[CUDA] Include purec header from CUDA (shibatch#394)

With this patch, the pure C inline header can be included from CUDA programs along with the CUDA inline header.
blapie · Jan 17, 2021 · bf50d61 · bf50d61
1 parent 295a8b6
commit bf50d61
Show file tree

Hide file tree

Showing 27 changed files with 238 additions and 195 deletions.
diff --git a/Configure.cmake b/Configure.cmake
@@ -740,11 +740,13 @@ if(SLEEF_SHOW_ERROR_LOG)
   if (EXISTS ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeError.log)
     file(READ ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeError.log FILE_CONTENT)
     message("")
-    message("=====  Content of CMakeError.log  =====")
+    message("")
+    message("======  Content of CMakeError.log  ======")
     message("")
     message("${FILE_CONTENT}")
     message("")
-    message("=======================================")
+    message("========  End of CMakeError.log  ========")
+    message("")
     message("")
   endif()
 endif(SLEEF_SHOW_ERROR_LOG)

diff --git a/appveyor.yml b/appveyor.yml
@@ -22,7 +22,9 @@ build_script:
   - if "%DO_TEST%" == "TRUE" call p.bat
   - if "%DO_TEST%" == "TRUE" "C:\\Cygwin64\\bin\\bash" -c 'mkdir build-mingw;cd build-mingw;CC=x86_64-w64-mingw32-gcc cmake -G Ninja .. -DRUNNING_ON_APPVEYOR=TRUE -DBUILD_SHARED_LIBS=FALSE -DBUILD_QUAD=TRUE -DBUILD_DFT=TRUE -DENFORCE_SSE2=TRUE -DENFORCE_SSE4=TRUE;ninja'
   - if "%DO_TEST%" == "TRUE" cd "%BUILDFOLDER%"
-  - if "%DO_TEST%" == "TRUE" "C:\\Cygwin64\\bin\\bash" -c 'mkdir build-cygwin;cd build-cygwin;cmake -G Ninja -DRUNNING_ON_APPVEYOR=TRUE -DBUILD_QUAD=TRUE -DBUILD_DFT=TRUE -DBUILD_INLINE_HEADERS=TRUE -DENFORCE_SSE2=TRUE -DENFORCE_SSE2=TRUE -DENFORCE_SSE4=TRUE ..;ninja'
+  - if "%DO_TEST%" == "TRUE" "C:\\Cygwin64\\bin\\bash" -c 'mkdir build-cygwin;cd build-cygwin;cmake -G Ninja -DRUNNING_ON_APPVEYOR=TRUE -DBUILD_QUAD=TRUE -DBUILD_DFT=FALSE -DBUILD_INLINE_HEADERS=TRUE -DENFORCE_SSE2=TRUE -DENFORCE_SSE2=TRUE -DENFORCE_SSE4=TRUE ..;ninja'
+  - if "%DO_TEST%" == "TRUE" cd "%BUILDFOLDER%"
+  - if "%DO_TEST%" == "TRUE" "C:\\Cygwin64\\bin\\bash" -c 'cd build-cygwin;ctest -j `nproc`'
   - if "%DO_TEST%" == "TRUE" cd "%BUILDFOLDER%"
   - if "%DO_TEST%" == "TRUE" del /Q /F build-cygwin\bin\iut*
   - if "%DO_TEST%" == "TRUE" echo PATH %ORGPATH%;c:\Cygwin64\bin;c:\Cygwin64\usr\bin;%CD%\build-cygwin\bin;%CD%\build\bin > q.bat

diff --git a/include/sleefdft.h b/include/sleefdft.h
@@ -46,24 +46,6 @@ IMPORT struct SleefDFT *SleefDFT_float_init1d(uint32_t n, const float *in, float
 IMPORT struct SleefDFT *SleefDFT_float_init2d(uint32_t n, uint32_t m, const float *in, float *out, uint64_t mode);
 IMPORT void SleefDFT_float_execute(struct SleefDFT *ptr, const float *in, float *out);
 
-IMPORT struct SleefDFT *SleefDFT_longdouble_init1d(uint32_t n, const long double *in, long double *out, uint64_t mode);
-IMPORT struct SleefDFT *SleefDFT_longdouble_init2d(uint32_t n, uint32_t m, const long double *in, long double *out, uint64_t mode);
-IMPORT void SleefDFT_longdouble_execute(struct SleefDFT *ptr, const long double *in, long double *out);
-
-#if defined(ENABLEFLOAT128) && !defined(Sleef_quad2_DEFINED)
-#define Sleef_quad2_DEFINED
-typedef __float128 Sleef_quad;
-typedef struct {
-  Sleef_quad x, y;
-} Sleef_quad2;
-#endif
-
-#if defined(Sleef_quad2_DEFINED)
-IMPORT struct SleefDFT *SleefDFT_quad_init1d(uint32_t n, const Sleef_quad *in, Sleef_quad *out, uint64_t mode);
-IMPORT struct SleefDFT *SleefDFT_quad_init2d(uint32_t n, uint32_t m, const Sleef_quad *in, Sleef_quad *out, uint64_t mode);
-IMPORT void SleefDFT_quad_execute(struct SleefDFT *ptr, const Sleef_quad *in, Sleef_quad *out);
-#endif
-
 IMPORT void SleefDFT_dispose(struct SleefDFT *ptr);
 
