working on NSIMD=2/4/8 options

2020-08-08 00:22:32 +02:00 · 2020-08-08 00:22:32 +02:00 · 606cd59108
parent 09d192eed2
commit 606cd59108
5 changed files with 298 additions and 47 deletions
--- a/configure.ac
+++ b/configure.ac
@ -1267,45 +1267,8 @@ if test "x$enable_modsimd" = xyes; then
    fi
 fi

-# do we want to use SLEEF ?
-if test "x$enable_modsimd" = xyes; then
-  if test "x$with_sleef" = xyes; then
-    if test "x$nsimd" = x4; then
-      AC_CHECK_HEADERS([sleef.h])
-      AC_CHECK_LIB([sleef], [Sleef_logd4_u35],
-        [AC_DEFINE([HAVE_LIBSLEEF], [], [Have SLEEF vector math library])
-         LIBS="$LIBS -lsleef"
-	 have_sleef=yes
-	 AC_MSG_RESULT([Use SLEEF])
-	 ], [have_sleef=no]
-	 )
-    fi
-    if test "x$nsimd" = x2; then
-      AC_CHECK_HEADERS([sleef.h])
-      AC_CHECK_LIB([sleef], [Sleef_logd2_u35],
-        [AC_DEFINE([HAVE_LIBSLEEF], [], [Have SLEEF vector math library])
-         LIBS="$LIBS -lsleef"
-	 have_sleef=yes
-	 AC_MSG_RESULT([Use SLEEF])
-	 ], [have_sleef=no]
-	 )
-    fi
-    if test "x$nsimd" = x8; then
-      AC_CHECK_HEADERS([sleef.h])
-      AC_CHECK_LIB([sleef], [Sleef_logd8_u35],
-        [AC_DEFINE([HAVE_LIBSLEEF], [], [Have SLEEF vector math library])
-         LIBS="$LIBS -lsleef"
-	 have_sleef=yes
-	 AC_MSG_RESULT([Use SLEEF])
-	 ], [have_sleef=no]
-	 )
-    fi
-  fi
-fi
-
-# otherwise test if we can use libmvec or equivalent
+# test if we can use libmvec or equivalent
 if test "x$have_intrinsics" = xyes; then
-if test "x$have_sleef" != xyes; then
    AC_MSG_CHECKING([vector math library libmvec])
    have_libmvec=no
    if test "x$nsimd" = x4; then
@ -1350,8 +1313,64 @@ if test "x$have_sleef" != xyes; then
      have_libmvec=no
    ])
    fi
-    
 fi
+
+# do we want to use SLEEF ?
+if test "x$have_libmvec" = xno; then
+if test "x$with_sleef" != xno && test "x$with_sleef" != xyes; then
+   with_sleef=try;
+fi
+fi
+
+if test "x$enable_modsimd" = xyes; then
+  if test "x$with_sleef" = xyes || test "x$with_sleef" = xtry; then
+    if test "x$nsimd" = x4; then
+      AC_CHECK_LIB([sleef], [Sleef_logd4_u35],
+        [AC_DEFINE([HAVE_LIBSLEEF], [], [Have SLEEF vector math library])
+         LIBS="$LIBS -lsleef"
+	 have_sleef=yes
+	 AC_MSG_RESULT([Use SLEEF])
+	 ], [have_sleef=no]
+	 )
+    fi
+    if test "x$nsimd" = x2; then
+      AC_CHECK_HEADERS([sleef.h])
+      AC_CHECK_LIB([sleef], [Sleef_logd2_u35],
+        [AC_DEFINE([HAVE_LIBSLEEF], [], [Have SLEEF vector math library])
+         LIBS="$LIBS -lsleef"
+	 have_sleef=yes
+	 AC_MSG_RESULT([Use SLEEF])
+	 ], [have_sleef=no]
+	 )
+    fi
+    if test "x$nsimd" = x8; then
+      AC_CHECK_HEADERS([sleef.h])
+      AC_CHECK_LIB([sleef], [Sleef_logd8_u35])
+      AC_MSG_CHECKING([vector math library sleef with vector width 8])
+      AC_LINK_IFELSE([
+      AC_LANG_PROGRAM(
+        [[#include <sleef.h>]],
+        [[__m512d x; x=Sleef_logd8_u35(x);]]
+      )] ,[
+        AC_MSG_RESULT([yes])
+        have_sleef=yes
+      ],[
+        AC_MSG_RESULT([no])
+        have_sleef=no
+      ] )
+    fi
+  fi
+fi
+
+if test "x$have_sleef" = xno; then
+if test "x$with_sleef" = xyes; then
+  AC_MSG_ERROR([SLEEF requested but can't be used, try without-sleef])
+fi
+fi
+
+#disable libmvec if specified to use sleef
+if test "x$have_sleef" = xyes; then
+  have_libmvec=no
 fi

