Mercurial > hg > dmlib

changeset 283:4d42b8910d7e

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Using SSE inline asm is not worth it in single operations, it hinders compiler optimizations too much.
author Matti Hamalainen <ccr@tnsp.org>
date Thu, 11 Oct 2012 08:50:03 +0300
parents 175328b20341
children d47e8cafaa63
files dmvecmat.c dmvecmat.h
diffstat 2 files changed, 48 insertions(+), 40 deletions(-) [+]
line wrap: on
line diff
--- a/dmvecmat.c	Thu Oct 11 08:08:46 2012 +0300
+++ b/dmvecmat.c	Thu Oct 11 08:50:03 2012 +0300
@@ -11,7 +11,19 @@
 {
     int i;
     for (i = 0; i < nlist; i++)
+    {
+#ifdef DM_USE_SIMD
+    asm("movups      %2,     %%xmm1\n"
+        "movups      %1,     %%xmm2\n"
+        "addps       %%xmm2, %%xmm1\n"
+        "movups      %%xmm1, %0\n"
+        : "=m" (dst[i])
+        : "m" (dst[i]), "m" (src[i])
+        : "memory", "%xmm1", "%xmm2");
+#else
         dm_vector_add(dst + i, src + i);
+#endif
+    }
 }
 
 
@@ -19,7 +31,19 @@
 {
     int i;
     for (i = 0; i < nlist; i++)
+    {
+#ifdef DM_USE_SIMD
+    asm("movups      %2,     %%xmm1\n"
+        "movups      %1,     %%xmm2\n"
+        "addps       %%xmm2, %%xmm1\n"
+        "movups      %%xmm1, %0\n"
+        : "=m" (dst[i])
+        : "m" (src1[i]), "m" (src2[i])
+        : "memory", "%xmm1", "%xmm2");
+#else
         dm_vector_add_r(dst + i, src1 + i, src2 + i);
+#endif
+    }
 }
 
 
@@ -27,7 +51,19 @@
 {
     int i;
     for (i = 0; i < nlist; i++)
+    {
+#ifdef DM_USE_SIMD
+    asm("movups      %2,     %%xmm1\n"
+        "movups      %1,     %%xmm2\n"
+        "subps       %%xmm2, %%xmm1\n"
+        "movups      %%xmm1, %0\n"
+        : "=m" (dst[i])
+        : "m" (dst[i]), "m" (src[i])
+        : "memory", "%xmm1", "%xmm2");
+#else
         dm_vector_add(dst + i, src + i);
+#endif
+    }
 }
 
 
@@ -35,7 +71,19 @@
 {
     int i;
     for (i = 0; i < nlist; i++)
+    {
+#ifdef DM_USE_SIMD
+    asm("movups      %2,     %%xmm1\n"
+        "movups      %1,     %%xmm2\n"
+        "subps       %%xmm2, %%xmm1\n"
+        "movups      %%xmm1, %0\n"
+        : "=m" (dst[i])
+        : "m" (src1[i]), "m" (src2[i])
+        : "memory", "%xmm1", "%xmm2");
+#else
         dm_vector_sub_r(dst + i, src1 + i, src2 + i);
+#endif
+    }
 }
 
 
--- a/dmvecmat.h	Thu Oct 11 08:08:46 2012 +0300
+++ b/dmvecmat.h	Thu Oct 11 08:50:03 2012 +0300
@@ -57,73 +57,33 @@
 
 static inline void dm_vector_add(DMVector *vr, const DMVector *v2)
 {
-#ifdef DM_USE_SIMD
-    asm("movups      %2,     %%xmm1\n"
-        "movups      %1,     %%xmm2\n"
-        "addps       %%xmm2, %%xmm1\n"
-        "movups      %%xmm1, %0\n"
-        : "=m" (*vr)
-        : "m" (*vr), "m" (*v2)
-        : "memory", "%xmm1", "%xmm2");
-#else
     vr->x += v2->x;
     vr->y += v2->y;
     vr->z += v2->z;
-#endif
 }
 
 
 static inline void dm_vector_add_r(DMVector *vr, const DMVector *v1, const DMVector *v2)
 {
-#ifdef DM_USE_SIMD
-    asm("movups      %2,     %%xmm1\n"
-        "movups      %1,     %%xmm2\n"
-        "addps       %%xmm2, %%xmm1\n"
-        "movups      %%xmm1, %0\n"
-        : "=m" (*vr)
-        : "m" (*v1), "m" (*v2)
-        : "memory", "%xmm1", "%xmm2");
-#else
     vr->x = v1->x + v2->x;
     vr->y = v1->y + v2->y;
     vr->z = v1->z + v2->z;
-#endif
 }
 
 
 static inline void dm_vector_sub(DMVector *vr, const DMVector *v2)
 {
-#ifdef DM_USE_SIMD
-    asm("movups      %2,     %%xmm1\n"
-        "movups      %1,     %%xmm2\n"
-        "subps       %%xmm2, %%xmm1\n"
-        "movups      %%xmm1, %0\n"
-        : "=m" (*vr)
-        : "m" (*vr), "m" (*v2)
-        : "memory", "%xmm1", "%xmm2");
-#else
     vr->x -= v2->x;
     vr->y -= v2->y;
     vr->z -= v2->z;
-#endif
 }
 
 
 static inline void dm_vector_sub_r(DMVector *vr, const DMVector *v1, const DMVector *v2)
 {
-#ifdef DM_USE_SIMD
-    asm("movups      %2,     %%xmm1\n"
-        "movups      %1,     %%xmm2\n"
-        "subps       %%xmm2, %%xmm1\n"
-        "movups      %%xmm1, %0\n"
-        : "=m" (*vr)
-        : "m" (*v1), "m" (*v2)
-        : "memory", "%xmm1", "%xmm2");
-#else
     vr->x = v1->x - v2->x;
     vr->y = v1->y - v2->y;
     vr->z = v1->z - v2->z;
-#endif
 }