Attachment 913760 Details for Bug 1115042 – add NO_VPERMPD switch

[patch] add NO_VPERMPD switch

openblas-0.2.9-haswell.patch (text/plain), 2.33 KB, created by Dave Love on 2014-07-01 13:15:53 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Dave Love

Created: 2014-07-01 13:15:53 UTC

Size: 2.33 KB

patch

obsolete

>--- ./Makefile.system~	2014-06-10 14:55:47.000000000 +0100
>+++ ./Makefile.system	2014-06-25 23:15:41.611513846 +0100
>@@ -65,6 +65,10 @@
> GETARCH_FLAGS	+= -DNO_AVX
> endif
> 
>+ifeq ($(NO_VPERMPD), 1)
>+GETARCH_FLAGS	+= -DNO_VPERMPD
>+endif
>+
> ifeq ($(DEBUG), 1)
> GETARCH_FLAGS	+= -g
> endif
>@@ -337,14 +341,20 @@
> DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
> 	       CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
> ifneq ($(NO_AVX), 1)
>-DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER HASWELL
>+DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER
>+ifneq ($(NO_VPERMPD), 1)
>+DYNAMIC_CORE += HASWELL
>+endif
> endif
> endif
> 
> ifeq ($(ARCH), x86_64)
> DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
> ifneq ($(NO_AVX), 1)
>-DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER HASWELL
>+DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER
>+ifneq ($(NO_VPERMPD), 1)
>+DYNAMIC_CORE += HASWELL
>+endif
> endif
> endif
> 
>@@ -718,6 +729,10 @@
> CCOMMON_OPT	+= -DNO_AVX
> endif
> 
>+ifeq ($(NO_VPERMPD), 1)
>+CCOMMON_OPT	+= -DNO_VPERMPD
>+endif
>+
> ifdef SMP
> CCOMMON_OPT	+= -DSMP_SERVER
> 
>--- ./cpuid_x86.c~	2014-06-10 14:55:47.000000000 +0100
>+++ ./cpuid_x86.c	2014-06-25 23:07:36.229587213 +0100
>@@ -49,6 +49,9 @@
> #define CORE_BULLDOZER CORE_BARCELONA
> #define CPUTYPE_PILEDRIVER CPUTYPE_BARCELONA
> #define CORE_PILEDRIVER CORE_BARCELONA
>+#elif NO_VPERMPD		/* binutils too old */
>+#define CPUTYPE_HASWELL CPUTYPE_SANDYBRIDGE
>+#define CORE_HASWELL CORE_SANDYBRIDGE
> #endif
> 
> #ifndef CPUIDEMU
>--- ./Makefile.rule~	2014-06-10 14:55:47.000000000 +0100
>+++ ./Makefile.rule	2014-06-25 23:07:32.171671486 +0100
>@@ -88,6 +88,9 @@
> # and OS. However, the performance is low.
> # NO_AVX = 1
> 
>+# Don't use Haswell optimizations if binutils is too old (e.g. RHEL6)
>+# NO_VPERMPD = 1
>+
> # Don't use parallel make.
> # NO_PARALLEL_MAKE = 1
> 
>--- ./driver/others/dynamic.c~	2014-06-10 14:55:47.000000000 +0100
>+++ ./driver/others/dynamic.c	2014-06-25 23:15:45.424434765 +0100
>@@ -66,7 +66,11 @@
> extern gotoblas_t  gotoblas_SANDYBRIDGE;
> extern gotoblas_t  gotoblas_BULLDOZER;
> extern gotoblas_t  gotoblas_PILEDRIVER;
>+#if NO_VPERMPD
>+#define gotoblas_HASWELL gotoblas_SANDYBRIDGE
>+#else
> extern gotoblas_t  gotoblas_HASWELL;
>+#endif
> #else
> //Use NEHALEM kernels for sandy bridge
> #define gotoblas_SANDYBRIDGE gotoblas_NEHALEM

Actions: View | Diff

Attachments on bug 1115042: 913760 | 913761