 # adjust compiler flags for MODSIMD
@ -1403,8 +1422,13 @@ if test "x$enable_modsimd" = xyes; then
 	AC_MSG_RESULT([Use libmvec])
    fi
    
-    if test "x$have_libmvec" = xno; then
-    if test "x$have_sleef" = xno; then
+    if test "x$have_sleef" = xyes; then
+        AC_DEFINE([USE_LIBSLEEF],[1], [use SLEEF for vector math])
+	AC_MSG_RESULT([Use SLEEF])
+    fi
+    
+    if test "x$have_libmvec" != xyes; then
+    if test "x$have_sleef" != xyes; then
       AC_MSG_WARN([No math vector library detected (sleef or libmvec). Rely on compiler's auto-vectorization of math functions.])
    fi
    fi
--- a/src/include/ngspice/SIMD/simdop2.h
+++ b/src/include/ngspice/SIMD/simdop2.h
@ -0,0 +1,118 @@
+/*******************************************************************************
+ * Copyright 2020 Florian Ballenegger, Anamosic Ballenegger Design
+ *******************************************************************************
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ * this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ * may be used to endorse or promote products derived from this software without
+ * specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+
+#if USEX86INTRINSICS==1
+#include <x86intrin.h>
+#define vec2_MAX(a,b) _mm_max_pd(a,b)
+#define vecN_MAX vec2_MAX
+#define vec2_sqrt(a) _mm_sqrt_pd(a)
+#define vecN_sqrt vec2_sqrt
+static inline Vec2d vec2_blend(Vec2d fa, Vec2d tr, Vec2m mask)
+{
+	return _mm_blendv_pd(fa,tr, (Vec2d) mask);
+}
+#define vecN_blend vec2_blend
+#else
+#define vec2_blend vecN_blend
+#endif
+
+#ifdef USE_LIBSLEEF
+#include <sleef.h>
+#define vec2_exp(a) Sleef_expd2_u10(a)
+#define vecN_exp vec2_exp
+#define vec2_log(a) Sleef_logd2_u35(a)
+#define vecN_log vec2_log
+#ifndef USEX86INTRINSICS
+#define vec2_MAX(a,b) Sleef_fmaxd2(a,b)
+#define vecN_MAX vec2_MAX
+#define vec2_sqrt(a) Sleef_sqrtd2_u35(a)
+#define vecN_sqrt vec2_sqrt
+#endif
+#define vec2_fabs(a) Sleef_fabsd2(a)
+#define vecN_fabs vec2_fabs
+#define vec2_pow(a,b) Sleef_powd2_u10(a,vec2_SIMDTOVECTOR(b))
+#define vecN_pow vec2_pow
+
+#else
+
+#ifdef HAS_LIBMVEC
+Vec2d _ZGVbN2v_exp(Vec2d);
+Vec2d _ZGVbN2v_log(Vec2d);
+Vec2d _ZGVbN2vv_pow(Vec2d, Vec2d);
+
+#define vec2_exp(a) _ZGVbN2v_exp(a)
+#define vecN_exp vec2_exp
+#define vec2_log(a) _ZGVbN2v_log(a)
+#define vecN_log vec2_log
+#define vec2_pow(a,b) _ZGVbN2vv_pow(a,b)
+#define vecN_pow vec2_pow
+#define vec2_fabs vecN_fabs
+
+#endif /* HAS_LIBMVEC */
+#endif /* not USE_LIBSLEEF */
+
+#ifdef USE_SERIAL_FORM
+
+#define vec2_SIMDTOVECTOR vecN_SIMDTOVECTOR
+#define vec2_SIMDTOVECTORMASK vecN_SIMDTOVECTORMASK
+#define vec2_StateAccess vecN_StateAccess
+#define vec2_SIMDCOUNT vecN_SIMDCOUNT
+
+#else
+
+static inline Vec2d vec2_SIMDTOVECTOR(double val)
+{
+	return (Vec2d) {val,val};
+}
+static inline Vec2m vec2_SIMDTOVECTORMASK(int val)
+{
+	return (Vec2m) {val,val};
+}
+static inline Vec2d vec2_StateAccess(double* cktstate, Vec2m stateindexes)
+{
+	return (Vec2d) {
+	 cktstate[stateindexes[0]],
+	 cktstate[stateindexes[1]]
+	};
+}
+static inline int vec2_SIMDCOUNT(Vec2m mask) {
+	return (mask[0] ? 1 : 0) + (mask[1] ? 1 : 0) ;
+}
+#define vecN_SIMDTOVECTOR vec2_SIMDTOVECTOR
+#define vecN_SIMDTOVECTORMASK vec2_SIMDTOVECTORMASK
+#define vecN_StateAccess vec2_StateAccess
+#define vecN_SIMDCOUNT vec2_SIMDCOUNT
+
+#endif
+
+#define vec2_StateStore vecN_StateStore
+#define vec2_StateAdd vecN_StateAdd
+#define vec2_StateSub vecN_StateSub
+
--- a/src/include/ngspice/SIMD/simdop4.h
+++ b/src/include/ngspice/SIMD/simdop4.h
@ -43,7 +43,7 @@ static inline Vec4d vec4_blend(Vec4d fa, Vec4d tr, Vec4m mask)
 #define vec4_blend vecN_blend
 #endif

