summaryrefslogtreecommitdiffstats
path: root/libgsmhr/laforge-mmx.diff
blob: a846c25e636cd6eb686e773936768ba87dbef238 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
Only in refsrc: .downloaded
Only in refsrc: .sp_frm.c.swp
diff -u refsrc.orig/sp_frm.c refsrc/sp_frm.c
--- refsrc.orig/sp_frm.c	2015-12-27 19:22:13.966296058 +0100
+++ refsrc/sp_frm.c	2014-05-13 22:43:56.786205819 +0200
@@ -60,6 +60,13 @@
  *
  **************************************************************************/
 
+#include <stdio.h>
+
+#include <stdint.h>
+//#define HAVE_MMX
+extern int32_t mmx_mac_unsat(int16_t *_x, int16_t *_y);
+extern int32_t mmx_mac_sat(int16_t *_x, int16_t *_y);
+
 /*_________________________________________________________________________
  |                                                                         |
  |                            Include Files                                |
@@ -384,6 +391,7 @@
         /* get a vector */
         /*--------------*/
 
+	      // 16, 32 or 64 iteraitons
         getNextVec(pswRc);
 
         /* clear the limiter flag */
@@ -432,6 +440,8 @@
         for (iCnt = 0; iCnt < quantList.iNum; iCnt++)
         {
 
+	      // 4 * 16, 32 or 64 iteraitons
+
           /* get a vector */
           /*--------------*/
 
@@ -931,6 +941,7 @@
 
     for (i = 0; i <= bound; i++)
     {
+	    // 3-4 iterations
       L_sum = L_mac(L_ROUND, pswVOld[i], pswQntRc[j]);
       L_sum = L_mac(L_sum, pswVOld[-i], pswQntRc[j]);
       L_sum = L_mac(L_sum, pswPOld[i], pswQntRcSqd[j]);
@@ -949,6 +960,7 @@
 
     for (i = -bound; i < 0; i++)
     {
+	    // 3-4 iterations
       L_sum = L_msu(L_ROUND, pswVOld[i + 1], SW_MIN);
       L_sum = L_mac(L_sum, pswQntRcSqd[j], pswVOld[-i - 1]);
       L_sum = L_mac(L_sum, pswQntRc[j], pswPOld[-i - 1]);
@@ -958,6 +970,7 @@
 
     for (i = 0; i <= bound; i++)
     {
+	    // 3-4 iterations
       L_sum = L_msu(L_ROUND, pswVOld[i + 1], SW_MIN);
       L_sum = L_mac(L_sum, pswQntRcSqd[j], pswVOld[-i - 1]);
       L_sum = L_mac(L_sum, pswQntRc[j], pswPOld[i + 1]);
@@ -4536,6 +4549,11 @@
 
   pswScaledWSpeech = pswScaledWSpeechBuffer + LSMAX;
 
+#if 0
+  printf("G_FRAME_LEN=%d\n", G_FRAME_LEN);
+  printf("LSMIN=%d\n", LSMIN);
+  printf("LSMAX=%d\n", LSMAX);
+#endif
 /*_________________________________________________________________________
  |                                                                         |
  |                            Executable Code                              |
@@ -4633,11 +4651,20 @@
   /*---------------------------------------------------------------------*/
 
   L_G = 0;
+#ifndef HAVE_MMX
+  // 40 iterations (MMX: 5)
   for (i = -LSMAX; i < -LSMAX + S_LEN; i++)
     L_G = L_mac(L_G, pswScaledWSpeech[i], pswScaledWSpeech[i]);
+#else
+  for (i = -LSMAX; i < -LSMAX + S_LEN; i+=8)
+    L_G += mmx_mac_unsat(&pswScaledWSpeech[i], &pswScaledWSpeech[i]);
+#endif
 
   pswGFrame[G_FRAME_LEN - 1] = extract_h(L_G);
 
+//#ifndef HAVE_MMX
+#if 1
+  // 248 iterations (MMX: 31)
   for (i = -LSMAX; i < G_FRAME_LEN - LSMAX - 1; i++)
   {
 
@@ -4646,6 +4673,13 @@
                 pswScaledWSpeech[i + S_LEN]);
     pswGFrame[G_FRAME_LEN - LSMAX - 2 - i] = extract_h(L_G);
   }
+#else
+  for (i = -LSMAX; i < G_FRAME_LEN - LSMAX - 1; i+= 8) {
+	  L_G -= mmx_mac_sat(&pswScaledWSpeech[i], &pswScaledWSpeech[i]);
+	  L_G += mmx_mac_sat(&pswScaledWSpeech[i + S_LEN],
+			     &pswScaledWSpeech[i + S_LEN]);
+  } 
+#endif
 
   ppswGSfrm[0] = pswGFrame + 3 * S_LEN;
   ppswGSfrm[1] = pswGFrame + 2 * S_LEN;
@@ -4661,8 +4695,14 @@
   pswSfrmEng[2] = pswGFrame[G_FRAME_LEN - 1 - LSMAX - 2 * S_LEN];
 
   L_WSfrmEng = 0;
+#ifndef HAVE_MMX
+  // 40 iterations (MMX: 5)
   for (i = F_LEN - S_LEN; i < F_LEN; i++)
     L_WSfrmEng = L_mac(L_WSfrmEng, pswScaledWSpeech[i], pswScaledWSpeech[i]);
+#else
+  for (i = F_LEN - S_LEN; i < F_LEN; i+= 8)
+    L_WSfrmEng += mmx_mac_unsat(&pswScaledWSpeech[i], &pswScaledWSpeech[i]);
+#endif
 
   pswSfrmEng[3] = extract_h(L_WSfrmEng);
 
@@ -4671,19 +4711,26 @@
   /* as in the G buffer.)                                       */
   /*------------------------------------------------------------*/
 
+  // 4 iterations
   for (i = 0; i < N_SUB; i++)
   {
 
+    // 127 iterations
     for (j = LSMIN; j <= LSMAX; j++)
     {
-
       L_C = 0;
+#ifndef HAVE_MMX
+      // 4*127*40 iterations (MMX: 4*127*5)
       for (k = 0; k < S_LEN; k++)
       {
-
         L_C = L_mac(L_C, pswScaledWSpeech[i * S_LEN + k],
                     pswScaledWSpeech[i * S_LEN - j + k]);
       }
+#else
+      for (k = 0; k < S_LEN; k+= 8)
+	      L_C += mmx_mac_unsat(&pswScaledWSpeech[i*S_LEN + k],
+			      	   &pswScaledWSpeech[i*S_LEN - j + k]);
+#endif
 
       pswCFrame[i * CG_TERMS + j - LSMIN] = extract_h(L_C);
     }
@@ -4750,6 +4797,7 @@
 
   L_Voicing = 0;
   for (i = 0; i < N_SUB; i++)
+	  // 4 Iterations
     L_Voicing = L_mac(L_Voicing, pswSfrmEng[i], UV_SCALE0);
 
   L_Voicing = L_add(L_Voicing, L_deposit_h(swBestPG));
@@ -5085,6 +5133,7 @@
       siLowestSoFar = 2;
       for (i = 0; i < N_SUB; i++)
       {
+	      // 4 iterations
 
         /* Check this subframe against highest voicing threshold */
         /*-------------------------------------------------------*/