[Cleanup] Conform strict aliasing rules (#392)

This is a combined patch for removing potential problems with the strict aliasing rule.
It also drops long double support for DFT.

CMakeLists.txt

@@ -88,11 +88,6 @@ set(COSTOVERRIDE_SVENOFMA 10)
 
 cmake_minimum_required(VERSION 3.4.3)
 
-# Set to NEW when updating cmake_minimum_required to VERSION >= 3.7.2
-if(${CMAKE_VERSION} VERSION_GREATER "3.7.1")
-  cmake_policy(SET CMP0066 OLD)
-endif()
-
 if(${CMAKE_VERSION} VERSION_GREATER "3.14.99")
   cmake_policy(SET CMP0091 NEW)
 endif()

Jenkinsfile

@@ -53,7 +53,7 @@ pipeline {
 			 rm -rf build
  			 mkdir build
 			 cd build
-			 cmake -GNinja -DCMAKE_INSTALL_PREFIX=../install -DSLEEF_SHOW_CONFIG=1 -DBUILD_SHARED_LIBS=FALSE -DENFORCE_TESTER3=TRUE -DFORCE_AAVPCS=On -DENABLE_GNUABI=On -DBUILD_QUAD=TRUE -DBUILD_DFT=TRUE -DBUILD_INLINE_HEADERS=TRUE -DENABLE_CUDA=TRUE -DENFORCE_CUDA=TRUE -DENFORCE_SVE=TRUE -DEMULATOR=qemu-aarch64 ..
+			 cmake -GNinja -DCMAKE_INSTALL_PREFIX=../install -DSLEEF_SHOW_CONFIG=1 -DBUILD_SHARED_LIBS=FALSE -DENFORCE_TESTER3=TRUE -DFORCE_AAVPCS=On -DENABLE_GNUABI=On -DBUILD_QUAD=TRUE -DBUILD_DFT=TRUE -DBUILD_INLINE_HEADERS=FALSE -DENABLE_CUDA=TRUE -DENFORCE_CUDA=TRUE -DENFORCE_SVE=TRUE -DEMULATOR=qemu-aarch64 ..
 			 ninja
 			 export OMP_WAIT_POLICY=passive
 		         export CTEST_OUTPUT_ON_FAILURE=TRUE

src/arch/helperpurec_scalar.h

@@ -123,7 +123,7 @@ typedef struct {
 
 #if defined(ENABLEFLOAT128) && CONFIG != 3
 typedef __float128 vargquad;
-#elif defined(__SIZEOF_LONG_DOUBLE__) && defined(__aarch64__) && CONFIG != 3
+#elif defined(__SIZEOF_LONG_DOUBLE__) && __SIZEOF_LONG_DOUBLE__ == 16 && defined(__aarch64__) && CONFIG != 3
 typedef long double vargquad;
 #else
 typedef vquad vargquad;
@@ -444,19 +444,19 @@ static INLINE vargquad cast_aq_vq(vquad vq) {
 #else
 static vquad loadu_vq_p(void *p) {
   vquad vq;
-  memcpy(&vq, p, 16);
+  memcpy(&vq, p, sizeof(vq));
   return vq;
 }
 
 static INLINE vquad cast_vq_aq(vargquad aq) {
   vquad vq;
-  memcpy(&vq, &aq, 16);
+  memcpy(&vq, &aq, sizeof(vq));
   return vq;
 }
 
 static INLINE vargquad cast_aq_vq(vquad vq) {
   vargquad aq;
-  memcpy(&aq, &vq, 16);
+  memcpy(&aq, &vq, sizeof(aq));
   return aq;
 }
 #endif

src/common/misc.h

@@ -150,10 +150,6 @@
 #define stringify(s) stringify_(s)
 #define stringify_(s) #s
 
-#if !defined(SLEEF_GENHEADER)
-typedef long double longdouble;
-#endif
-
 #if !defined(Sleef_double2_DEFINED) && !defined(SLEEF_GENHEADER)
 #define Sleef_double2_DEFINED
 typedef struct {
@@ -168,70 +164,19 @@ typedef struct {
 } Sleef_float2;
 #endif
 
