Add IBM System/390 support

.travis.yml

@@ -67,6 +67,15 @@ matrix:
       env:
         - LABEL=common
         - OMP_WAIT_POLICY=passive CTEST_OUTPUT_ON_FAILURE=TRUE
+    - os: linux
+      dist: bionic
+      arch: s390x
+      addons:
+        apt:
+          packages:
+            - g++-8 libmpfr-dev libssl-dev libfftw3-dev
+      env:
+        - LABEL="s390x-gcc"
 
 before_install:
      - export PATH=$PATH:/usr/bin

Configure.cmake

@@ -54,6 +54,7 @@ set(SLEEF_SUPPORTED_EXTENSIONS
   ADVSIMD ADVSIMDNOFMA SVE SVENOFMA                     # Aarch64
   NEON32 NEON32VFPV4                                    # Aarch32
   VSX VSXNOFMA                                          # PPC64
+  ZVECTOR ZVECTORNOFMA ZVECTOR2 ZVECTOR2NOFMA		                # IBM Z
   PUREC_SCALAR PURECFMA_SCALAR                          # Generic type
   CACHE STRING "List of SIMD architectures supported by libsleef."
   )
@@ -197,6 +198,31 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(powerpc|ppc)64")
   set(TESTER3_DEFINITIONS_VSX      ATR=finz_ DPTYPE=__vector_double SPTYPE=__vector_float DPTYPESPEC=d2 SPTYPESPEC=f4 EXTSPEC=vsx)
   set(TESTER3_DEFINITIONS_VSXNOFMA ATR=cinz_ DPTYPE=__vector_double SPTYPE=__vector_float DPTYPESPEC=d2 SPTYPESPEC=f4 EXTSPEC=vsxnofma)
 
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "s390x")
+  set(SLEEF_ARCH_S390X ON CACHE INTERNAL "True for IBM Z architecture.")
+
+  set(SLEEF_HEADER_LIST
+    ZVECTOR_
+    ZVECTOR
+    ZVECTORNOFMA
+    ZVECTOR2
+    ZVECTOR2NOFMA
+    PUREC_SCALAR
+    PURECFMA_SCALAR
+  )
+
+  set(HEADER_PARAMS_ZVECTOR_      finz_ 2 4 "SLEEF_VECTOR_DOUBLE" "SLEEF_VECTOR_FLOAT" "SLEEF_VECTOR_INT" "SLEEF_VECTOR_INT" __VEC__)
+  set(HEADER_PARAMS_ZVECTOR       finz_ 2 4 "SLEEF_VECTOR_DOUBLE" "SLEEF_VECTOR_FLOAT" "SLEEF_VECTOR_INT" "SLEEF_VECTOR_INT" __VEC__ zvector)
+  set(HEADER_PARAMS_ZVECTORNOFMA  cinz_ 2 4 "SLEEF_VECTOR_DOUBLE" "SLEEF_VECTOR_FLOAT" "SLEEF_VECTOR_INT" "SLEEF_VECTOR_INT" __VEC__ zvectornofma)
+  set(HEADER_PARAMS_ZVECTOR2      finz_ 2 4 "SLEEF_VECTOR_DOUBLE" "SLEEF_VECTOR_FLOAT" "SLEEF_VECTOR_INT" "SLEEF_VECTOR_INT" __VEC__ zvector2)
+  set(HEADER_PARAMS_ZVECTOR2NOFMA cinz_ 2 4 "SLEEF_VECTOR_DOUBLE" "SLEEF_VECTOR_FLOAT" "SLEEF_VECTOR_INT" "SLEEF_VECTOR_INT" __VEC__ zvector2nofma)
+
+  set(CLANG_FLAGS_ENABLE_PURECFMA_SCALAR "-march=z13")
+
+  set(TESTER3_DEFINITIONS_ZVECTOR      ATR=finz_ DPTYPE=SLEEF_VECTOR_DOUBLE SPTYPE=SLEEF_VECTOR_FLOAT DPTYPESPEC=d2 SPTYPESPEC=f4 EXTSPEC=zvector)
+  set(TESTER3_DEFINITIONS_ZVECTORNOFMA ATR=cinz_ DPTYPE=SLEEF_VECTOR_DOUBLE SPTYPE=SLEEF_VECTOR_FLOAT DPTYPESPEC=d2 SPTYPESPEC=f4 EXTSPEC=zvectornofma)
+  set(TESTER3_DEFINITIONS_ZVECTOR2      ATR=finz_ DPTYPE=SLEEF_VECTOR_DOUBLE SPTYPE=SLEEF_VECTOR_FLOAT DPTYPESPEC=d2 SPTYPESPEC=f4 EXTSPEC=zvector2)
+  set(TESTER3_DEFINITIONS_ZVECTOR2NOFMA ATR=cinz_ DPTYPE=SLEEF_VECTOR_DOUBLE SPTYPE=SLEEF_VECTOR_FLOAT DPTYPESPEC=d2 SPTYPESPEC=f4 EXTSPEC=zvector2nofma)
 endif()
 
