ref: fec20805f773e3a2c4b98f488f2cbf133ef72a28
parent: d3d22aa7c9ff19af3af2cf21cd7c7739fc2c6263
parent: 5757b919c8b6374c4d7327b5c4b1a8cfc707c2bf
author: Johann Koenig <[email protected]>
date: Tue Nov 27 13:47:33 EST 2018
Merge "rename quantize_x86.h"
--- a/vp9/encoder/x86/vp9_quantize_avx2.c
+++ b/vp9/encoder/x86/vp9_quantize_avx2.c
@@ -15,7 +15,7 @@
#include "vpx/vpx_integer.h"
#include "vpx_dsp/vpx_dsp_common.h"
#include "vpx_dsp/x86/bitdepth_conversion_avx2.h"
-#include "vpx_dsp/x86/quantize_x86.h"
+#include "vpx_dsp/x86/quantize_sse2.h"
// Zero fill 8 positions in the output buffer.
static INLINE void store_zero_tran_low(tran_low_t *a) {
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -296,8 +296,8 @@
DSP_SRCS-yes += quantize.c
DSP_SRCS-yes += quantize.h
-DSP_SRCS-$(HAVE_SSE2) += x86/quantize_x86.h
DSP_SRCS-$(HAVE_SSE2) += x86/quantize_sse2.c
+DSP_SRCS-$(HAVE_SSE2) += x86/quantize_sse2.h
DSP_SRCS-$(HAVE_SSSE3) += x86/quantize_ssse3.c
DSP_SRCS-$(HAVE_AVX) += x86/quantize_avx.c
DSP_SRCS-$(HAVE_NEON) += arm/quantize_neon.c
--- a/vpx_dsp/x86/quantize_avx.c
+++ b/vpx_dsp/x86/quantize_avx.c
@@ -17,7 +17,7 @@
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
-#include "vpx_dsp/x86/quantize_x86.h"
+#include "vpx_dsp/x86/quantize_sse2.h"
void vpx_quantize_b_avx(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
--- a/vpx_dsp/x86/quantize_sse2.c
+++ b/vpx_dsp/x86/quantize_sse2.c
@@ -15,7 +15,7 @@
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
-#include "vpx_dsp/x86/quantize_x86.h"
+#include "vpx_dsp/x86/quantize_sse2.h"
void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
--- /dev/null
+++ b/vpx_dsp/x86/quantize_sse2.h
@@ -1,0 +1,82 @@
+/*
+ * Copyright (c) 2017 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.
+ */
+
+#ifndef VPX_VPX_DSP_X86_QUANTIZE_SSE2_H_
+#define VPX_VPX_DSP_X86_QUANTIZE_SSE2_H_
+
+#include <emmintrin.h>
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+
+static INLINE void load_b_values(const int16_t *zbin_ptr, __m128i *zbin,
+ const int16_t *round_ptr, __m128i *round,
+ const int16_t *quant_ptr, __m128i *quant,
+ const int16_t *dequant_ptr, __m128i *dequant,
+ const int16_t *shift_ptr, __m128i *shift) {
+ *zbin = _mm_load_si128((const __m128i *)zbin_ptr);
+ *round = _mm_load_si128((const __m128i *)round_ptr);
+ *quant = _mm_load_si128((const __m128i *)quant_ptr);
+ *zbin = _mm_sub_epi16(*zbin, _mm_set1_epi16(1));
+ *dequant = _mm_load_si128((const __m128i *)dequant_ptr);
+ *shift = _mm_load_si128((const __m128i *)shift_ptr);
+}
+
+// With ssse3 and later abs() and sign() are preferred.
+static INLINE __m128i invert_sign_sse2(__m128i a, __m128i sign) {
+ a = _mm_xor_si128(a, sign);
+ return _mm_sub_epi16(a, sign);
+}
+
+static INLINE void calculate_qcoeff(__m128i *coeff, const __m128i round,
+ const __m128i quant, const __m128i shift) {
+ __m128i tmp, qcoeff;
+ qcoeff = _mm_adds_epi16(*coeff, round);
+ tmp = _mm_mulhi_epi16(qcoeff, quant);
+ qcoeff = _mm_add_epi16(tmp, qcoeff);
+ *coeff = _mm_mulhi_epi16(qcoeff, shift);
+}
+
+static INLINE __m128i calculate_dqcoeff(__m128i qcoeff, __m128i dequant) {
+ return _mm_mullo_epi16(qcoeff, dequant);
+}
+
+// Scan 16 values for eob reference in scan. Use masks (-1) from comparing to
+// zbin to add 1 to the index in 'scan'.
