cf5 100644
--- a/test/invalid_file_test.cc
+++ b/test/invalid_file_test.cc
@@ -147,6 +147,7 @@
{4, "invalid-"
"vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf"},
{2, "invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf"},
+ {4, "invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf"},
};
INSTANTIATE_TEST_CASE_P(
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index 7956e50..d2bc5a8 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -685,3 +685,5 @@
0a3884edb3fd8f9d9b500223e650f7de257b67d8 invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf.res
fac89b5735be8a86b0dc05159f996a5c3208ae32 invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf
22e0ee8babe574722baf4ef6d7ff5d7cf80d386c invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf.res
+4506dfdcdf8ee4250924b075a0dcf1f070f72e5a invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf
+d3ea592c8d7b05d14c7ed48befc0a3aaf7709b7a invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf.res
diff --git a/test/test.mk b/test/test.mk
index fe5a1cf..e592a70 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -807,6 +807,8 @@
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm.res
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf.res
ifeq ($(CONFIG_DECODE_PERF_TESTS),yes)
# BBB VP9 streams
diff --git a/test/variance_test.cc b/test/variance_test.cc
index 7d81182..925ff03 100644
--- a/test/variance_test.cc
+++ b/test/variance_test.cc
@@ -485,21 +485,6 @@
make_tuple(6, 5, subpel_avg_variance64x32_c),
make_tuple(6, 6, subpel_avg_variance64x64_c)));
-#if HAVE_MMX
-const vp9_variance_fn_t variance4x4_mmx = vp9_variance4x4_mmx;
-const vp9_variance_fn_t variance8x8_mmx = vp9_variance8x8_mmx;
-const vp9_variance_fn_t variance8x16_mmx = vp9_variance8x16_mmx;
-const vp9_variance_fn_t variance16x8_mmx = vp9_variance16x8_mmx;
-const vp9_variance_fn_t variance16x16_mmx = vp9_variance16x16_mmx;
-INSTANTIATE_TEST_CASE_P(
- MMX, VP9VarianceTest,
- ::testing::Values(make_tuple(2, 2, variance4x4_mmx),
- make_tuple(3, 3, variance8x8_mmx),
- make_tuple(3, 4, variance8x16_mmx),
- make_tuple(4, 3, variance16x8_mmx),
- make_tuple(4, 4, variance16x16_mmx)));
-#endif
-
#if HAVE_SSE2
#if CONFIG_USE_X86INC
const vp9_variance_fn_t variance4x4_sse2 = vp9_variance4x4_sse2;
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 92f9318..5dadee906 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -420,19 +420,19 @@
specialize qw/vp9_variance64x64 avx2/, "$sse2_x86inc";
add_proto qw/unsigned int vp9_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
-specialize qw/vp9_variance16x16 mmx avx2 neon/, "$sse2_x86inc";
+specialize qw/vp9_variance16x16 avx2 neon/, "$sse2_x86inc";
add_proto qw/unsigned int vp9_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
-specialize qw/vp9_variance16x8 mmx/, "$sse2_x86inc";
+specialize qw/vp9_variance16x8/, "$sse2_x86inc";
add_proto qw/unsigned int vp9_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
-specialize qw/vp9_variance8x16 mmx/, "$sse2_x86inc";
+specialize qw/vp9_variance8x16/, "$sse2_x86inc";
add_proto qw/unsigned int vp9_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
-specialize qw/vp9_variance8x8 mmx neon/, "$sse2_x86inc";
+specialize qw/vp9_variance8x8 neon/, "$sse2_x86inc";
add_proto qw/void vp9_get8x8var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
-specialize qw/vp9_get8x8var mmx neon/, "$sse2_x86inc";
+specialize qw/vp9_get8x8var neon/, "$sse2_x86inc";
add_proto qw/void vp9_get16x16var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
specialize qw/vp9_get16x16var avx2 neon/, "$sse2_x86inc";
@@ -444,7 +444,7 @@
specialize qw/vp9_variance4x8/, "$sse2_x86inc";
add_proto qw/unsigned int vp9_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
-specialize qw/vp9_variance4x4 mmx/, "$sse2_x86inc";
+specialize qw/vp9_variance4x4/, "$sse2_x86inc";
