ref: 170a4bb4b8869c919f6f631eeb073357859a1003
parent: fea1bd10a4d2005de5dc61cf227b8a44b96e6b24
parent: 7779e9c9118222c6c2dee7bdd0e2511da0c4b359
author: Johann Koenig <johannkoenig@google.com>
date: Fri Dec 21 14:29:25 EST 2018
Merge changes I48b9a9cd,I92504ed4 * changes: subpixel_8t ssse3: resolve missing declarations subpixel_8t avx2: resolve missing declarations
--- a/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c
+++ b/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c
@@ -328,9 +328,10 @@
height, filter, 1);
}
-void vpx_filter_block1d16_h4_avx2(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d16_h4_avx2(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will cast the kernel from 16-bit words to 8-bit words, and then extract
// the middle four elements of the kernel into two registers in the form
// ... k[3] k[2] k[3] k[2]
@@ -432,9 +433,10 @@
}
}
-void vpx_filter_block1d16_v4_avx2(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d16_v4_avx2(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will load two rows of pixels as 8-bit words, rearrange them into the
// form
// ... s[1,0] s[0,0] s[0,0] s[-1,0]
@@ -537,9 +539,10 @@
}
}
-void vpx_filter_block1d8_h4_avx2(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d8_h4_avx2(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will cast the kernel from 16-bit words to 8-bit words, and then extract
// the middle four elements of the kernel into two registers in the form
// ... k[3] k[2] k[3] k[2]
@@ -631,9 +634,10 @@
}
}
-void vpx_filter_block1d8_v4_avx2(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d8_v4_avx2(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will load two rows of pixels as 8-bit words, rearrange them into the
// form
// ... s[1,0] s[0,0] s[0,0] s[-1,0]
@@ -725,9 +729,10 @@
}
}
-void vpx_filter_block1d4_h4_avx2(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d4_h4_avx2(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will cast the kernel from 16-bit words to 8-bit words, and then extract
// the middle four elements of the kernel into a single register in the form
// k[5:2] k[5:2] k[5:2] k[5:2]
@@ -805,9 +810,10 @@
}
}
-void vpx_filter_block1d4_v4_avx2(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d4_v4_avx2(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will load two rows of pixels as 8-bit words, rearrange them into the
// form
// ... s[3,0] s[2,0] s[1,0] s[0,0] s[2,0] s[1,0] s[0,0] s[-1,0]
--- a/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c
+++ b/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c
@@ -12,21 +12,17 @@
#include <string.h>
+#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/vpx_filter.h"
#include "vpx_dsp/x86/convolve.h"
-#include "vpx_dsp/x86/convolve_ssse3.h"
#include "vpx_dsp/x86/convolve_sse2.h"
+#include "vpx_dsp/x86/convolve_ssse3.h"
#include "vpx_dsp/x86/mem_sse2.h"
#include "vpx_dsp/x86/transpose_sse2.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
-// These are reused by the avx2 intrinsics.
-// vpx_filter_block1d8_v8_intrin_ssse3()
-// vpx_filter_block1d8_h8_intrin_ssse3()
-// vpx_filter_block1d4_h8_intrin_ssse3()
-
static INLINE __m128i shuffle_filter_convolve8_8_ssse3(
const __m128i *const s, const int16_t *const filter) {__m128i f[4];
@@ -34,6 +30,22 @@
return convolve8_8_ssse3(s, f);
}
+// Used by the avx2 implementation.
+#if ARCH_X86_64
+// Use the intrinsics below
+filter8_1dfunction vpx_filter_block1d4_h8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d8_v8_intrin_ssse3;
+#define vpx_filter_block1d4_h8_ssse3 vpx_filter_block1d4_h8_intrin_ssse3
+#define vpx_filter_block1d8_h8_ssse3 vpx_filter_block1d8_h8_intrin_ssse3
+#define vpx_filter_block1d8_v8_ssse3 vpx_filter_block1d8_v8_intrin_ssse3
+#else // ARCH_X86
+// Use the assembly in vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm.