-#if !defined(Sleef_longdouble2_DEFINED) && !defined(SLEEF_GENHEADER)
-#define Sleef_longdouble2_DEFINED
-typedef struct {
-  long double x, y;
-} Sleef_longdouble2;
-#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__) && defined(__aarch64__)
+#elif defined(__SIZEOF_LONG_DOUBLE__) && __SIZEOF_LONG_DOUBLE__ == 16 && defined(__aarch64__)
 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(Sleef_quad1_DEFINED) && !defined(SLEEF_GENHEADER)
-#define Sleef_quad1_DEFINED
-typedef union {
-  struct {
-    Sleef_quad x;
-  };
-  Sleef_quad s[1];
-} Sleef_quad1;
-#endif
-
-#if !defined(Sleef_quad2_DEFINED) && !defined(SLEEF_GENHEADER)
-#define Sleef_quad2_DEFINED
-typedef union {
-  struct {
-    Sleef_quad x, y;
-  };
-  Sleef_quad s[2];
-} Sleef_quad2;
-#endif
-
-#if !defined(Sleef_quad4_DEFINED) && !defined(SLEEF_GENHEADER)
-#define Sleef_quad4_DEFINED
-typedef union {
-  struct {
-    Sleef_quad x, y, z, w;
-  };
-  Sleef_quad s[4];
-} Sleef_quad4;
-#endif
-
-#if !defined(Sleef_quad8_DEFINED) && !defined(SLEEF_GENHEADER)
-#define Sleef_quad8_DEFINED
-typedef union {
-  Sleef_quad s[8];
-} Sleef_quad8;
-#endif
-
-#if defined(__ARM_FEATURE_SVE) && !defined(Sleef_quadx_DEFINED) && !defined(SLEEF_GENHEADER)
-#define Sleef_quadx_DEFINED
-typedef union {
-  Sleef_quad s[32];
-} Sleef_quadx;
-#endif
-
 //
 
 #if (defined (__GNUC__) || defined (__clang__) || defined(__INTEL_COMPILER)) && !defined(_MSC_VER)

src/dft/CMakeLists.txt

@@ -12,16 +12,14 @@ if (SLEEFDFT_MAXBUTWIDTH GREATER 7)
 endif()
 
 option(SLEEFDFT_ENABLE_STREAM "Streaming instructions are utilized in DFT." OFF)
-option(SLEEFDFT_ENABLE_LONGDOUBLE "Long double routines will be compiled in." OFF)
-option(SLEEFDFT_ENABLE_QUAD "Quad precision routines will be compiled in." OFF)
 
 # Settings
 
 # Constants definition
 
-set(LISTSHORTTYPENAME "dp" "sp" "ld" "qp")
-set(LISTLONGTYPENAME "double" "float" "longdouble" "Sleef_quad")
-set(LISTTYPEID "1" "2" "3" "4")
+set(LISTSHORTTYPENAME "dp" "sp")
+set(LISTLONGTYPENAME "double" "float")
+set(LISTTYPEID "1" "2")
 
 set(MACRODEF_vecextdp BASETYPEID=1 ENABLE_VECEXT CONFIG=1)
 set(CFLAGS_vecextdp ${FLAGS_ENABLE_VECEXT})
@@ -105,22 +103,6 @@ if(CMAKE_C_COMPILER_ID MATCHES "(GNU|Clang)")
   set(ISALIST_DP vecextdp)
 endif(CMAKE_C_COMPILER_ID MATCHES "(GNU|Clang)")
 
-if (COMPILER_SUPPORTS_LONG_DOUBLE AND SLEEFDFT_ENABLE_LONGDOUBLE)
-  set(LIST_SUPPORTED_FPTYPE ${LIST_SUPPORTED_FPTYPE} 2)
-  set(ISALIST_QP purecld)
-  if(CMAKE_C_COMPILER_ID MATCHES "(GNU|Clang)")
-    set(ISALIST_LD vecextld)
-  endif(CMAKE_C_COMPILER_ID MATCHES "(GNU|Clang)")
-endif(COMPILER_SUPPORTS_LONG_DOUBLE AND SLEEFDFT_ENABLE_LONGDOUBLE)
-
-if (COMPILER_SUPPORTS_FLOAT128 AND SLEEFDFT_ENABLE_QUAD)
-  set(LIST_SUPPORTED_FPTYPE ${LIST_SUPPORTED_FPTYPE} 3)
-  set(ISALIST_QP purecqp)
-  if(CMAKE_C_COMPILER_ID MATCHES "(GNU|Clang)")
-    set(ISALIST_QP vecextqp)
-  endif(CMAKE_C_COMPILER_ID MATCHES "(GNU|Clang)")
-endif(COMPILER_SUPPORTS_FLOAT128 AND SLEEFDFT_ENABLE_QUAD)
-
 # List all available vector data types
 