add_proto qw/unsigned int vp9_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
specialize qw/vp9_sub_pixel_variance64x64 avx2/, "$sse2_x86inc", "$ssse3_x86inc";
@@ -693,7 +693,7 @@
specialize qw/vp9_sad4x4x4d sse/;
add_proto qw/unsigned int vp9_mse16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int recon_stride, unsigned int *sse";
-specialize qw/vp9_mse16x16 mmx avx2/, "$sse2_x86inc";
+specialize qw/vp9_mse16x16 avx2/, "$sse2_x86inc";
add_proto qw/unsigned int vp9_mse8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int recon_stride, unsigned int *sse";
specialize qw/vp9_mse8x16/;
@@ -705,7 +705,7 @@
specialize qw/vp9_mse8x8/;
add_proto qw/unsigned int vp9_get_mb_ss/, "const int16_t *";
-specialize qw/vp9_get_mb_ss mmx sse2/;
+specialize qw/vp9_get_mb_ss sse2/;
# ENCODEMB INVOKE
add_proto qw/int64_t vp9_block_error/, "const int16_t *coeff, const int16_t *dqcoeff, intptr_t block_size, int64_t *ssz";
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index 72e5edf..aae56c1 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -996,6 +996,7 @@
// Reset tile decoding hook
for (n = 0; n < num_workers; ++n) {
+ winterface->sync(&pbi->tile_workers[n]);
pbi->tile_workers[n].hook = (VP9WorkerHook)tile_worker_hook;
}
diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index e79dcf3..ae0da79 100644
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -96,10 +96,8 @@
}
vpx_free(pbi->tile_workers);
- if (pbi->num_tile_workers) {
- const int sb_rows =
- mi_cols_aligned_to_sb(cm->mi_rows) >> MI_BLOCK_SIZE_LOG2;
- vp9_loop_filter_dealloc(&pbi->lf_row_sync, sb_rows);
+ if (pbi->num_tile_workers > 0) {
+ vp9_loop_filter_dealloc(&pbi->lf_row_sync);
}
vp9_remove_common(cm);
diff --git a/vp9/decoder/vp9_dthread.c b/vp9/decoder/vp9_dthread.c
index cf5cc8c..b82ea6a 100644
--- a/vp9/decoder/vp9_dthread.c
+++ b/vp9/decoder/vp9_dthread.c
@@ -148,16 +148,7 @@
// Allocate memory used in thread synchronization.
// This always needs to be done even if frame_filter_level is 0.
if (!lf_sync->sync_range || cm->last_height != cm->height) {
- if (cm->last_height != cm->height) {
- const int aligned_last_height =
- ALIGN_POWER_OF_TWO(cm->last_height, MI_SIZE_LOG2);
- const int last_sb_rows =
- mi_cols_aligned_to_sb(aligned_last_height >> MI_SIZE_LOG2) >>
- MI_BLOCK_SIZE_LOG2;
-
- vp9_loop_filter_dealloc(lf_sync, last_sb_rows);
- }
-
+ vp9_loop_filter_dealloc(lf_sync);
vp9_loop_filter_alloc(cm, lf_sync, sb_rows, cm->width);
}
@@ -227,19 +218,22 @@
// Allocate memory for lf row synchronization
void vp9_loop_filter_alloc(VP9_COMMON *cm, VP9LfSync *lf_sync, int rows,
int width) {
+ lf_sync->rows = rows;
#if CONFIG_MULTITHREAD
- int i;
+ {
+ int i;
- CHECK_MEM_ERROR(cm, lf_sync->mutex_,
- vpx_malloc(sizeof(*lf_sync->mutex_) * rows));
- for (i = 0; i < rows; ++i) {
- pthread_mutex_init(&lf_sync->mutex_[i], NULL);
- }
+ CHECK_MEM_ERROR(cm, lf_sync->mutex_,
+ vpx_malloc(sizeof(*lf_sync->mutex_) * rows));
+ for (i = 0; i < rows; ++i) {
+ pthread_mutex_init(&lf_sync->mutex_[i], NULL);
+ }
- CHECK_MEM_ERROR(cm, lf_sync->cond_,
- vpx_malloc(sizeof(*lf_sync->cond_) * rows));
- for (i = 0; i < rows; ++i) {
- pthread_cond_init(&lf_sync->cond_[i], NULL);
+ CHECK_MEM_ERROR(cm, lf_sync->cond_,
+ vpx_malloc(sizeof(*lf_sync->cond_) * rows));
+ for (i = 0; i < rows; ++i) {
+ pthread_cond_init(&lf_sync->cond_[i], NULL);
+ }
}
#endif // CONFIG_MULTITHREAD
@@ -251,23 +245,19 @@
}
// Deallocate lf synchronization related mutex and data
-void vp9_loop_filter_dealloc(VP9LfSync *lf_sync, int rows) {
-#if !CONFIG_MULTITHREAD
- (void)rows;
-#endif // !CONFIG_MULTITHREAD
-
+void vp9_loop_filter_dealloc(VP9LfSync *lf_sync) {
if (lf_sync != NULL) {
#if CONFIG_MULTITHREAD
int i;
if (lf_sync->mutex_ != NULL) {
- for (i = 0; i < rows; ++i) {
+ for (i = 0; i < lf_sync->rows; ++i) {
pthread_mutex_destroy(&lf_sync->mutex_[i]);
}
vpx_free(lf_sync->mutex_);
}
if (lf_sync->cond_ != NULL) {
