foo.diff

commit cc4411da548016a7bcd94678e63b89170b852012
Author: Zeno Albisser <[email protected]>
Date:   Thu Sep 12 22:36:42 2013 +0200

    debug that one

diff --git a/common.c b/common.c
index dec4589..7079c2b 100644
--- a/common.c
+++ b/common.c
@@ -6,6 +6,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <emmintrin.h>
 
 #include "c63.h"
 #include "tables.h"
@@ -14,29 +15,44 @@ void dequantize_idct_row(int16_t *in_data, uint8_t *prediction, int w, int h,
     int y, uint8_t *out_data, uint8_t *quantization)
 {
   int x;
-
   int16_t block[8*8];
+  __m128i in;
+  __m128i pred;
+  __m128i mask_min = _mm_set1_epi16(255);
+  __m128i mask_max = _mm_setzero_si128();
 
   /* Perform the dequantization and iDCT */
   for(x = 0; x < w; x += 8)
   {
-    int i, j;
-
     dequant_idct_block_8x8(in_data+(x*8), block, quantization);
 
-    for (i = 0; i < 8; ++i)
+    int i;
+    for (i = 0; i < 8; i++)
     {
-      for (j = 0; j < 8; ++j)
-      {
-        /* Add prediction block. Note: DCT is not precise -
-           Clamp to legal values */
-        int16_t tmp = block[i*8+j] + (int16_t)prediction[i*w+j+x];
-
-        if (tmp < 0) { tmp = 0; }
-        else if (tmp > 255) { tmp = 255; }
-
-        out_data[i*w+j+x] = tmp;
-      }
+      // Fetch
+      in = _mm_loadl_epi64(((__m128i*)&block[i*8])); // Convert the type by changing pointer
+      in = _mm_unpacklo_epi8(in, _mm_setzero_si128());       // and then get values
+
+
+      // in = _mm_loadu_si128((__m128i*)&block[i*8]);
+      pred = _mm_loadl_epi64(((__m128i*)&prediction[i*w+x])); // Convert the type by changing pointer
+      pred = _mm_unpacklo_epi8(pred, _mm_setzero_si128());       // and then get values
+
+      // Compute
+      in = _mm_adds_epi16(in, pred);
+
+      // Clamp
+      // in = _mm_max_epi16(in, mask_max);
+      // in = _mm_min_epi16(in, mask_min);
+
+      // Save
+      in = _mm_packus_epi16 (in, _mm_setzero_si128());
+      uint8_t* p = &in;
+      fprintf(stderr, "%d %d %d %d   %d %d %d %d   ", p[15], p[14], p[13], p[12], p[11], p[10], p[9], p[8]);
+      fprintf(stderr, "%d %d %d %d   %d %d %d %d\n", p[7], p[6], p[5], p[4], p[3], p[2], p[1], p[0]);
+      // Save
+      _mm_storel_epi64((__m128i*)(out_data+(i*w+x)), in);
+      // _mm_storeu_si128((__m128i*)&out_data[i*w+x], in);
     }
   }
 }