 IMPORT void SleefDFT_setPath(struct SleefDFT *ptr, char *pathStr);

diff --git a/src/arch/helperpurec_scalar.h b/src/arch/helperpurec_scalar.h
@@ -117,16 +117,12 @@ typedef int64_t vint2;
 typedef int64_t vint64;
 typedef uint64_t vuint64;
 
-typedef struct {
-  vmask x, y;
-} vquad;
-
-#if defined(ENABLEFLOAT128) && CONFIG != 3
-typedef __float128 vargquad;
-#elif defined(__SIZEOF_LONG_DOUBLE__) && __SIZEOF_LONG_DOUBLE__ == 16 && defined(__aarch64__) && CONFIG != 3
-typedef long double vargquad;
+typedef Sleef_uint64_2t vquad;
+
+#if CONFIG != 3
+typedef Sleef_quad vargquad;
 #else
-typedef vquad vargquad;
+typedef Sleef_uint64_2t vargquad;
 #endif
 
 //

diff --git a/src/common/keywords.txt b/src/common/keywords.txt
@@ -679,3 +679,6 @@ fma_tdx_tdx_tdx_tdx
 hypot_tdx_tdx_tdx
 ilogb_vi_tdx
 ldexp_tdx_tdx_vi
+Sleef_rempitabsp
+Sleef_rempitabdp
+Sleef_rempitabqp
diff --git a/src/common/misc.h b/src/common/misc.h
@@ -164,19 +164,6 @@ typedef struct {
 } Sleef_float2;
 #endif
 
-#if !defined(Sleef_quad_DEFINED) && !defined(SLEEF_GENHEADER)
-#define Sleef_quad_DEFINED
-#if defined(ENABLEFLOAT128)
-typedef __float128 Sleef_quad;
-#define SLEEF_QUAD_C(x) (x ## Q)
-#elif defined(__SIZEOF_LONG_DOUBLE__) && __SIZEOF_LONG_DOUBLE__ == 16 && (defined(__aarch64__) || defined(__zarch__))
-typedef long double Sleef_quad;
-#define SLEEF_QUAD_C(x) (x ## L)
-#else
-typedef struct { uint64_t x, y; } Sleef_quad;
-#endif
-#endif
-
 //
 
 #if (defined (__GNUC__) || defined (__clang__) || defined(__INTEL_COMPILER)) && !defined(_MSC_VER)