 if (COMPILER_SUPPORTS_SSE4)
@@ -204,14 +186,6 @@ else()
   set(COMMON_TARGET_DEFINITIONS ${COMMON_TARGET_DEFINITIONS} ENABLE_STREAM=0)
 endif()
 
-if (COMPILER_SUPPORTS_FLOAT128 AND NOT (SLEEF_ARCH_32BIT AND SLEEF_ARCH_X86))
-  set(COMMON_TARGET_DEFINITIONS ${COMMON_TARGET_DEFINITIONS} ENABLEFLOAT128)
-endif()
-
-if (COMPILER_SUPPORTS_LONG_DOUBLE AND NOT (SLEEF_ARCH_32BIT AND SLEEF_ARCH_X86))
-  set(COMMON_TARGET_DEFINITIONS ${COMMON_TARGET_DEFINITIONS} ENABLE_LONGDOUBLE)
-endif()
-
 if(COMPILER_SUPPORTS_OPENMP)
   set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
 endif(COMPILER_SUPPORTS_OPENMP)

src/dft/dft.c

@@ -68,50 +68,6 @@ typedef Sleef_float2 sc_t;
 #define BUTB butb_float
 #define SINCOSPI Sleef_sincospif_u05
 #include "dispatchsp.h"
-#elif BASETYPEID == 3
-typedef long double real;
-typedef Sleef_longdouble2 sc_t;
-#define BASETYPESTRING "long double"
-#define MAGIC 0x14142135
-#define MAGIC2D 0x26457513
-#define INIT SleefDFT_longdouble_init1d
-#define EXECUTE SleefDFT_longdouble_execute
-#define INIT2D SleefDFT_longdouble_init2d
-#define CTBL ctbl_longdouble
-#define REALSUB0 realSub0_longdouble
-#define REALSUB1 realSub1_longdouble
-#define GETINT getInt_longdouble
-#define GETPTR getPtr_longdouble
-#define DFTF dftf_longdouble
-#define DFTB dftb_longdouble
-#define TBUTF tbutf_longdouble
-#define TBUTB tbutb_longdouble
-#define BUTF butf_longdouble
-#define BUTB butb_longdouble
-#define SINCOSPI Sleef_sincospil_u05
-#include "dispatchld.h"
-#elif BASETYPEID == 4
-typedef Sleef_quad real;
-typedef Sleef_quad2 sc_t;
-#define BASETYPESTRING "Sleef_quad"
-#define MAGIC 0x33166247
-#define MAGIC2D 0x36055512
-#define INIT SleefDFT_quad_init1d
-#define EXECUTE SleefDFT_quad_execute
-#define INIT2D SleefDFT_quad_init2d
-#define CTBL ctbl_Sleef_quad
-#define REALSUB0 realSub0_Sleef_quad
-#define REALSUB1 realSub1_Sleef_quad
-#define GETINT getInt_Sleef_quad
-#define GETPTR getPtr_Sleef_quad
-#define DFTF dftf_Sleef_quad
-#define DFTB dftb_Sleef_quad
-#define TBUTF tbutf_Sleef_quad
-#define TBUTB tbutb_Sleef_quad
-#define BUTF butf_Sleef_quad
-#define BUTB butb_Sleef_quad
-#define SINCOSPI Sleef_sincospiq_u05
-#include "dispatchqp.h"
 #else
 #error No BASETYPEID specified
 #endif
@@ -121,47 +77,6 @@ typedef Sleef_quad2 sc_t;
 