-#ifdef HAVE_LIBSLEEF
+#ifdef USE_LIBSLEEF
 #include <sleef.h>
 #define vec4_exp(a) Sleef_expd4_u10(a)
 #define vecN_exp vec4_exp
@ -76,7 +76,7 @@ Vec4d _ZGVdN4vv_pow(Vec4d, Vec4d);
 #define vec4_fabs vecN_fabs

 #endif /* HAS_LIBMVEC */
-#endif /* not HAVE_LIBSLEEF */
+#endif /* not USE_LIBSLEEF */

 #ifdef USE_SERIAL_FORM

--- a/src/include/ngspice/SIMD/simdop8.h
+++ b/src/include/ngspice/SIMD/simdop8.h
@ -0,0 +1,109 @@
+/*******************************************************************************
+ * Copyright 2020 Florian Ballenegger, Anamosic Ballenegger Design
+ *******************************************************************************
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ * this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its contributors
+ * may be used to endorse or promote products derived from this software without
+ * specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ ******************************************************************************/
+
+#ifdef USE_LIBSLEEF
+#include <sleef.h>
+#define vec8_exp(a) Sleef_expd8_u10(a)
+#define vecN_exp vec8_exp
+#define vec8_log(a) Sleef_logd8_u35(a)
+#define vecN_log vec8_log
+#ifndef USEX86INTRINSICS
+#define vec8_MAX(a,b) Sleef_fmaxd8(a,b)
+#define vecN_MAX vec8_MAX
+#define vec8_sqrt(a) Sleef_sqrtd8_u35(a)
+#define vecN_sqrt vec8_sqrt
+#endif
+#define vec8_fabs(a) Sleef_fabsd8(a)
+#define vecN_fabs vec8_fabs
+#define vec8_pow(a,b) Sleef_powd8_u10(a,vec8_SIMDTOVECTOR(b))
+#define vecN_pow vec8_pow
+
+#else
+
+#define vec8_exp vecN_exp
+#define vec8_log vecN_log
+#define vec8_MAX vecN_MAX
+#define vec8_sqrt vecN_sqrt
+#define vec8_fabs vecN_fabs
+#define vec8_pow vecN_pow
+
+#endif /* USE_LIBSLEEF */
+
+#ifdef USE_SERIAL_FORM
+
+#define vec8_SIMDTOVECTOR vecN_SIMDTOVECTOR
+#define vec8_SIMDTOVECTORMASK vecN_SIMDTOVECTORMASK
+#define vec8_StateAccess vecN_StateAccess
+#define vec8_SIMDCOUNT vecN_SIMDCOUNT
+
+#else
+
+static inline Vec8d vec8_SIMDTOVECTOR(double val)
+{
+	return (Vec8d) {val,val,val,val,val,val,val,val};