- for (i = 0; i < rows; ++i) {
+ for (i = 0; i < lf_sync->rows; ++i) {
pthread_cond_destroy(&lf_sync->cond_[i]);
}
vpx_free(lf_sync->cond_);
diff --git a/vp9/decoder/vp9_dthread.h b/vp9/decoder/vp9_dthread.h
index 423bd88..8b02ef7 100644
--- a/vp9/decoder/vp9_dthread.h
+++ b/vp9/decoder/vp9_dthread.h
@@ -38,6 +38,7 @@
// The optimal sync_range for different resolution and platform should be
// determined by testing. Currently, it is chosen to be a power-of-2 number.
int sync_range;
+ int rows;
} VP9LfSync;
// Allocate memory for loopfilter row synchronization.
@@ -45,7 +46,7 @@
int rows, int width);
// Deallocate loopfilter synchronization related mutex and data.
-void vp9_loop_filter_dealloc(VP9LfSync *lf_sync, int rows);
+void vp9_loop_filter_dealloc(VP9LfSync *lf_sync);
// Multi-threaded loopfilter that uses the tile threads.
void vp9_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame,
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index 9f82a5a..b26c8c7 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -390,7 +390,7 @@
search_site_config ss_cfg;
int mbmode_cost[INTRA_MODES];
- unsigned inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
+ unsigned int inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
int intra_uv_mode_cost[FRAME_TYPES][INTRA_MODES];
int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
diff --git a/vp9/encoder/x86/vp9_variance_impl_mmx.asm b/vp9/encoder/x86/vp9_variance_impl_mmx.asm
deleted file mode 100644
index 3501cf1..0000000
--- a/vp9/encoder/x86/vp9_variance_impl_mmx.asm
+++ /dev/null
@@ -1,510 +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_ports/x86_abi_support.asm"
-
-;unsigned int vp9_get_mb_ss_mmx( short *src_ptr )
-global sym(vp9_get_mb_ss_mmx) PRIVATE
-sym(vp9_get_mb_ss_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- GET_GOT rbx
- push rsi
- push rdi
- sub rsp, 8
- ; end prolog
-
- mov rax, arg(0) ;src_ptr
- mov rcx, 16
- pxor mm4, mm4
-
-.NEXTROW:
- movq mm0, [rax]
- movq mm1, [rax+8]
- movq mm2, [rax+16]
- movq mm3, [rax+24]
- pmaddwd mm0, mm0
- pmaddwd mm1, mm1
- pmaddwd mm2, mm2
- pmaddwd mm3, mm3
-
- paddd mm4, mm0
- paddd mm4, mm1
- paddd mm4, mm2
- paddd mm4, mm3
-
- add rax, 32
- dec rcx
- ja .NEXTROW
- movq QWORD PTR [rsp], mm4
-
- ;return sum[0]+sum[1];
- movsxd rax, dword ptr [rsp]
- movsxd rcx, dword ptr [rsp+4]
- add rax, rcx
-
-
- ; begin epilog
- add rsp, 8
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-;unsigned int vp9_get8x8var_mmx
-;(
-; unsigned char *src_ptr,
-; int source_stride,
-; unsigned char *ref_ptr,
-; int recon_stride,
-; unsigned int *SSE,
-; int *Sum
-;)
-global sym(vp9_get8x8var_mmx) PRIVATE
-sym(vp9_get8x8var_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- push rbx
- sub rsp, 16
- ; end prolog
-
-
- pxor mm5, mm5 ; Blank mmx6
- pxor mm6, mm6 ; Blank mmx7
- pxor mm7, mm7 ; Blank mmx7
-
- mov rax, arg(0) ;[src_ptr] ; Load base addresses
- mov rbx, arg(2) ;[ref_ptr]
- movsxd rcx, dword ptr arg(1) ;[source_stride]
- movsxd rdx, dword ptr arg(3) ;[recon_stride]
-
- ; Row 1
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm1, [rbx] ; Copy eight bytes to mm1
- movq mm2, mm0 ; Take copies
- movq mm3, mm1 ; Take copies
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- punpckhbw mm2, mm6 ; unpack to higher prrcision
- punpckhbw mm3, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- psubsw mm2, mm3 ; A-B (high order) to MM2
-
- paddw mm5, mm0 ; accumulate differences in mm5
- paddw mm5, mm2 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- pmaddwd mm2, mm2 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
- paddd mm7, mm2 ; accumulate in mm7
-
-
- ; Row 2
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm2, mm0 ; Take copies
- movq mm3, mm1 ; Take copies
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- punpckhbw mm2, mm6 ; unpack to higher prrcision
