Commit: 3ec168465d31c90cc75b5ea70de492dbeb4ac992
Author: Sv. Lockal
Date: Sun May 10 19:04:24 2015 +0000
Branches: master
https://developer.blender.org/rB3ec168465d31c90cc75b5ea70de492dbeb4ac992
Cycles: fix compilation on 32-bit Windows for half-floats
Reported by IRC user HG1.
===================================================================
M intern/cycles/util/util_half.h
===================================================================
diff --git a/intern/cycles/util/util_half.h b/intern/cycles/util/util_half.h
index 9642f8e..f4bac98 100644
--- a/intern/cycles/util/util_half.h
+++ b/intern/cycles/util/util_half.h
@@ -54,10 +54,12 @@ ccl_device_inline void float4_store_half(half *h, float4 f,
float scale)
for(int i = 0; i < 4; i++) {
/* optimized float to half for pixels:
* assumes no negative, no nan, no inf, and sets denormal to 0
*/
+ union { uint i; float f; } in;
float fscale = f[i] * scale;
- float x = min(max(fscale, 0.0f), 65504.0f);
+ in.f = (fscale > 0.0f)? ((fscale < 65504.0f)? fscale:
65504.0f): 0.0f;
+ int x = in.i;
- int absolute = __float_as_uint(in) & 0x7FFFFFFF;
+ int absolute = x & 0x7FFFFFFF;
int Z = absolute + 0xC8000000;
int result = (absolute < 0x38800000)? 0: Z;
int rshift = (result >> 13);
_______________________________________________
Bf-blender-cvs mailing list
[email protected]
http://lists.blender.org/mailman/listinfo/bf-blender-cvs