+}
+static inline Vec8m vec8_SIMDTOVECTORMASK(int val)
+{
+	return (Vec8m) {val,val,val,val,val,val,val,val};
+}
+static inline Vec8d vec8_StateAccess(double* cktstate, Vec8m stateindexes)
+{
+	return (Vec8d) {
+	 cktstate[stateindexes[0]],
+	 cktstate[stateindexes[1]],
+	 cktstate[stateindexes[2]],
+	 cktstate[stateindexes[3]],
+	 cktstate[stateindexes[4]],
+	 cktstate[stateindexes[5]],
+	 cktstate[stateindexes[6]],
+	 cktstate[stateindexes[7]],
+	};
+}
+static inline int vec8_SIMDCOUNT(Vec8m mask) {
+	return (   mask[0] ? 1 : 0)
+		+ (mask[1] ? 1 : 0)
+		+ (mask[2] ? 1 : 0)
+		+ (mask[3] ? 1 : 0)
+		+ (mask[4] ? 1 : 0)
+		+ (mask[5] ? 1 : 0)
+		+ (mask[6] ? 1 : 0)
+		+ (mask[7] ? 1 : 0);
+}
+#define vecN_SIMDTOVECTOR vec8_SIMDTOVECTOR
+#define vecN_SIMDTOVECTORMASK vec8_SIMDTOVECTORMASK
+#define vecN_StateAccess vec8_StateAccess
+#define vecN_SIMDCOUNT vec8_SIMDCOUNT
+
+#endif
+
+#define vec8_StateStore vecN_StateStore
+#define vec8_StateAdd vecN_StateAdd
+#define vec8_StateSub vecN_StateSub
+
--- a/src/spicelib/devices/bsim3v32simd/b3v32ldsimd4d.c
+++ b/src/spicelib/devices/bsim3v32simd/b3v32ldsimd4d.c
@ -39,7 +39,7 @@
 #define vec4_powMJSWG(x,p) vec4_pow(x,p)


-#ifdef HAVE_LIBSLEEF
+#ifdef USE_LIBSLEEF
 #include <sleef.h>
 #define vec4_exp(a) Sleef_expd4_u10(a)
 #define vec4_log(a) Sleef_logd4_u35(a)
@ -50,7 +50,7 @@
 #endif


-/* HAS_LIBMVEC and/or HAVE_LIBSLEEF defined from configure.ac */
+/* HAS_LIBMVEC and/or USE_LIBSLEEF defined from configure.ac */

 /* USE_SERIAL_FORM can be defined but has no performance influence */

@ -73,7 +73,7 @@ static inline Vec4d vec4_blend(Vec4d fa, Vec4d tr, Vec4m mask)
 }
 #endif

-#ifndef HAVE_LIBSLEEF
+#ifndef USE_LIBSLEEF
 /******* vec4_exp, vec4_log *******/
 #ifdef HAS_LIBMVEC
 Vec4d _ZGVdN4v_exp(Vec4d x);
@ -161,7 +161,7 @@ static inline Vec4d vec4_pow(Vec4d x, double p)
 	return vec4_exp(vec4_log(x)*p);
 }

-#endif /* HAVE_LIBSLEEF */
+#endif /* USE_LIBSLEEF */

 /******* vec4_SIMDTOVECTOR, vec4_SIMDTOVECTORMASK *******/
 #ifdef USE_SERIAL_FORM