- punpckhbw mm3, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- psubsw mm2, mm3 ; A-B (high order) to MM2
-
- paddw mm5, mm0 ; accumulate differences in mm5
- paddw mm5, mm2 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- pmaddwd mm2, mm2 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
- paddd mm7, mm2 ; accumulate in mm7
-
- ; Row 3
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm2, mm0 ; Take copies
- movq mm3, mm1 ; Take copies
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- punpckhbw mm2, mm6 ; unpack to higher prrcision
- punpckhbw mm3, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- psubsw mm2, mm3 ; A-B (high order) to MM2
-
- paddw mm5, mm0 ; accumulate differences in mm5
- paddw mm5, mm2 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- pmaddwd mm2, mm2 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
- paddd mm7, mm2 ; accumulate in mm7
-
- ; Row 4
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm2, mm0 ; Take copies
- movq mm3, mm1 ; Take copies
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- punpckhbw mm2, mm6 ; unpack to higher prrcision
- punpckhbw mm3, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- psubsw mm2, mm3 ; A-B (high order) to MM2
-
- paddw mm5, mm0 ; accumulate differences in mm5
- paddw mm5, mm2 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- pmaddwd mm2, mm2 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
- paddd mm7, mm2 ; accumulate in mm7
-
- ; Row 5
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm2, mm0 ; Take copies
- movq mm3, mm1 ; Take copies
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- punpckhbw mm2, mm6 ; unpack to higher prrcision
- punpckhbw mm3, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- psubsw mm2, mm3 ; A-B (high order) to MM2
-
- paddw mm5, mm0 ; accumulate differences in mm5
- paddw mm5, mm2 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- pmaddwd mm2, mm2 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
- ; movq mm4, [rbx + rdx]
- paddd mm7, mm0 ; accumulate in mm7
- paddd mm7, mm2 ; accumulate in mm7
-
- ; Row 6
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm2, mm0 ; Take copies
- movq mm3, mm1 ; Take copies
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- punpckhbw mm2, mm6 ; unpack to higher prrcision
- punpckhbw mm3, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- psubsw mm2, mm3 ; A-B (high order) to MM2
-
- paddw mm5, mm0 ; accumulate differences in mm5
- paddw mm5, mm2 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- pmaddwd mm2, mm2 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
- paddd mm7, mm2 ; accumulate in mm7
-
- ; Row 7
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm2, mm0 ; Take copies
- movq mm3, mm1 ; Take copies
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- punpckhbw mm2, mm6 ; unpack to higher prrcision
- punpckhbw mm3, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- psubsw mm2, mm3 ; A-B (high order) to MM2
-
- paddw mm5, mm0 ; accumulate differences in mm5
- paddw mm5, mm2 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- pmaddwd mm2, mm2 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
- paddd mm7, mm2 ; accumulate in mm7
-
- ; Row 8
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm2, mm0 ; Take copies
- movq mm3, mm1 ; Take copies
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- punpckhbw mm2, mm6 ; unpack to higher prrcision
- punpckhbw mm3, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- psubsw mm2, mm3 ; A-B (high order) to MM2
-
- paddw mm5, mm0 ; accumulate differences in mm5
- paddw mm5, mm2 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- pmaddwd mm2, mm2 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- paddd mm7, mm0 ; accumulate in mm7
- paddd mm7, mm2 ; accumulate in mm7
-
- ; Now accumulate the final results.
