ref: 979f0c0e5a901e9d6d21bcc5d566024acaaa22f9
parent: e9ba43543522f0fd2822627b2cb384119d1d17f2
parent: 73930f97634bfd163882151a30cb52b95eb69bec
author: Chi Yo Tsai <chiyotsai@google.com>
date: Wed Oct 24 12:36:20 EDT 2018
Merge "Clean up vpx_dsp/x86/convolve_sse2.h"
--- a/vpx_dsp/x86/convolve_sse2.h
+++ b/vpx_dsp/x86/convolve_sse2.h
@@ -11,7 +11,6 @@
#ifndef VPX_VPX_DSP_X86_CONVOLVE_SSE2_H_
#define VPX_VPX_DSP_X86_CONVOLVE_SSE2_H_
-#include <assert.h>
#include <emmintrin.h> // SSE2
#include "./vpx_config.h"
@@ -30,12 +29,13 @@
return _mm_unpacklo_epi64(tmp, tmp);
}
-// Interprets src as 8-bit words, pads each word with zeroes to form 16-bit
-// words, then multiplies with ker and add the adjacent results to form 32-bit
-// words. Finally adds the result from 1 and 2 together.
-static INLINE __m128i pad_multiply_add_add_epi8_sse2(
- const __m128i *const src_1, const __m128i *const src_2,
- const __m128i *const ker_1, const __m128i *const ker_2) {+// Interprets src as 8-bit words, zero extends to form 16-bit words, then
+// multiplies with ker and add the adjacent results to form 32-bit words.
+// Finally adds the result from 1 and 2 together.
+static INLINE __m128i multiply_add_epi8_sse2(const __m128i *const src_1,
+ const __m128i *const src_2,
+ const __m128i *const ker_1,
+ const __m128i *const ker_2) {const __m128i src_1_half = _mm_unpacklo_epi8(*src_1, _mm_setzero_si128());
const __m128i src_2_half = _mm_unpacklo_epi8(*src_2, _mm_setzero_si128());
const __m128i madd_1 = _mm_madd_epi16(src_1_half, *ker_1);
@@ -50,6 +50,8 @@
const __m128i madd_2 = _mm_madd_epi16(*src_1, *ker);
return _mm_packs_epi32(madd_1, madd_2);
}
+
+// Interleaves src_1 and src_2
static INLINE __m128i combine_epi32_sse2(const __m128i *const src_1,
const __m128i *const src_2) {const __m128i tmp_1 = _mm_unpacklo_epi32(*src_1, *src_2);
--- a/vpx_dsp/x86/vpx_subpixel_4t_intrin_sse2.c
+++ b/vpx_dsp/x86/vpx_subpixel_4t_intrin_sse2.c
@@ -54,12 +54,12 @@
src_reg_shift_3 = _mm_srli_si128(src_reg, 3);
// Output 6 4 2 0
- even = pad_multiply_add_add_epi8_sse2(&src_reg, &src_reg_shift_2,
- &kernel_reg_23, &kernel_reg_45);
+ even = multiply_add_epi8_sse2(&src_reg, &src_reg_shift_2, &kernel_reg_23,
+ &kernel_reg_45);
// Output 7 5 3 1
- odd = pad_multiply_add_add_epi8_sse2(&src_reg_shift_1, &src_reg_shift_3,
- &kernel_reg_23, &kernel_reg_45);
+ odd = multiply_add_epi8_sse2(&src_reg_shift_1, &src_reg_shift_3,
+ &kernel_reg_23, &kernel_reg_45);
// Combine to get the first half of the dst
dst_first = combine_epi32_sse2(&even, &odd);
@@ -71,12 +71,12 @@
src_reg_shift_3 = _mm_srli_si128(src_reg, 3);
// Output 14 12 10 8
- even = pad_multiply_add_add_epi8_sse2(&src_reg, &src_reg_shift_2,
- &kernel_reg_23, &kernel_reg_45);
+ even = multiply_add_epi8_sse2(&src_reg, &src_reg_shift_2, &kernel_reg_23,
+ &kernel_reg_45);
// Output 15 13 11 9
- odd = pad_multiply_add_add_epi8_sse2(&src_reg_shift_1, &src_reg_shift_3,
- &kernel_reg_23, &kernel_reg_45);
+ odd = multiply_add_epi8_sse2(&src_reg_shift_1, &src_reg_shift_3,
+ &kernel_reg_23, &kernel_reg_45);
// Combine to get the second half of the dst
dst_second = combine_epi32_sse2(&even, &odd);
@@ -288,12 +288,12 @@
src_reg_shift_3 = _mm_srli_si128(src_reg, 3);
// Output 6 4 2 0
- even = pad_multiply_add_add_epi8_sse2(&src_reg, &src_reg_shift_2,
- &kernel_reg_23, &kernel_reg_45);
+ even = multiply_add_epi8_sse2(&src_reg, &src_reg_shift_2, &kernel_reg_23,
+ &kernel_reg_45);
// Output 7 5 3 1
- odd = pad_multiply_add_add_epi8_sse2(&src_reg_shift_1, &src_reg_shift_3,
- &kernel_reg_23, &kernel_reg_45);
+ odd = multiply_add_epi8_sse2(&src_reg_shift_1, &src_reg_shift_3,
+ &kernel_reg_23, &kernel_reg_45);
// Combine to get the first half of the dst
dst_first = combine_epi32_sse2(&even, &odd);