Merge "Extract high bit depth helper functions"

3 files changed, 64 insertions, 42 deletions

diff --git a/vpx_dsp/vpx_dsp.mk b/vpx_dsp/vpx_dsp.mk
index 8c91b141f..2909beb0f 100644
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -252,6 +252,7 @@ ifeq ($(CONFIG_VP9_ENCODER),yes)
 DSP_SRCS-yes            += quantize.c
 DSP_SRCS-yes            += quantize.h
 
+DSP_SRCS-$(HAVE_SSE2)   += x86/fdct.h
 DSP_SRCS-$(HAVE_SSE2)   += x86/quantize_sse2.c
 ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
 DSP_SRCS-$(HAVE_SSE2)   += x86/highbd_quantize_intrin_sse2.c
diff --git a/vpx_dsp/x86/fdct.h b/vpx_dsp/x86/fdct.h
new file mode 100644
index 000000000..cad73f82a
--- /dev/null
+++ b/vpx_dsp/x86/fdct.h
@@ -0,0 +1,46 @@
+/*
+ *  Copyright (c) 2016 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_DSP_X86_FDCT_H_
+#define VPX_DSP_X86_FDCT_H_
+
+#include <xmmintrin.h>
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_dsp/vpx_dsp_common.h"
+
+// Load 8 16 bit values. If the source is 32 bits then cast down.
+// This does not saturate values. It only truncates.
+static INLINE __m128i load_tran_low(const tran_low_t *a) {
+#if CONFIG_VP9_HIGHBITDEPTH
+  return _mm_setr_epi16((int16_t)a[0], (int16_t)a[1], (int16_t)a[2],
+                        (int16_t)a[3], (int16_t)a[4], (int16_t)a[5],
+                        (int16_t)a[6], (int16_t)a[7]);
+#else
+  return _mm_load_si128((const __m128i *)a);
+#endif
+}
+
+// Store 8 16 bit values. If the destination is 32 bits then sign extend the
+// values by multiplying by 1.
+static INLINE void store_tran_low(__m128i a, tran_low_t *b) {
+#if CONFIG_VP9_HIGHBITDEPTH
+  const __m128i one = _mm_set1_epi16(1);
+  const __m128i a_hi = _mm_mulhi_epi16(a, one);
+  const __m128i a_lo = _mm_mullo_epi16(a, one);
+  const __m128i a_1 = _mm_unpacklo_epi16(a_lo, a_hi);
+  const __m128i a_2 = _mm_unpackhi_epi16(a_lo, a_hi);
+  _mm_store_si128((__m128i *)(b), a_1);
+  _mm_store_si128((__m128i *)(b + 4), a_2);
+#else
+  _mm_store_si128((__m128i *)(b), a);
+#endif
+}
+#endif  // VPX_DSP_X86_FDCT_H_
diff --git a/vpx_dsp/x86/quantize_sse2.c b/vpx_dsp/x86/quantize_sse2.c
index 2c7e431c7..0580a7bd7 100644
--- a/vpx_dsp/x86/quantize_sse2.c
+++ b/vpx_dsp/x86/quantize_sse2.c
@@ -13,32 +13,7 @@
 
 #include "./vpx_dsp_rtcd.h"
 #include "vpx/vpx_integer.h"
-
-static INLINE __m128i load_coefficients(const tran_low_t *coeff_ptr) {
-#if CONFIG_VP9_HIGHBITDEPTH
-  return _mm_setr_epi16((int16_t)coeff_ptr[0], (int16_t)coeff_ptr[1],
-                        (int16_t)coeff_ptr[2], (int16_t)coeff_ptr[3],
-                        (int16_t)coeff_ptr[4], (int16_t)coeff_ptr[5],
-                        (int16_t)coeff_ptr[6], (int16_t)coeff_ptr[7]);
-#else
-  return _mm_load_si128((const __m128i *)coeff_ptr);
-#endif
-}
-
-static INLINE void store_coefficients(__m128i coeff_vals,
-                                      tran_low_t *coeff_ptr) {
-#if CONFIG_VP9_HIGHBITDEPTH
-  __m128i one = _mm_set1_epi16(1);
-  __m128i coeff_vals_hi = _mm_mulhi_epi16(coeff_vals, one);
-  __m128i coeff_vals_lo = _mm_mullo_epi16(coeff_vals, one);
-  __m128i coeff_vals_1 = _mm_unpacklo_epi16(coeff_vals_lo, coeff_vals_hi);
-  __m128i coeff_vals_2 = _mm_unpackhi_epi16(coeff_vals_lo, coeff_vals_hi);
-  _mm_store_si128((__m128i *)(coeff_ptr), coeff_vals_1);
-  _mm_store_si128((__m128i *)(coeff_ptr + 4), coeff_vals_2);
-#else
-  _mm_store_si128((__m128i *)(coeff_ptr), coeff_vals);
-#endif
-}
+#include "vpx_dsp/x86/fdct.h"
 