- movq QWORD PTR [rsp+8], mm5 ; copy back accumulated results into normal memory
- movq QWORD PTR [rsp], mm7 ; copy back accumulated results into normal memory
- movsx rdx, WORD PTR [rsp+8]
- movsx rcx, WORD PTR [rsp+10]
- movsx rbx, WORD PTR [rsp+12]
- movsx rax, WORD PTR [rsp+14]
- add rdx, rcx
- add rbx, rax
- add rdx, rbx ;XSum
- movsxd rax, DWORD PTR [rsp]
- movsxd rcx, DWORD PTR [rsp+4]
- add rax, rcx ;XXSum
- mov rsi, arg(4) ;SSE
- mov rdi, arg(5) ;Sum
- mov dword ptr [rsi], eax
- mov dword ptr [rdi], edx
- xor rax, rax ; return 0
-
-
- ; begin epilog
- add rsp, 16
- pop rbx
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-
-;unsigned int
-;vp9_get4x4var_mmx
-;(
-; unsigned char *src_ptr,
-; int source_stride,
-; unsigned char *ref_ptr,
-; int recon_stride,
-; unsigned int *SSE,
-; int *Sum
-;)
-global sym(vp9_get4x4var_mmx) PRIVATE
-sym(vp9_get4x4var_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- push rbx
- sub rsp, 16
- ; end prolog
-
-
- pxor mm5, mm5 ; Blank mmx6
- pxor mm6, mm6 ; Blank mmx7
- pxor mm7, mm7 ; Blank mmx7
-
- mov rax, arg(0) ;[src_ptr] ; Load base addresses
- mov rbx, arg(2) ;[ref_ptr]
- movsxd rcx, dword ptr arg(1) ;[source_stride]
- movsxd rdx, dword ptr arg(3) ;[recon_stride]
-
- ; Row 1
- movd mm0, [rax] ; Copy 4 bytes to mm0
- movd mm1, [rbx] ; Copy 4 bytes to mm1
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- paddw mm5, mm0 ; accumulate differences in mm5
- pmaddwd mm0, mm0 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movd mm1, [rbx] ; Copy 4 bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
-
-
- ; Row 2
- movd mm0, [rax] ; Copy 4 bytes to mm0
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- paddw mm5, mm0 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movd mm1, [rbx] ; Copy 4 bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
-
- ; Row 3
- movd mm0, [rax] ; Copy 4 bytes to mm0
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- paddw mm5, mm0 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movd mm1, [rbx] ; Copy 4 bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
-
- ; Row 4
- movd mm0, [rax] ; Copy 4 bytes to mm0
-
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
-
- paddw mm5, mm0 ; accumulate differences in mm5
-
- pmaddwd mm0, mm0 ; square and accumulate
- paddd mm7, mm0 ; accumulate in mm7
-
-
- ; Now accumulate the final results.
