From eb76eebf177b170ff1fa139318fa9e139d8a6e50 Mon Sep 17 00:00:00 2001
From: Florian Ballenegger <florian.ballenegger@anamosic.com>
Date: Wed, 29 Jul 2020 23:15:44 +0200
Subject: [PATCH] use libmvec only if AVX is available

---
 configure.ac                                  |  3 +-
 .../devices/bsim3v32simd/b3v32ldsimd4d.c      |  3 +-
 .../devices/bsim3v32simd/b3v32ldsimd8d.c      | 46 +++++++++++++++----
 3 files changed, 40 insertions(+), 12 deletions(-)

diff --git a/configure.ac b/configure.ac
index 7d6a68b70..b1a4bfe1e 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1226,10 +1226,11 @@ if test "x$enable_modsimd" = xyes; then
       AC_MSG_RESULT([no])
       have_intrinsics=no
     ])
+    have_libmvec=no
 fi
 
 # test if we can use libmvec or equivalent
-if test "x$enable_modsimd" = xyes; then
+if test "x$have_intrinsics" = xyes; then
     AC_MSG_CHECKING([vector math library])
     AC_TRY_COMPILE([
     	#include <math.h>
diff --git a/src/spicelib/devices/bsim3v32simd/b3v32ldsimd4d.c b/src/spicelib/devices/bsim3v32simd/b3v32ldsimd4d.c
index 44325f110..13b340aca 100644
--- a/src/spicelib/devices/bsim3v32simd/b3v32ldsimd4d.c
+++ b/src/spicelib/devices/bsim3v32simd/b3v32ldsimd4d.c
@@ -329,7 +329,7 @@ static inline int vec4_SIMDCOUNT(Vec4m mask) {
 }
 
 
-
+#if 0
 /* some debug utils functions */
 void vec4_printd(const char* msg, const char* name, Vec4d vecd)
 {
@@ -352,3 +352,4 @@ void vec4_CheckCollisions(Vec4m stateindexes, const char* msg)
 		raise(SIGINT);
 	}
 }
+#endif
diff --git a/src/spicelib/devices/bsim3v32simd/b3v32ldsimd8d.c b/src/spicelib/devices/bsim3v32simd/b3v32ldsimd8d.c
index 4c8fec468..f0f706907 100644
--- a/src/spicelib/devices/bsim3v32simd/b3v32ldsimd8d.c
+++ b/src/spicelib/devices/bsim3v32simd/b3v32ldsimd8d.c
@@ -28,10 +28,18 @@
  * POSSIBILITY OF SUCH DAMAGE.
  ******************************************************************************/
 
+/* disable omp simd for GCC, as it slow down a bit */
+#if !defined(__GNUC__) || defined(__clang__) || defined(__INTEL_COMPILER)
+#define USE_OMPSIMD
+#endif
+
+
 static inline Vec8d vec8_blend(Vec8d fa, Vec8d tr, Vec8m mask)
 {
 	Vec8d r;
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int i=0;i<8;i++)
 		r[i] = (mask[i]==0 ? fa[i] : tr[i]);
 	return r;
@@ -40,7 +48,9 @@ static inline Vec8d vec8_blend(Vec8d fa, Vec8d tr, Vec8m mask)
 static inline Vec8d vec8_exp(Vec8d x)
 {
 	Vec8d r;
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int i=0;i<8;i++)
 		r[i] = exp(x[i]);
 	return r;
@@ -49,7 +59,9 @@ static inline Vec8d vec8_exp(Vec8d x)
 static inline Vec8d vec8_log(Vec8d x)
 {
 	Vec8d r;
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int i=0;i<8;i++)
 		r[i] = log(x[i]);
 	return r;
@@ -58,7 +70,9 @@ static inline Vec8d vec8_log(Vec8d x)
 static inline Vec8d vec8_max(Vec8d x, Vec8d y)
 {
 	Vec8d r;
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int i=0;i<8;i++)
 		r[i] = MAX(x[i],y[i]);
 	return r;
@@ -67,7 +81,9 @@ static inline Vec8d vec8_max(Vec8d x, Vec8d y)
 static inline Vec8d vec8_sqrt(Vec8d x)
 {
 	Vec8d r;
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int i=0;i<8;i++)
 		r[i] = sqrt(x[i]);
 	return r;
@@ -76,7 +92,9 @@ static inline Vec8d vec8_sqrt(Vec8d x)
 static inline Vec8d vec8_fabs(Vec8d x)
 {
 	Vec8d r;
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int i=0;i<8;i++)
 		r[i] = fabs(x[i]);
 	return r;
@@ -111,7 +129,9 @@ static inline Vec8d vec8_SIMDLOADDATA(int idx, double data[7][8])
 static inline Vec8d vec8_BSIM3v32_StateAccess(double* cktstate, Vec8m stateindexes)
 {
 	Vec8d r;
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int i=0;i<8;i++)
 		r[i] =  cktstate[stateindexes[i]];
 	return r;
@@ -120,7 +140,9 @@ static inline Vec8d vec8_BSIM3v32_StateAccess(double* cktstate, Vec8m stateindex
 
 static inline void vec8_BSIM3v32_StateStore(double* cktstate, Vec8m stateindexes, Vec8d values)
 {
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int idx=0;idx<8;idx++)
 	{
 		cktstate[stateindexes[idx]] = values[idx];
@@ -129,7 +151,9 @@ static inline void vec8_BSIM3v32_StateStore(double* cktstate, Vec8m stateindexes
 
 static inline void vec8_BSIM3v32_StateAdd(double* cktstate, Vec8m stateindexes, Vec8d values)
 {
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int idx=0;idx<8;idx++)
 	{
 		cktstate[stateindexes[idx]] += values[idx];
@@ -138,7 +162,9 @@ static inline void vec8_BSIM3v32_StateAdd(double* cktstate, Vec8m stateindexes,
 
 static inline void vec8_BSIM3v32_StateSub(double* cktstate, Vec8m stateindexes, Vec8d values)
 {
-	#pragma omp simd
+	#ifdef USE_OMPSIMD
+	#pragma omp simd simdlen(4)
+	#endif
 	for(int idx=0;idx<8;idx++)
 	{
 		cktstate[stateindexes[idx]] -= values[idx];