Merge remote branch 'origin/haswell' into develop

Makefile.system

@@ -336,14 +336,14 @@ ifeq ($(ARCH), x86)
 DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
 	       CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
 ifneq ($(NO_AVX), 1)
-DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER
+DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER HASWELL
 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
+DYNAMIC_CORE += SANDYBRIDGE BULLDOZER PILEDRIVER HASWELL
 endif
 endif
 

cpuid.h

@@ -107,7 +107,7 @@
 #define CORE_BOBCAT     21
 #define CORE_BULLDOZER  22
 #define CORE_PILEDRIVER  23
-#define CORE_HASWELL CORE_SANDYBRIDGE
+#define CORE_HASWELL 24
 
 #define HAVE_SSE      (1 <<  0)
 #define HAVE_SSE2     (1 <<  1)
@@ -200,7 +200,6 @@ typedef struct {
 #define CPUTYPE_BOBCAT                  45
 #define CPUTYPE_BULLDOZER               46
 #define CPUTYPE_PILEDRIVER              47
-// this define is because BLAS doesn't have haswell specific optimizations yet
-#define CPUTYPE_HASWELL CPUTYPE_SANDYBRIDGE 
+#define CPUTYPE_HASWELL 48
 
 #endif

cpuid_x86.c

@@ -1243,6 +1243,7 @@ static char *cpuname[] = {
   "BOBCAT",
   "BULLDOZER",
   "PILEDRIVER",
+  "HASWELL",
 };
 
 static char *lowercpuname[] = {
@@ -1293,6 +1294,7 @@ static char *lowercpuname[] = {
   "bobcat",
   "bulldozer",
   "piledriver",
+  "haswell",
 };
 
 static char *corename[] = {
@@ -1320,6 +1322,7 @@ static char *corename[] = {
   "BOBCAT",
   "BULLDOZER",
   "PILEDRIVER",
+  "HASWELL",
 };
 
 static char *corename_lower[] = {
@@ -1347,6 +1350,7 @@ static char *corename_lower[] = {
   "bobcat",
   "bulldozer",
   "piledriver",
+  "haswell",
 };
 
 

driver/others/dynamic.c

@@ -65,14 +65,15 @@ extern gotoblas_t  gotoblas_BOBCAT;
 extern gotoblas_t  gotoblas_SANDYBRIDGE;
 extern gotoblas_t  gotoblas_BULLDOZER;
 extern gotoblas_t  gotoblas_PILEDRIVER;
+extern gotoblas_t  gotoblas_HASWELL;
 #else
 //Use NEHALEM kernels for sandy bridge
 #define gotoblas_SANDYBRIDGE gotoblas_NEHALEM
+#define gotoblas_HASWELL gotoblas_NEHALEM
 #define gotoblas_BULLDOZER gotoblas_BARCELONA
 #define gotoblas_PILEDRIVER gotoblas_BARCELONA
 #endif
-//Use sandy bridge kernels for haswell.
-#define gotoblas_HASWELL gotoblas_SANDYBRIDGE
+
 
 #define VENDOR_INTEL      1
 #define VENDOR_AMD        2
@@ -297,6 +298,7 @@ static char *corename[] = {
     "Bobcat",
     "Bulldozer",
     "Piledriver",
+    "Haswell",
 };
 
 char *gotoblas_corename(void) {
@@ -319,7 +321,8 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_SANDYBRIDGE)  return corename[16];
   if (gotoblas == &gotoblas_BOBCAT)       return corename[17];
   if (gotoblas == &gotoblas_BULLDOZER)    return corename[18];
-  if (gotoblas == &gotoblas_PILEDRIVER)    return corename[19];
+  if (gotoblas == &gotoblas_PILEDRIVER)   return corename[19];
+  if (gotoblas == &gotoblas_HASWELL)      return corename[20];
 
   return corename[0];
 }

getarch.c

@@ -298,6 +298,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CORENAME  "SANDYBRIDGE"
 #endif
 