 command_arguments(HEADER_PARAMS_PUREC_SCALAR    cinz_ 1 1 double float int32_t int32_t __STDC__ purec)
@@ -226,6 +252,10 @@ command_arguments(RENAME_PARAMS_NEON32          cinz_ 2 4 neon)
 command_arguments(RENAME_PARAMS_NEON32VFPV4     finz_ 2 4 neonvfpv4)
 command_arguments(RENAME_PARAMS_VSX             finz_ 2 4 vsx)
 command_arguments(RENAME_PARAMS_VSXNOFMA        cinz_ 2 4 vsxnofma)
+command_arguments(RENAME_PARAMS_ZVECTOR             finz_ 2 4 zvector)
+command_arguments(RENAME_PARAMS_ZVECTORNOFMA        cinz_ 2 4 zvectornofma)
+command_arguments(RENAME_PARAMS_ZVECTOR2             finz_ 2 4 zvector2)
+command_arguments(RENAME_PARAMS_ZVECTOR2NOFMA        cinz_ 2 4 zvector2nofma)
 command_arguments(RENAME_PARAMS_PUREC_SCALAR    cinz_ 1 1 purec)
 command_arguments(RENAME_PARAMS_PURECFMA_SCALAR finz_ 1 1 purecfma)
 # The vector length parameters in SVE, for SP and DP, are chosen for
@@ -282,6 +312,11 @@ set(CLANG_FLAGS_ENABLE_SVENOFMA "-march=armv8-a+sve")
 # PPC64
 set(CLANG_FLAGS_ENABLE_VSX "-mcpu=power8")
 set(CLANG_FLAGS_ENABLE_VSXNOFMA "-mcpu=power8")
+# IBM z
+set(CLANG_FLAGS_ENABLE_ZVECTOR "-march=z13;-mzvector")
+set(CLANG_FLAGS_ENABLE_ZVECTORNOFMA "-march=z13;-mzvector")
+set(CLANG_FLAGS_ENABLE_ZVECTOR2 "-march=z14;-mzvector")
+set(CLANG_FLAGS_ENABLE_ZVECTOR2NOFMA "-march=z14;-mzvector")
 
 # All variables storing compiler flags should be prefixed with FLAGS_
 if(CMAKE_C_COMPILER_ID MATCHES "(GNU|Clang)")
@@ -407,7 +442,7 @@ option(DISABLE_SSE2 "Disable SSE2" OFF)
 option(ENFORCE_SSE2 "Build fails if SSE2 is not supported by the compiler" OFF)
 
 if(SLEEF_ARCH_X86 AND NOT DISABLE_SSE2)
-  set (CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_SSE2})
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_SSE2})
   CHECK_C_SOURCE_COMPILES("
   #if defined(_MSC_VER)
   #include <intrin.h>
@@ -429,7 +464,7 @@ option(DISABLE_SSE4 "Disable SSE4" OFF)
 option(ENFORCE_SSE4 "Build fails if SSE4 is not supported by the compiler" OFF)
 
 if(SLEEF_ARCH_X86 AND NOT DISABLE_SSE4)
-  set (CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_SSE4})
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_SSE4})
   CHECK_C_SOURCE_COMPILES("
   #if defined(_MSC_VER)
   #include <intrin.h>
@@ -451,7 +486,7 @@ option(ENFORCE_AVX "Disable AVX" OFF)
 option(ENFORCE_AVX "Build fails if AVX is not supported by the compiler" OFF)
 
 if(SLEEF_ARCH_X86 AND NOT DISABLE_AVX)
-  set (CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_AVX})
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_AVX})
   CHECK_C_SOURCE_COMPILES("
   #if defined(_MSC_VER)
   #include <intrin.h>
@@ -473,7 +508,7 @@ option(DISABLE_FMA4 "Disable FMA4" OFF)
 option(ENFORCE_FMA4 "Build fails if FMA4 is not supported by the compiler" OFF)
 
