From 86c8f70db10a584aa78e4d5f397ad3543fdb77d2 Mon Sep 17 00:00:00 2001
From: Roland Scheidegger <sroland@vmware.com>
Date: Thu, 3 Dec 2009 23:26:13 +0100
Subject: mesa: use _mesa_memcpy for COPY_4FV macro

Gets rid of one of the worst strict-aliasing offenders, and actually
produces faster code (at least in some cases, when compiler can use
for instance 64bit moves for memcpy).
(note _mesa_memcpy should get inlined)
---
 src/mesa/main/macros.h | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

(limited to 'src/mesa/main')

diff --git a/src/mesa/main/macros.h b/src/mesa/main/macros.h
index 3d9a1aba98..f0ea463fb9 100644
--- a/src/mesa/main/macros.h
+++ b/src/mesa/main/macros.h
@@ -202,17 +202,12 @@ do {                                \
 #endif
 
 /**
- * Copy a 4-element float vector (avoid using FPU registers)
- * XXX Could use two 64-bit moves on 64-bit systems
+ * Copy a 4-element float vector
+ * memcpy seems to be most efficient
  */
 #define COPY_4FV( DST, SRC )                  \
 do {                                          \
-   const GLuint *_s = (const GLuint *) (SRC); \
-   GLuint *_d = (GLuint *) (DST);             \
-   _d[0] = _s[0];                             \
-   _d[1] = _s[1];                             \
-   _d[2] = _s[2];                             \
-   _d[3] = _s[3];                             \
+   _mesa_memcpy(DST, SRC, sizeof(GLfloat) * 4);       \
 } while (0)
 
 /** Copy \p SZ elements into a 4-element vector */
-- 
cgit v1.2.3


From c36d1aacf4c70d76165c91cd7048c0f9f43b8571 Mon Sep 17 00:00:00 2001
From: Roland Scheidegger <sroland@vmware.com>
Date: Mon, 7 Dec 2009 20:11:46 +0100
Subject: mesa: fix strict aliasing issues in half-to-float/float-to-half
 conversions

use union instead of casts
---
 src/mesa/main/imports.c | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

(limited to 'src/mesa/main')

diff --git a/src/mesa/main/imports.c b/src/mesa/main/imports.c
index c9e00cf752..f2ef84f35c 100644
--- a/src/mesa/main/imports.c
+++ b/src/mesa/main/imports.c
@@ -460,7 +460,7 @@ _mesa_inv_sqrtf(float n)
 #if 0 /* not used, see below -BP */
         float r3, x3, y3;
 #endif
-        union { float f; unsigned int i; } u;
+        fi_type u;
         unsigned int magic;
 
         /*
@@ -649,10 +649,10 @@ _mesa_bitcount(unsigned int n)
 GLhalfARB
 _mesa_float_to_half(float val)
 {
-   const int flt = *((int *) (void *) &val);
-   const int flt_m = flt & 0x7fffff;
-   const int flt_e = (flt >> 23) & 0xff;
-   const int flt_s = (flt >> 31) & 0x1;
+   const fi_type fi = {val};
+   const int flt_m = fi.i & 0x7fffff;
+   const int flt_e = (fi.i >> 23) & 0xff;
+   const int flt_s = (fi.i >> 31) & 0x1;
    int s, e, m = 0;
    GLhalfARB result;
    
@@ -739,7 +739,8 @@ _mesa_half_to_float(GLhalfARB val)
    const int m = val & 0x3ff;
    const int e = (val >> 10) & 0x1f;
    const int s = (val >> 15) & 0x1;
-   int flt_m, flt_e, flt_s, flt;
+   int flt_m, flt_e, flt_s;
+   fi_type fi;
    float result;
 
    /* sign bit */
@@ -774,8 +775,8 @@ _mesa_half_to_float(GLhalfARB val)
       flt_m = m << 13;
    }
 
-   flt = (flt_s << 31) | (flt_e << 23) | flt_m;
-   result = *((float *) (void *) &flt);
+   fi.i = (flt_s << 31) | (flt_e << 23) | flt_m;
+   result = fi.f;
    return result;
 }
 
-- 
cgit v1.2.3