+static INLINE __m128i scan_for_eob(__m128i *coeff0, __m128i *coeff1,
+ const __m128i zbin_mask0,
+ const __m128i zbin_mask1,
+ const int16_t *scan, const int index,
+ const __m128i zero) {
+ const __m128i zero_coeff0 = _mm_cmpeq_epi16(*coeff0, zero);
+ const __m128i zero_coeff1 = _mm_cmpeq_epi16(*coeff1, zero);
+ __m128i scan0 = _mm_load_si128((const __m128i *)(scan + index));
+ __m128i scan1 = _mm_load_si128((const __m128i *)(scan + index + 8));
+ __m128i eob0, eob1;
+ // Add one to convert from indices to counts
+ scan0 = _mm_sub_epi16(scan0, zbin_mask0);
+ scan1 = _mm_sub_epi16(scan1, zbin_mask1);
+ eob0 = _mm_andnot_si128(zero_coeff0, scan0);
+ eob1 = _mm_andnot_si128(zero_coeff1, scan1);
+ return _mm_max_epi16(eob0, eob1);
+}
+
+static INLINE int16_t accumulate_eob(__m128i eob) {
+ __m128i eob_shuffled;
+ eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
+ eob = _mm_max_epi16(eob, eob_shuffled);
+ eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
+ eob = _mm_max_epi16(eob, eob_shuffled);
+ eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
+ eob = _mm_max_epi16(eob, eob_shuffled);
+ return _mm_extract_epi16(eob, 1);
+}
+
+#endif // VPX_VPX_DSP_X86_QUANTIZE_SSE2_H_
--- a/vpx_dsp/x86/quantize_ssse3.c
+++ b/vpx_dsp/x86/quantize_ssse3.c
@@ -14,7 +14,7 @@
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
-#include "vpx_dsp/x86/quantize_x86.h"
+#include "vpx_dsp/x86/quantize_sse2.h"
void vpx_quantize_b_ssse3(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
--- a/vpx_dsp/x86/quantize_x86.h
+++ /dev/null
@@ -1,82 +1,0 @@
-/*
- * Copyright (c) 2017 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.
- */
-
-#ifndef VPX_VPX_DSP_X86_QUANTIZE_X86_H_
-#define VPX_VPX_DSP_X86_QUANTIZE_X86_H_
-
-#include <emmintrin.h>
-
-#include "./vpx_config.h"
-#include "vpx/vpx_integer.h"
-
-static INLINE void load_b_values(const int16_t *zbin_ptr, __m128i *zbin,
- const int16_t *round_ptr, __m128i *round,
- const int16_t *quant_ptr, __m128i *quant,
- const int16_t *dequant_ptr, __m128i *dequant,
- const int16_t *shift_ptr, __m128i *shift) {
- *zbin = _mm_load_si128((const __m128i *)zbin_ptr);
- *round = _mm_load_si128((const __m128i *)round_ptr);
- *quant = _mm_load_si128((const __m128i *)quant_ptr);
- *zbin = _mm_sub_epi16(*zbin, _mm_set1_epi16(1));
- *dequant = _mm_load_si128((const __m128i *)dequant_ptr);
- *shift = _mm_load_si128((const __m128i *)shift_ptr);
-}
-
-// With ssse3 and later abs() and sign() are preferred.
-static INLINE __m128i invert_sign_sse2(__m128i a, __m128i sign) {
- a = _mm_xor_si128(a, sign);
- return _mm_sub_epi16(a, sign);
-}
-
-static INLINE void calculate_qcoeff(__m128i *coeff, const __m128i round,
- const __m128i quant, const __m128i shift) {
- __m128i tmp, qcoeff;
- qcoeff = _mm_adds_epi16(*coeff, round);
- tmp = _mm_mulhi_epi16(qcoeff, quant);
- qcoeff = _mm_add_epi16(tmp, qcoeff);
- *coeff = _mm_mulhi_epi16(qcoeff, shift);
-}
-
-static INLINE __m128i calculate_dqcoeff(__m128i qcoeff, __m128i dequant) {
- return _mm_mullo_epi16(qcoeff, dequant);
-}
-
-// Scan 16 values for eob reference in scan. Use masks (-1) from comparing to
-// zbin to add 1 to the index in 'scan'.
-static INLINE __m128i scan_for_eob(__m128i *coeff0, __m128i *coeff1,
- const __m128i zbin_mask0,
- const __m128i zbin_mask1,
- const int16_t *scan, const int index,
- const __m128i zero) {
- const __m128i zero_coeff0 = _mm_cmpeq_epi16(*coeff0, zero);
- const __m128i zero_coeff1 = _mm_cmpeq_epi16(*coeff1, zero);
- __m128i scan0 = _mm_load_si128((const __m128i *)(scan + index));
- __m128i scan1 = _mm_load_si128((const __m128i *)(scan + index + 8));
- __m128i eob0, eob1;
- // Add one to convert from indices to counts
- scan0 = _mm_sub_epi16(scan0, zbin_mask0);
- scan1 = _mm_sub_epi16(scan1, zbin_mask1);
- eob0 = _mm_andnot_si128(zero_coeff0, scan0);
- eob1 = _mm_andnot_si128(zero_coeff1, scan1);
- return _mm_max_epi16(eob0, eob1);
-}
-
-static INLINE int16_t accumulate_eob(__m128i eob) {
- __m128i eob_shuffled;
- eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
- eob = _mm_max_epi16(eob, eob_shuffled);
- return _mm_extract_epi16(eob, 1);
-}
-
-#endif // VPX_VPX_DSP_X86_QUANTIZE_X86_H_