diff --git a/src/common/quaddef.h b/src/common/quaddef.h
@@ -0,0 +1,54 @@
+//   Copyright Naoki Shibata and contributors 2010 - 2021.
+// Distributed under the Boost Software License, Version 1.0.
+//    (See accompanying file LICENSE.txt or copy at
+//          http://www.boost.org/LICENSE_1_0.txt)
+
+#if !defined(SLEEF_GENHEADER)
+
+#if (defined(__SIZEOF_FLOAT128__) && __SIZEOF_FLOAT128__ == 16) || (defined(__linux__) && defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))) || (defined(__PPC64__) && defined(__GNUC__) && !defined(__clang__) && __GNUC__ >= 8)
+#define SLEEF_FLOAT128_IS_IEEEQP
+#endif
+
+#if !defined(SLEEF_FLOAT128_IS_IEEEQP) && defined(__SIZEOF_LONG_DOUBLE__) && __SIZEOF_LONG_DOUBLE__ == 16 && (defined(__aarch64__) || defined(__zarch__))
+#define SLEEF_LONGDOUBLE_IS_IEEEQP
+#endif
+
+#if !defined(Sleef_quad_DEFINED)
+#define Sleef_quad_DEFINED
+typedef struct { uint64_t x, y; } Sleef_uint64_2t;
+#if defined(SLEEF_FLOAT128_IS_IEEEQP)
+typedef __float128 Sleef_quad;
+#define SLEEF_QUAD_C(x) (x ## Q)
+#elif defined(SLEEF_LONGDOUBLE_IS_IEEEQP)
+typedef long double Sleef_quad;
+#define SLEEF_QUAD_C(x) (x ## L)
+#else
+typedef Sleef_uint64_2t Sleef_quad;
+#endif
+#endif
+
+#else // #if !defined(SLEEF_GENHEADER)
+
+SLEEFSHARPif (defined(SLEEFXXX__SIZEOF_FLOAT128__) && SLEEFXXX__SIZEOF_FLOAT128__ == 16) || (defined(SLEEFXXX__linux__) && defined(SLEEFXXX__GNUC__) && (defined(SLEEFXXX__i386__) || defined(SLEEFXXX__x86_64__))) || (defined(SLEEFXXX__PPC64__) && defined(SLEEFXXX__GNUC__) && !defined(SLEEFXXX__clang__) && SLEEFXXX__GNUC__ >= 8)
+SLEEFSHARPdefine SLEEFXXXSLEEF_FLOAT128_IS_IEEEQP
+SLEEFSHARPendif
+
+SLEEFSHARPif !defined(SLEEFXXXSLEEF_FLOAT128_IS_IEEEQP) && defined(SLEEFXXX__SIZEOF_LONG_DOUBLE__) && SLEEFXXX__SIZEOF_LONG_DOUBLE__ == 16 && (defined(SLEEFXXX__aarch64__) || defined(SLEEFXXX__zarch__))
+SLEEFSHARPdefine SLEEFXXXSLEEF_LONGDOUBLE_IS_IEEEQP
+SLEEFSHARPendif
+
+SLEEFSHARPif !defined(SLEEFXXXSleef_quad_DEFINED)
+SLEEFSHARPdefine SLEEFXXXSleef_quad_DEFINED
+typedef struct { uint64_t x, y; } Sleef_uint64_2t;
+SLEEFSHARPif defined(SLEEFXXXSLEEF_FLOAT128_IS_IEEEQP)
+typedef __float128 Sleef_quad;
+SLEEFSHARPdefine SLEEFXXXSLEEF_QUAD_C(x) (x ## Q)
+SLEEFSHARPelif defined(SLEEFXXXSLEEF_LONGDOUBLE_IS_IEEEQP)
+typedef long double Sleef_quad;
+SLEEFSHARPdefine SLEEFXXXSLEEF_QUAD_C(x) (x ## L)
+SLEEFSHARPelse
+typedef Sleef_uint64_2t Sleef_quad;
+SLEEFSHARPendif
+SLEEFSHARPendif
+
+#endif // #if !defined(SLEEF_GENHEADER)
diff --git a/src/libm-tester/CMakeLists.txt b/src/libm-tester/CMakeLists.txt
@@ -131,7 +131,7 @@ set(IUT_LIST ${TARGET_IUT})
 if (BUILD_INLINE_HEADERS AND SED_COMMAND AND CMAKE_CUDA_COMPILER)
   add_executable(iutcuda iutcuda.cu)
   set_target_properties(iutcuda PROPERTIES LINKER_LANGUAGE CUDA)
-  target_compile_options(iutcuda PRIVATE "--fmad=false")
+  target_compile_options(iutcuda PRIVATE "--fmad=false;-Xcompiler;-ffp-contract=off")
   add_dependencies(iutcuda ${TARGET_INLINE_HEADERS})
   add_test_iut(iutcuda 20.0)
   list(APPEND IUT_LIST iutcuda)

diff --git a/src/libm-tester/iutcuda.cu b/src/libm-tester/iutcuda.cu
@@ -12,6 +12,7 @@
 #include <stdint.h>
 #include <cuda.h>
 
+#include "sleefinline_purec_scalar.h"
 #include "sleefinline_cuda.h"
 
 #define STDIN_FILENO 0

diff --git a/src/libm-tester/iutsimd.c b/src/libm-tester/iutsimd.c
@@ -21,6 +21,7 @@
 #include <signal.h>
 #endif
 
+#include "quaddef.h"
 #include "misc.h"
 
 #if !defined(USE_INLINE_HEADER)

diff --git a/src/libm-tester/tester2simddp.c b/src/libm-tester/tester2simddp.c
@@ -20,6 +20,7 @@
 #endif
 
 #include "sleef.h"
+#include "quaddef.h"
 #include "testerutil.h"
 
 #ifdef ENABLE_SSE2

diff --git a/src/libm-tester/tester2simdsp.c b/src/libm-tester/tester2simdsp.c
@@ -17,10 +17,10 @@
 #define _GNU_SOURCE
 #include <unistd.h>
 #include <sys/syscall.h>
-#include <linux/random.h>
 #endif
 
 #include "sleef.h"
+#include "quaddef.h"
 #include "testerutil.h"
 
 #ifdef ENABLE_SSE2