- movq QWORD PTR [rsp+8], mm5 ; copy back accumulated results into normal memory
- movq QWORD PTR [rsp], mm7 ; copy back accumulated results into normal memory
- movsx rdx, WORD PTR [rsp+8]
- movsx rcx, WORD PTR [rsp+10]
- movsx rbx, WORD PTR [rsp+12]
- movsx rax, WORD PTR [rsp+14]
- add rdx, rcx
- add rbx, rax
- add rdx, rbx ;XSum
- movsxd rax, DWORD PTR [rsp]
- movsxd rcx, DWORD PTR [rsp+4]
- add rax, rcx ;XXSum
- mov rsi, arg(4) ;SSE
- mov rdi, arg(5) ;Sum
- mov dword ptr [rsi], eax
- mov dword ptr [rdi], edx
- xor rax, rax ; return 0
-
-
- ; begin epilog
- add rsp, 16
- pop rbx
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-
-;unsigned int
-;vp9_get4x4sse_cs_mmx
-;(
-; unsigned char *src_ptr,
-; int source_stride,
-; unsigned char *ref_ptr,
-; int recon_stride
-;)
-global sym(vp9_get4x4sse_cs_mmx) PRIVATE
-sym(vp9_get4x4sse_cs_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
- push rsi
- push rdi
- push rbx
- ; end prolog
-
-
- pxor mm6, mm6 ; Blank mmx7
- pxor mm7, mm7 ; Blank mmx7
-
- mov rax, arg(0) ;[src_ptr] ; Load base addresses
- mov rbx, arg(2) ;[ref_ptr]
- movsxd rcx, dword ptr arg(1) ;[source_stride]
- movsxd rdx, dword ptr arg(3) ;[recon_stride]
- ; Row 1
- movd mm0, [rax] ; Copy eight bytes to mm0
- movd mm1, [rbx] ; Copy eight bytes to mm1
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- pmaddwd mm0, mm0 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movd mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
-
- ; Row 2
- movd mm0, [rax] ; Copy eight bytes to mm0
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- pmaddwd mm0, mm0 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movd mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
-
- ; Row 3
- movd mm0, [rax] ; Copy eight bytes to mm0
- punpcklbw mm1, mm6
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- psubsw mm0, mm1 ; A-B (low order) to MM0
-
- pmaddwd mm0, mm0 ; square and accumulate
- add rbx,rdx ; Inc pointer into ref data
- add rax,rcx ; Inc pointer into the new data
- movd mm1, [rbx] ; Copy eight bytes to mm1
- paddd mm7, mm0 ; accumulate in mm7
-
- ; Row 4
- movd mm0, [rax] ; Copy eight bytes to mm0
- punpcklbw mm0, mm6 ; unpack to higher prrcision
- punpcklbw mm1, mm6
- psubsw mm0, mm1 ; A-B (low order) to MM0
- pmaddwd mm0, mm0 ; square and accumulate
- paddd mm7, mm0 ; accumulate in mm7
-
- movq mm0, mm7 ;
- psrlq mm7, 32
-
- paddd mm0, mm7
- movq rax, mm0
-
-
- ; begin epilog
- pop rbx
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
diff --git a/vp9/encoder/x86/vp9_variance_mmx.c b/vp9/encoder/x86/vp9_variance_mmx.c
deleted file mode 100644
index ce1c832..0000000
--- a/vp9/encoder/x86/vp9_variance_mmx.c
+++ /dev/null
@@ -1,103 +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 "vp9/encoder/vp9_variance.h"
-#include "vpx_ports/mem.h"
-
-unsigned int vp9_get8x8var_mmx(const uint8_t *src, int src_stride,
- const uint8_t *ref, int ref_stride,
- unsigned int *sse, int *sum);
-
-unsigned int vp9_get4x4var_mmx(const uint8_t *src, int src_stride,
- const uint8_t *ref, int ref_stride,
- unsigned int *SSE, int *sum);
-
-unsigned int vp9_variance4x4_mmx(const uint8_t *src, int src_stride,
- const uint8_t *ref, int ref_stride,