 if(SLEEF_ARCH_X86 AND NOT DISABLE_FMA4)
-  set (CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_FMA4})
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_FMA4})
   CHECK_C_SOURCE_COMPILES("
   #if defined(_MSC_VER)
   #include <intrin.h>
@@ -495,7 +530,7 @@ option(DISABLE_AVX2 "Disable AVX2" OFF)
 option(ENFORCE_AVX2 "Build fails if AVX2 is not supported by the compiler" OFF)
 
 if(SLEEF_ARCH_X86 AND NOT DISABLE_AVX2)
-  set (CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_AVX2})
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_AVX2})
   CHECK_C_SOURCE_COMPILES("
   #if defined(_MSC_VER)
   #include <intrin.h>
@@ -522,7 +557,7 @@ option(DISABLE_AVX512F "Disable AVX512F" OFF)
 option(ENFORCE_AVX512F "Build fails if AVX512F is not supported by the compiler" OFF)
 
 if(SLEEF_ARCH_X86 AND NOT DISABLE_AVX512F)
-  set (CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_AVX512F})
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_AVX512F})
   CHECK_C_SOURCE_COMPILES("
   #if defined(_MSC_VER)
   #include <intrin.h>
@@ -554,7 +589,7 @@ option(DISABLE_SVE "Disable SVE" OFF)
 option(ENFORCE_SVE "Build fails if SVE is not supported by the compiler" OFF)
 
 if(SLEEF_ARCH_AARCH64 AND NOT DISABLE_SVE)
-  set (CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_SVE})
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_SVE})
   CHECK_C_SOURCE_COMPILES("
   #include <arm_sve.h>
   int main() {
@@ -576,7 +611,7 @@ option(DISABLE_VSX "Disable VSX" OFF)
 option(ENFORCE_VSX "Build fails if VSX is not supported by the compiler" OFF)
 
 if(SLEEF_ARCH_PPC64 AND NOT DISABLE_VSX)
-  set (CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_VSX})
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS ${FLAGS_ENABLE_VSX})
   CHECK_C_SOURCE_COMPILES("
   #include <altivec.h>
   #ifndef __LITTLE_ENDIAN__
@@ -600,6 +635,44 @@ if (ENFORCE_VSX AND NOT COMPILER_SUPPORTS_VSX)
   message(FATAL_ERROR "ENFORCE_VSX is specified and that feature is disabled or not supported by the compiler")
 endif()
 
+# IBM Z
+
+option(DISABLE_ZVECTOR "Disable ZVECTOR" OFF)
+option(ENFORCE_ZVECTOR "Build fails if ZVECTOR is not supported by the compiler" OFF)
+
+if(SLEEF_ARCH_S390X AND NOT DISABLE_ZVECTOR)
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS "${FLAGS_ENABLE_ZVECTOR}")
+  CHECK_C_SOURCE_COMPILES("
+  #include <vecintrin.h>
+  int main() {
+    __vector double d;
+    d = __builtin_s390_vfidb(d, 4, 4);
+  }"
+    COMPILER_SUPPORTS_ZVECTOR)
+endif()
+
+if (ENFORCE_ZVECTOR AND NOT COMPILER_SUPPORTS_ZVECTOR)
+  message(FATAL_ERROR "ENFORCE_ZVECTOR is specified and that feature is disabled or not supported by the compiler")
+endif()
+
+option(DISABLE_ZVECTOR2 "Disable ZVECTOR2" OFF)
+option(ENFORCE_ZVECTOR2 "Build fails if ZVECTOR2 is not supported by the compiler" OFF)
+
+if(SLEEF_ARCH_S390X AND NOT DISABLE_ZVECTOR2)
+  string (REPLACE ";" " " CMAKE_REQUIRED_FLAGS "${FLAGS_ENABLE_ZVECTOR2}")
+  CHECK_C_SOURCE_COMPILES("
+  #include <vecintrin.h>
+  int main() {
+    __vector float d;
+    d = vec_sqrt(d);
+  }"
+    COMPILER_SUPPORTS_ZVECTOR2)
+endif()
+
+if (ENFORCE_ZVECTOR2 AND NOT COMPILER_SUPPORTS_ZVECTOR2)
+  message(FATAL_ERROR "ENFORCE_ZVECTOR2 is specified and that feature is disabled or not supported by the compiler")
+endif()
+
 # OpenMP
 