 //
 
-#if BASETYPEID == 4
-real CTBL[] = {
-  0.7071067811865475243818940365159164684883Q, -0.7071067811865475243818940365159164684883Q,
-  0.9238795325112867561014214079495587839119Q, -0.382683432365089771723257530688933059082Q,
-  0.382683432365089771723257530688933059082Q, -0.9238795325112867561014214079495587839119Q,
-#if MAXBUTWIDTH >= 5
-  0.9807852804032304491190993878113602022495Q, -0.1950903220161282678433729148581576851029Q,
-  0.5555702330196022247573058028269343822103Q, -0.8314696123025452370808655033762590846891Q,
-  0.8314696123025452370808655033762590846891Q, -0.5555702330196022247573058028269343822103Q,
-  0.1950903220161282678433729148581576851029Q, -0.9807852804032304491190993878113602022495Q,
-#endif
-#if MAXBUTWIDTH >= 6
-  0.9951847266721968862310254699821143731242Q, -0.09801714032956060199569840382660679267701Q,
-  0.6343932841636454982026105398063009488396Q, -0.7730104533627369607965383602188325085081Q,
-  0.881921264348355029715105513066220055407Q, -0.4713967368259976485449225247492677226546Q,
-  0.2902846772544623676448431737195932100803Q, -0.9569403357322088649310892760624369657307Q,
-  0.9569403357322088649310892760624369657307Q, -0.2902846772544623676448431737195932100803Q,
-  0.4713967368259976485449225247492677226546Q, -0.881921264348355029715105513066220055407Q,
-  0.7730104533627369607965383602188325085081Q, -0.6343932841636454982026105398063009488396Q,
-  0.09801714032956060199569840382660679267701Q, -0.9951847266721968862310254699821143731242Q,
-#endif
-#if MAXBUTWIDTH >= 7
-  0.9987954562051723927007702841240899260811Q, -0.04906767432741801425355085940205324135377Q,
-  0.6715589548470184006194634573905233310143Q, -0.7409511253549590911932944126139233276263Q,
-  0.9039892931234433315823215138173907234886Q, -0.427555093430282094315230886905077056781Q,
-  0.336889853392220050702686798271834334173Q, -0.9415440651830207783906830087961026265475Q,
-  0.9700312531945439926159106824865574481009Q, -0.2429801799032638899447731489766866275204Q,
-  0.5141027441932217266072797923204262815489Q, -0.8577286100002720698929313536407192941624Q,
-  0.8032075314806449097991200569701675249235Q, -0.5956993044924333434615715265891822127742Q,
-  0.1467304744553617516588479505190711904561Q, -0.9891765099647809734561415551112872890371Q,
-  0.9891765099647809734561415551112872890371Q, -0.1467304744553617516588479505190711904561Q,
-  0.5956993044924333434615715265891822127742Q, -0.8032075314806449097991200569701675249235Q,
-  0.8577286100002720698929313536407192941624Q, -0.5141027441932217266072797923204262815489Q,
-  0.2429801799032638899447731489766866275204Q, -0.9700312531945439926159106824865574481009Q,
-  0.9415440651830207783906830087961026265475Q, -0.336889853392220050702686798271834334173Q,
-  0.427555093430282094315230886905077056781Q, -0.9039892931234433315823215138173907234886Q,
-  0.7409511253549590911932944126139233276263Q, -0.6715589548470184006194634573905233310143Q,
-  0.04906767432741801425355085940205324135377Q, -0.9987954562051723927007702841240899260811Q,
-#endif
-};
-#else
 real CTBL[] = {
   0.7071067811865475243818940365159164684883L, -0.7071067811865475243818940365159164684883L,
   0.9238795325112867561014214079495587839119L, -0.382683432365089771723257530688933059082L,
@@ -201,7 +116,6 @@ real CTBL[] = {
   0.04906767432741801425355085940205324135377L, -0.9987954562051723927007702841240899260811L,
 #endif
 };
-#endif
 
 #ifndef ENABLE_STREAM
 #error ENABLE_STREAM not defined

src/dft/dftcommon.c

@@ -29,13 +29,9 @@
 
 #define MAGIC_FLOAT 0x31415926
 #define MAGIC_DOUBLE 0x27182818
-#define MAGIC_LONGDOUBLE 0x14142135
-#define MAGIC_QUAD 0x33166247
 
 #define MAGIC2D_FLOAT 0x22360679
 #define MAGIC2D_DOUBLE 0x17320508
-#define MAGIC2D_LONGDOUBLE 0x26457513
-#define MAGIC2D_QUAD 0x36055512
 
 const char *configStr[] = { "ST", "ST stream", "MT", "MT stream" };
 