 void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                          int skip_block, const int16_t *zbin_ptr,
@@ -81,8 +56,8 @@ void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
         __m128i qtmp0, qtmp1;
         __m128i cmp_mask0, cmp_mask1;
         // Do DC and first 15 AC
-        coeff0 = load_coefficients(coeff_ptr + n_coeffs);
-        coeff1 = load_coefficients(coeff_ptr + n_coeffs + 8);
+        coeff0 = load_tran_low(coeff_ptr + n_coeffs);
+        coeff1 = load_tran_low(coeff_ptr + n_coeffs + 8);
 
         // Poor man's sign extract
         coeff0_sign = _mm_srai_epi16(coeff0, 15);
@@ -117,15 +92,15 @@ void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
         qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
         qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
 
-        store_coefficients(qcoeff0, qcoeff_ptr + n_coeffs);
-        store_coefficients(qcoeff1, qcoeff_ptr + n_coeffs + 8);
+        store_tran_low(qcoeff0, qcoeff_ptr + n_coeffs);
+        store_tran_low(qcoeff1, qcoeff_ptr + n_coeffs + 8);
 
         coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
         dequant = _mm_unpackhi_epi64(dequant, dequant);
         coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
 
-        store_coefficients(coeff0, dqcoeff_ptr + n_coeffs);
-        store_coefficients(coeff1, dqcoeff_ptr + n_coeffs + 8);
+        store_tran_low(coeff0, dqcoeff_ptr + n_coeffs);
+        store_tran_low(coeff1, dqcoeff_ptr + n_coeffs + 8);
       }
 
       {
@@ -159,8 +134,8 @@ void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
         __m128i qtmp0, qtmp1;
         __m128i cmp_mask0, cmp_mask1;
 
-        coeff0 = load_coefficients(coeff_ptr + n_coeffs);
-        coeff1 = load_coefficients(coeff_ptr + n_coeffs + 8);
+        coeff0 = load_tran_low(coeff_ptr + n_coeffs);
+        coeff1 = load_tran_low(coeff_ptr + n_coeffs + 8);
 
         // Poor man's sign extract
         coeff0_sign = _mm_srai_epi16(coeff0, 15);
@@ -191,14 +166,14 @@ void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
         qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
         qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
 
-        store_coefficients(qcoeff0, qcoeff_ptr + n_coeffs);
-        store_coefficients(qcoeff1, qcoeff_ptr + n_coeffs + 8);
+        store_tran_low(qcoeff0, qcoeff_ptr + n_coeffs);
+        store_tran_low(qcoeff1, qcoeff_ptr + n_coeffs + 8);
 
         coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
         coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
 
-        store_coefficients(coeff0, dqcoeff_ptr + n_coeffs);
-        store_coefficients(coeff1, dqcoeff_ptr + n_coeffs + 8);
+        store_tran_low(coeff0, dqcoeff_ptr + n_coeffs);
+        store_tran_low(coeff1, dqcoeff_ptr + n_coeffs + 8);
       }
 
       {
@@ -237,10 +212,10 @@ void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
     }
   } else {
     do {
-      store_coefficients(zero, dqcoeff_ptr + n_coeffs);
-      store_coefficients(zero, dqcoeff_ptr + n_coeffs + 8);
-      store_coefficients(zero, qcoeff_ptr + n_coeffs);
-      store_coefficients(zero, qcoeff_ptr + n_coeffs + 8);
+      store_tran_low(zero, dqcoeff_ptr + n_coeffs);
+      store_tran_low(zero, dqcoeff_ptr + n_coeffs + 8);
+      store_tran_low(zero, qcoeff_ptr + n_coeffs);
+      store_tran_low(zero, qcoeff_ptr + n_coeffs + 8);
       n_coeffs += 8 * 2;
     } while (n_coeffs < 0);
     *eob_ptr = 0;