shithub: libvpx

Download patch

ref: 3735def66700a90ef09d0a36e285f18479a9511c
parent: c98f8e04e573bdab8dbd2e9ab23354af58b2af68
author: James Zern <[email protected]>
date: Fri Apr 1 08:35:33 EDT 2016 

vpx_fdctNxN_1_sse2: reduce store size

only output[0] needs to be set, store_output is more involved than a
movdqa in the high bitdepth case

Change-Id: I2cbd85d7cf74688bdf47eb767934fe42e02bff67


--- a/vpx_dsp/x86/fwd_txfm_sse2.c
+++ b/vpx_dsp/x86/fwd_txfm_sse2.c
@@ -40,7 +40,7 @@
 
   in1 = _mm_add_epi32(tmp, in0);
   in0 = _mm_slli_epi32(in1, 1);
-  store_output(&in0, output);
+  output[0] = (tran_low_t)_mm_cvtsi128_si32(in0);
 }
 
 void vpx_fdct8x8_1_sse2(const int16_t *input, tran_low_t *output, int stride) {
@@ -80,7 +80,7 @@
   in0 = _mm_srli_si128(sum, 8);
 
   in1 = _mm_add_epi32(sum, in0);
-  store_output(&in1, output);
+  output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
 }
 
 void vpx_fdct16x16_1_sse2(const int16_t *input, tran_low_t *output,
@@ -149,7 +149,7 @@
 
   in1 = _mm_add_epi32(sum, in0);
   in1 = _mm_srai_epi32(in1, 1);
-  store_output(&in1, output);
+  output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
 }
 
 void vpx_fdct32x32_1_sse2(const int16_t *input, tran_low_t *output,
@@ -221,7 +221,7 @@
 
   in1 = _mm_add_epi32(sum, in0);
   in1 = _mm_srai_epi32(in1, 3);
-  store_output(&in1, output);
+  output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
 }
 
 #define DCT_HIGH_BIT_DEPTH 0