@@ -76,7 +72,7 @@ static int parsePathStr(char *p, int *path, int *config, int pathLenMax, int log
 }
 
 EXPORT void SleefDFT_setPath(SleefDFT *p, char *pathStr) {
-  assert(p != NULL && (p->magic == MAGIC_FLOAT || p->magic == MAGIC_DOUBLE || p->magic == MAGIC_LONGDOUBLE || p->magic == MAGIC_QUAD));
+  assert(p != NULL && (p->magic == MAGIC_FLOAT || p->magic == MAGIC_DOUBLE));
 
   int path[32], config[32];
   int pathLen = parsePathStr(pathStr, path, config, 31, p->log2len);
@@ -116,7 +112,7 @@ void freeTables(SleefDFT *p) {
 }
 
 EXPORT void SleefDFT_dispose(SleefDFT *p) {
-  if (p != NULL && (p->magic == MAGIC2D_FLOAT || p->magic == MAGIC2D_DOUBLE || p->magic == MAGIC2D_LONGDOUBLE || p->magic == MAGIC2D_QUAD)) {
+  if (p != NULL && (p->magic == MAGIC2D_FLOAT || p->magic == MAGIC2D_DOUBLE)) {
     Sleef_free(p->tBuf);
     SleefDFT_dispose(p->instH);
     if (p->hlen != p->vlen) SleefDFT_dispose(p->instV);
@@ -126,7 +122,7 @@ EXPORT void SleefDFT_dispose(SleefDFT *p) {
     return;
   }
 
-  assert(p != NULL && (p->magic == MAGIC_FLOAT || p->magic == MAGIC_DOUBLE || p->magic == MAGIC_LONGDOUBLE || p->magic == MAGIC_QUAD));
+  assert(p != NULL && (p->magic == MAGIC_FLOAT || p->magic == MAGIC_DOUBLE));
 
   if (p->log2len <= 1) {
     p->magic = 0;
@@ -322,7 +318,7 @@ static void planMap_putU64(uint64_t key, uint64_t value) {
 }
 
 int PlanManager_loadMeasurementResultsP(SleefDFT *p, int pathCat) {
-  assert(p != NULL && (p->magic == MAGIC_FLOAT || p->magic == MAGIC_DOUBLE || p->magic == MAGIC_LONGDOUBLE || p->magic == MAGIC_QUAD));
+  assert(p != NULL && (p->magic == MAGIC_FLOAT || p->magic == MAGIC_DOUBLE));
 
   initPlanMapLock();
 
@@ -357,7 +353,7 @@ int PlanManager_loadMeasurementResultsP(SleefDFT *p, int pathCat) {
 }
 
 void PlanManager_saveMeasurementResultsP(SleefDFT *p, int pathCat) {
-  assert(p != NULL && (p->magic == MAGIC_FLOAT || p->magic == MAGIC_DOUBLE || p->magic == MAGIC_LONGDOUBLE || p->magic == MAGIC_QUAD));
+  assert(p != NULL && (p->magic == MAGIC_FLOAT || p->magic == MAGIC_DOUBLE));
 
   initPlanMapLock();
 
@@ -388,7 +384,7 @@ void PlanManager_saveMeasurementResultsP(SleefDFT *p, int pathCat) {
 }
 
 int PlanManager_loadMeasurementResultsT(SleefDFT *p) {
-  assert(p != NULL && (p->magic == MAGIC2D_FLOAT || p->magic == MAGIC2D_DOUBLE || p->magic == MAGIC2D_LONGDOUBLE || p->magic == MAGIC2D_QUAD));
+  assert(p != NULL && (p->magic == MAGIC2D_FLOAT || p->magic == MAGIC2D_DOUBLE));
 
   initPlanMapLock();
 
@@ -409,7 +405,7 @@ int PlanManager_loadMeasurementResultsT(SleefDFT *p) {
 }
 
 void PlanManager_saveMeasurementResultsT(SleefDFT *p) {
-  assert(p != NULL && (p->magic == MAGIC2D_FLOAT || p->magic == MAGIC2D_DOUBLE || p->magic == MAGIC2D_LONGDOUBLE || p->magic == MAGIC2D_QUAD));
+  assert(p != NULL && (p->magic == MAGIC2D_FLOAT || p->magic == MAGIC2D_DOUBLE));
 
   initPlanMapLock();