On x86 32bits, gcc 4.4.3 generated wrong codes (movsd) from movlps in zdot_sse2.S...

This would casue zdotu & zdotc failures. Instead, use movlpd to walk around it. Fixed #8. Fixed #9.

diff --git a/Changelog.txt b/Changelog.txt
index b3c4384..463e03d 100644 (file)
--- a/Changelog.txt
+++ b/Changelog.txt
@@ -13,6 +13,9 @@ common:
        * Imported GotoBLAS2 1.13 BSD version
 
 x86/x86 64:
+       * On x86 32bits, gcc 4.4.3 generated wrong codes (movsd) from movlps 
+         in zdot_sse2.S line 191. This would casue zdotu & zdotc failures.
+         Instead,Walk around it. (Refs issue #8 #9 on github)
        * Modified ?axpy functions to return same netlib BLAS results 
          when incx==0 or incy==0 (Refs issue #7 on github)
        * Modified ?swap functions to return same netlib BLAS results 

diff --git a/kernel/x86/zdot_sse2.S b/kernel/x86/zdot_sse2.S
index 6304f01..0f7acff 100644 (file)
--- a/kernel/x86/zdot_sse2.S
+++ b/kernel/x86/zdot_sse2.S
@@ -1188,8 +1188,8 @@
        testl   $1, N
        jle     .L48
 
-       movlps  -16 * SIZE(X), %xmm4
-       movlps  -16 * SIZE(Y), %xmm6
+       movlpd  -16 * SIZE(X), %xmm4
+       movlpd  -16 * SIZE(Y), %xmm6
 
        pshufd  $0x4e,  %xmm6, %xmm3 
        mulpd   %xmm4,  %xmm6