+#ifdef FORCE_HASWELL
+#define FORCE
+#define FORCE_INTEL
+#define ARCHITECTURE    "X86"
+#define SUBARCHITECTURE "HASWELL"
+#define ARCHCONFIG   "-DHASWELL " \
+		     "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
+		     "-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
+		     "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
+		     "-DHAVE_CMOV -DHAVE_MMX -DHAVE_SSE -DHAVE_SSE2 -DHAVE_SSE3 -DHAVE_SSSE3 -DHAVE_SSE4_1 -DHAVE_SSE4_2 -DHAVE_AVX " \
+                     "-DFMA3"
+#define LIBNAME   "haswell"
+#define CORENAME  "HASWELL"
+#endif
+
 #ifdef FORCE_ATOM
 #define FORCE
 #define FORCE_INTEL

kernel/x86/KERNEL.HASWELL

@@ -0,0 +1 @@
+include $(KERNELDIR)/KERNEL.PENRYN

kernel/x86/trsm_kernel_LN_2x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_LN_4x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_LT_2x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_LT_4x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_RT_2x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/trsm_kernel_RT_4x4_penryn.S

@@ -62,7 +62,7 @@
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH     prefetcht0
 #define PREFETCHSIZE  (8 * 21 + 4)
 #endif

kernel/x86/ztrsm_kernel_LN_2x2_penryn.S

@@ -61,7 +61,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_LT_1x2_penryn.S

@@ -63,7 +63,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_LT_2x2_penryn.S

@@ -61,7 +61,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_RT_1x2_penryn.S

@@ -63,7 +63,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86/ztrsm_kernel_RT_2x2_penryn.S

@@ -61,7 +61,7 @@
 #define PREFETCHSIZE 84
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht1
 #define PREFETCHSIZE 84
 #endif

kernel/x86_64/KERNEL.HASWELL

@@ -0,0 +1,63 @@
+SGEMMKERNEL    =  sgemm_kernel_16x4_haswell.S
+SGEMMINCOPY    =  ../generic/gemm_ncopy_16.c
+SGEMMITCOPY    =  ../generic/gemm_tcopy_16.c
+SGEMMONCOPY    =  ../generic/gemm_ncopy_4.c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+DGEMMKERNEL    =  dgemm_kernel_4x4_haswell.S
+DGEMMINCOPY    =  
+DGEMMITCOPY    =  
+DGEMMONCOPY    =  ../generic/gemm_ncopy_4.c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+DGEMMINCOPYOBJ =  
+DGEMMITCOPYOBJ =  
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CGEMMKERNEL    =  cgemm_kernel_8x2_haswell.S
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_8.c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_8.c
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+ZGEMMKERNEL    =  zgemm_kernel_4x2_haswell.S
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_4.c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_4.c
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CGEMM3MKERNEL    =  zgemm3m_kernel_4x8_nehalem.S
+ZGEMM3MKERNEL    =  zgemm3m_kernel_2x8_nehalem.S
+

kernel/x86_64/symv_L_sse.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 12)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 12)

kernel/x86_64/symv_L_sse2.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 12)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 12)

kernel/x86_64/symv_U_sse.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 12)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 12)

kernel/x86_64/symv_U_sse2.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 12)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)

kernel/x86_64/zsymv_L_sse.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 24)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)

kernel/x86_64/zsymv_L_sse2.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 24)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)

kernel/x86_64/zsymv_U_sse.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 24)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)

kernel/x86_64/zsymv_U_sse2.S

@@ -57,7 +57,7 @@
 #define PREFETCHSIZE	(16 * 24)
 #endif
 
-#if defined(NEHALEM) || defined(SANDYBRIDGE)
+#if defined(NEHALEM) || defined(NEHALEM_OPTIMIZATION)
 #define PREFETCH	prefetcht0
 #define PREFETCHW	prefetcht0
 #define PREFETCHSIZE	(16 * 24)

param.h

@@ -1154,6 +1154,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #ifdef HASWELL
 
+<<<<<<< HEAD
 #define SNUMOPT         8
 #define DNUMOPT         4
 
@@ -1164,6 +1165,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SYMV_P  8
 
 #define SWITCH_RATIO    4
+=======
+#define SNUMOPT		8
+#define DNUMOPT		4
+
+#define GEMM_DEFAULT_OFFSET_A	  0
+#define GEMM_DEFAULT_OFFSET_B     0
+#define GEMM_DEFAULT_ALIGN 0x03fffUL
+
+#define SYMV_P	8
+
+#define SWITCH_RATIO	4
+>>>>>>> origin/haswell
 
 #ifdef ARCH_X86
 
@@ -1233,6 +1246,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ZGEMM_DEFAULT_Q 128
 
 #define SGEMM_DEFAULT_R sgemm_r
+<<<<<<< HEAD
+=======
+//#define DGEMM_DEFAULT_R dgemm_r
+>>>>>>> origin/haswell
 #define DGEMM_DEFAULT_R 13824
 #define CGEMM_DEFAULT_R cgemm_r
 #define ZGEMM_DEFAULT_R zgemm_r