 option(DISABLE_OPENMP "Disable OPENMP" OFF)

src/arch/helperpurec_scalar.h

@@ -43,7 +43,7 @@
 #define ENABLE_FMA_DP
 #define ENABLE_FMA_SP
 
-#if defined(__AVX2__) || defined(__aarch64__) || defined(__arm__) || defined(__powerpc64__)
+#if defined(__AVX2__) || defined(__aarch64__) || defined(__arm__) || defined(__powerpc64__) || defined(__zarch__)
 #ifndef FP_FAST_FMA
 #define FP_FAST_FMA
 #endif
@@ -290,16 +290,21 @@ static INLINE vfloat vtruncate_vf_vf(vfloat vd) { return vcast_vf_vi2(vtruncate_
 static INLINE vfloat vcast_vf_f(float f) { return f; }
 static INLINE vmask vreinterpret_vm_vf(vfloat vf) { union { vfloat vf; vmask vm; } cnv; cnv.vf = vf; return cnv.vm; }
 static INLINE vfloat vreinterpret_vf_vm(vmask vm) { union { vfloat vf; vmask vm; } cnv; cnv.vm = vm; return cnv.vf; }
+#if defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+static INLINE vfloat vreinterpret_vf_vi2(vint2 vi) { union { vfloat vf[2]; vint2 vi2; } cnv; cnv.vi2 = vi; return cnv.vf[1]; }
+static INLINE vint2 vreinterpret_vi2_vf(vfloat vf) { union { vfloat vf[2]; vint2 vi2; } cnv; cnv.vi2 = 0; cnv.vf[1] = vf; return cnv.vi2; }
+#else
 static INLINE vfloat vreinterpret_vf_vi2(vint2 vi) { union { vfloat vf; vint2 vi2; } cnv; cnv.vi2 = vi; return cnv.vf; }
 static INLINE vint2 vreinterpret_vi2_vf(vfloat vf) { union { vfloat vf; vint2 vi2; } cnv; cnv.vi2 = 0; cnv.vf = vf; return cnv.vi2; }
+#endif
 
 static INLINE vfloat vadd_vf_vf_vf(vfloat x, vfloat y) { return x + y; }
 static INLINE vfloat vsub_vf_vf_vf(vfloat x, vfloat y) { return x - y; }
 static INLINE vfloat vmul_vf_vf_vf(vfloat x, vfloat y) { return x * y; }
 static INLINE vfloat vdiv_vf_vf_vf(vfloat x, vfloat y) { return x / y; }
 static INLINE vfloat vrec_vf_vf   (vfloat x)           { return 1 / x; }
 
-static INLINE vfloat vabs_vf_vf(vfloat x) { versatileVector v = { .f = x }; v.x &= 0x7fffffff; return v.f; }
+static INLINE vfloat vabs_vf_vf(vfloat x) { versatileVector v = { .f = x }; v.i[0] &= 0x7fffffff; return v.f; }
 static INLINE vfloat vneg_vf_vf(vfloat x) { return -x; }
 
 static INLINE vfloat vmax_vf_vf_vf(vfloat x, vfloat y) { return x > y ? x : y; }
@@ -329,7 +334,7 @@ static INLINE vopmask vge_vo_vf_vf(vfloat x, vfloat y)  { return x >= y ? ~(uint
 
 static INLINE vint2 vadd_vi2_vi2_vi2(vint2 x, vint2 y) { versatileVector v = { .i2 = x }, w = { .i2 = y }; v.i[0] += w.i[0]; v.i[1] += w.i[1]; return v.i2; }
 static INLINE vint2 vsub_vi2_vi2_vi2(vint2 x, vint2 y) { versatileVector v = { .i2 = x }, w = { .i2 = y }; v.i[0] -= w.i[0]; v.i[1] -= w.i[1]; return v.i2; }
-static INLINE vint2 vneg_vi2_vi2(vint2 x)              { versatileVector v = { .i2 = x }; v.i[0] = -v.i[0]; return v.i2; }
+static INLINE vint2 vneg_vi2_vi2(vint2 x)              { versatileVector v = { .i2 = x }; v.i[0] = -v.i[0]; v.i[1] = -v.i[1]; return v.i2; }
 
 static INLINE vint2 vand_vi2_vi2_vi2(vint2 x, vint2 y)    { return x & y; }
 static INLINE vint2 vandnot_vi2_vi2_vi2(vint2 x, vint2 y) { return y & ~x; }