+filter8_1dfunction vpx_filter_block1d4_h8_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h8_ssse3;
+filter8_1dfunction vpx_filter_block1d8_v8_ssse3;
+#endif
+
void vpx_filter_block1d4_h8_intrin_ssse3(
const uint8_t *src_ptr, ptrdiff_t src_pitch, uint8_t *output_ptr,
ptrdiff_t output_pitch, uint32_t output_height, const int16_t *filter) {@@ -186,9 +198,11 @@
}
}
-void vpx_filter_block1d16_h4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d16_h4_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_stride,
+ uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will cast the kernel from 16-bit words to 8-bit words, and then extract
// the middle four elements of the kernel into two registers in the form
// ... k[3] k[2] k[3] k[2]
@@ -258,9 +272,11 @@
}
}
-void vpx_filter_block1d16_v4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d16_v4_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_stride,
+ uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will load two rows of pixels as 8-bit words, rearrange them into the
// form
// ... s[0,1] s[-1,1] s[0,0] s[-1,0]
@@ -372,9 +388,10 @@
}
}
-void vpx_filter_block1d8_h4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d8_h4_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will cast the kernel from 16-bit words to 8-bit words, and then extract
// the middle four elements of the kernel into two registers in the form
// ... k[3] k[2] k[3] k[2]
@@ -432,9 +449,10 @@
}
}
-void vpx_filter_block1d8_v4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d8_v4_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will load two rows of pixels as 8-bit words, rearrange them into the
// form
// ... s[0,1] s[-1,1] s[0,0] s[-1,0]
@@ -524,9 +542,10 @@
}
}
-void vpx_filter_block1d4_h4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d4_h4_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will cast the kernel from 16-bit words to 8-bit words, and then extract
// the middle four elements of the kernel into a single register in the form
// k[5:2] k[5:2] k[5:2] k[5:2]
@@ -574,9 +593,10 @@
}
}
-void vpx_filter_block1d4_v4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_stride,
- uint8_t *dst_ptr, ptrdiff_t dst_stride,
- uint32_t height, const int16_t *kernel) {+static void vpx_filter_block1d4_v4_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_stride, uint8_t *dst_ptr,
+ ptrdiff_t dst_stride, uint32_t height,
+ const int16_t *kernel) {// We will load two rows of pixels as 8-bit words, rearrange them into the
// form
// ... s[2,0] s[1,0] s[0,0] s[-1,0]
@@ -668,12 +688,10 @@
}
}
+// From vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm
filter8_1dfunction vpx_filter_block1d16_v8_ssse3;
filter8_1dfunction vpx_filter_block1d16_h8_ssse3;
-filter8_1dfunction vpx_filter_block1d8_v8_ssse3;
-filter8_1dfunction vpx_filter_block1d8_h8_ssse3;
filter8_1dfunction vpx_filter_block1d4_v8_ssse3;
-filter8_1dfunction vpx_filter_block1d4_h8_ssse3;
filter8_1dfunction vpx_filter_block1d16_v8_avg_ssse3;
filter8_1dfunction vpx_filter_block1d16_h8_avg_ssse3;
filter8_1dfunction vpx_filter_block1d8_v8_avg_ssse3;
@@ -681,6 +699,7 @@
filter8_1dfunction vpx_filter_block1d4_v8_avg_ssse3;
filter8_1dfunction vpx_filter_block1d4_h8_avg_ssse3;
+// Use the [vh]8 version because there is no [vh]4 implementation.
#define vpx_filter_block1d16_v4_avg_ssse3 vpx_filter_block1d16_v8_avg_ssse3
#define vpx_filter_block1d16_h4_avg_ssse3 vpx_filter_block1d16_h8_avg_ssse3
#define vpx_filter_block1d8_v4_avg_ssse3 vpx_filter_block1d8_v8_avg_ssse3
@@ -688,6 +707,7 @@
#define vpx_filter_block1d4_v4_avg_ssse3 vpx_filter_block1d4_v8_avg_ssse3
#define vpx_filter_block1d4_h4_avg_ssse3 vpx_filter_block1d4_h8_avg_ssse3
+// From vpx_dsp/x86/vpx_subpixel_bilinear_ssse3.asm
filter8_1dfunction vpx_filter_block1d16_v2_ssse3;
filter8_1dfunction vpx_filter_block1d16_h2_ssse3;
filter8_1dfunction vpx_filter_block1d8_v2_ssse3;
@@ -1061,7 +1081,7 @@
}
}
-// void vp9_convolve8_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+// void vpx_convolve8_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,