From: José Fonseca <jfonseca@vmware.com>
Date: Tue, 15 Dec 2009 12:13:43 +0000 (+0000)
Subject: llvmpipe: Fix bad SI -> FP conversion into lp_build_log2_approx.
X-Git-Tag: mesa-7.8~2111
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=846e38f1c33c3b2e46227886da57beda27b82f0c;p=platform%2Fupstream%2Fmesa.git

llvmpipe: Fix bad SI -> FP conversion into lp_build_log2_approx.

It should be a bitcast  as the integer value is actually an encoded FP
already.
---

diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.c b/src/gallium/drivers/llvmpipe/lp_bld_arit.c
index 9c59677..4fd459e 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_arit.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.c
@@ -1285,7 +1285,7 @@ lp_build_log2_approx(struct lp_build_context *bld,
       /* mant = (float) mantissa(x) */
       mant = LLVMBuildAnd(bld->builder, i, mantmask, "");
       mant = LLVMBuildOr(bld->builder, mant, one, "");
-      mant = LLVMBuildSIToFP(bld->builder, mant, vec_type, "");
+      mant = LLVMBuildBitCast(bld->builder, mant, vec_type, "");
 
       logmant = lp_build_polynomial(bld, mant, lp_build_log2_polynomial,
                                     Elements(lp_build_log2_polynomial));