- unsigned int *sse) {
- int sum;
- vp9_get4x4var_mmx(src, src_stride, ref, ref_stride, sse, &sum);
- return *sse - (((unsigned int)sum * sum) >> 4);
-}
-
-unsigned int vp9_variance8x8_mmx(const uint8_t *src, int src_stride,
- const uint8_t *ref, int ref_stride,
- unsigned int *sse) {
- int sum;
- vp9_get8x8var_mmx(src, src_stride, ref, ref_stride, sse, &sum);
- return *sse - (((unsigned int)sum * sum) >> 6);
-}
-
-unsigned int vp9_mse16x16_mmx(const uint8_t *src, int src_stride,
- const uint8_t *ref, int ref_stride,
- unsigned int *sse) {
- unsigned int sse0, sse1, sse2, sse3;
- int sum0, sum1, sum2, sum3;
-
- vp9_get8x8var_mmx(src, src_stride, ref, ref_stride, &sse0, &sum0);
- vp9_get8x8var_mmx(src + 8, src_stride, ref + 8, ref_stride, &sse1, &sum1);
- vp9_get8x8var_mmx(src + 8 * src_stride, src_stride,
- ref + 8 * ref_stride, ref_stride, &sse2, &sum2);
- vp9_get8x8var_mmx(src + 8 * src_stride + 8, src_stride,
- ref + 8 * ref_stride + 8, ref_stride, &sse3, &sum3);
-
- *sse = sse0 + sse1 + sse2 + sse3;
- return *sse;
-}
-
-
-unsigned int vp9_variance16x16_mmx(const uint8_t *src, int src_stride,
- const uint8_t *ref, int ref_stride,
- unsigned int *sse) {
- unsigned int sse0, sse1, sse2, sse3;
- int sum0, sum1, sum2, sum3, sum;
-
- vp9_get8x8var_mmx(src, src_stride, ref, ref_stride, &sse0, &sum0);
- vp9_get8x8var_mmx(src + 8, src_stride, ref + 8, ref_stride, &sse1, &sum1);
- vp9_get8x8var_mmx(src + 8 * src_stride, src_stride,
- ref + 8 * ref_stride, ref_stride, &sse2, &sum2);
- vp9_get8x8var_mmx(src + 8 * src_stride + 8, src_stride,
- ref + 8 * ref_stride + 8, ref_stride, &sse3, &sum3);
-
- *sse = sse0 + sse1 + sse2 + sse3;
- sum = sum0 + sum1 + sum2 + sum3;
- return *sse - (((unsigned int)sum * sum) >> 8);
-}
-
-unsigned int vp9_variance16x8_mmx(const uint8_t *src, int src_stride,
- const uint8_t *ref, int ref_stride,
- unsigned int *sse) {
- unsigned int sse0, sse1;
- int sum0, sum1, sum;
-
- vp9_get8x8var_mmx(src, src_stride, ref, ref_stride, &sse0, &sum0);
- vp9_get8x8var_mmx(src + 8, src_stride, ref + 8, ref_stride, &sse1, &sum1);
-
- *sse = sse0 + sse1;
- sum = sum0 + sum1;
- return *sse - (((unsigned int)sum * sum) >> 7);
-}
-
-
-unsigned int vp9_variance8x16_mmx(const uint8_t *src, int src_stride,
- const uint8_t *ref, int ref_stride,
- unsigned int *sse) {
- unsigned int sse0, sse1;
- int sum0, sum1, sum;
-
- vp9_get8x8var_mmx(src, src_stride, ref, ref_stride, &sse0, &sum0);
- vp9_get8x8var_mmx(src + 8 * src_stride, src_stride,
- ref + 8 * ref_stride, ref_stride, &sse1, &sum1);
-
- *sse = sse0 + sse1;
- sum = sum0 + sum1;
- return *sse - (((unsigned int)sum * sum) >> 7);
-}
diff --git a/vp9/vp9cx.mk b/vp9/vp9cx.mk
index dc46c4e..3bcecc2 100644
--- a/vp9/vp9cx.mk
+++ b/vp9/vp9cx.mk
@@ -93,8 +93,6 @@
VP9_CX_SRCS-yes += encoder/vp9_mbgraph.c
VP9_CX_SRCS-yes += encoder/vp9_mbgraph.h
-VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_variance_mmx.c
-VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_variance_impl_mmx.asm
VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_sad_mmx.asm
VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_variance_impl_sse2.asm
VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_variance_impl_intrin_avx2.c
