Merge branch 'master' into cmake2

Author: isuruf

include/flang/Error/errmsg-accel.n

@@ -84,3 +84,13 @@
 .MS S 1039 "OpenACC data clause expected after $"
 .MS S 1040 "OpenACC $ data clause may not follow a device_type clause."
 .MS S 1041 "OpenACC $ clause may not follow a device_type clause."
+.MS W 1042 "acc loop vector clause ignored because of inner vector loop."
+Nested acc loop vector directives will not be honored if the loops cannot be
+effectively collapsed or tiled.
+.MS W 1043 "acc loop worker clause ignored because of inner worker loop."
+Nested acc loop worker directives will not be honored if the loops cannot be
+effectively collapsed or tiled.
+.MS S 1044 "acc loop gang not allowed in an acc routine($) procedure."
+.MS S 1045 "call to acc routine(gang$) procedure not allowed in acc loop $."
+.MS S 1046 "OpenACC compute region ignored: no parallel kernels found."
+.MS S 1047 "Inner collapsed loop bounds are not invariant in outer loop."

include/flang/Error/errmsg-common.n

@@ -1,5 +1,5 @@
 .\"/*
-.\" * Copyright (c) 2010-2018, NVIDIA CORPORATION.  All rights reserved.
+.\" * Copyright (c) 2010-2019, NVIDIA CORPORATION.  All rights reserved.
 .\" *
 .\" * Licensed under the Apache License, Version 2.0 (the "License");
 .\" * you may not use this file except in compliance with the License.
@@ -28,3 +28,4 @@ We have one spot for the preinclude file in the gbl. structure.  This is not
 a user visible switch.
 .MS F 704 "Compilation aborted due to previous errors."
 Compilation will abort immediately in case of Severe or Fatal error.
+.MS F 705 "Half precision implementation missing support - $"

include/flang/Error/errmsg.n

@@ -1,5 +1,5 @@
 .\"/*
-.\" * Copyright (c) 1994-2018, NVIDIA CORPORATION.  All rights reserved.
+.\" * Copyright (c) 1994-2019, NVIDIA CORPORATION.  All rights reserved.
 .\" *
 .\" * Licensed under the Apache License, Version 2.0 (the "License");
 .\" * you may not use this file except in compliance with the License.
@@ -717,3 +717,4 @@ Please update your licenses to use this feature.
 .MS S 1200 "OpenMP GPU - [$] is used, it is not implemented yet."
 .MS S 1201 "OpenMP GPU - [$] is used with [$], this usage is not implemented yet."
 .MS S 1202 "OpenMP GPU - [$] is used independently than [$], this usage is not implemented yet."
+.MS S 1204 "OpenMP GPU - Internal compiler error. Reason: [$] at [$]"

lib/scutil/pgnewfil.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2006-2019, NVIDIA CORPORATION.  All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -79,7 +79,12 @@ addn(char *p, unsigned long val, int n)
   return p;
 } /* addn */
 
-static long pgrand = 0;
+#ifdef HOST_WIN
+#define LONG long long
+#else
+#define LONG long
+#endif
+static LONG pgrand = 0;
 static unsigned long pid = 0;
 
 #if defined(USETEMPNAM) || defined(HOST_WIN) || defined(WIN64)
@@ -109,15 +114,15 @@ gentmp(char *pfx, char *sfx)
   if (pgrand == 0) { /* first time, create seed */
     char *q;
     q = getenv("PATH");
-    pgrand = (long)q;
+    pgrand = (LONG)q;
     q = getenv("USER");
     if (q != NULL) {
       int n = 0;
       while (*q != '\0') {
-        pgrand ^= (long)(*q++) << n++;
+        pgrand ^= (LONG)(*q++) << n++;
       }
     }
-    pgrand ^= (long)filename >> 4;
+    pgrand ^= (LONG)filename >> 4;
     pgrand ^= time((long *)0);
     pid = getpid();
 #if DEBUG
@@ -153,7 +158,7 @@ gentmp(char *pfx, char *sfx)
   pgrand = (pgrand << 16) + pgrand * 3;
   p = addn(p, pgrand, 4);
   pgrand = (pgrand << 16) + pgrand * 3;
-  p = addn(p, pgrand ^ (((long)(pfx) ^ (long)(sfx)) >> 5), 4);
+  p = addn(p, pgrand ^ (((LONG)(pfx) ^ (LONG)(sfx)) >> 5), 4);
   if (sfx)
     p = add(p, sfx);
   return filename;

runtime/flang/CMakeLists.txt

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2017-2018, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2017-2019, NVIDIA CORPORATION.  All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -204,6 +204,8 @@ SET(FTN_INTRINSICS_DESC_DEP
   red_minloc.c
   red_maxval.c
   red_minval.c
+  red_norm2.c
+  red_norm2_stride1.c
   red_sum.c
   reduct.c
   scatter.c
@@ -419,24 +421,39 @@ SET(FTN_SUPPORT_DESC_DEP
   )
 
 set(I8_FILES_DIR I8_sources)
+separate_arguments(SEPARATED_CMAKE_Fortran_FLAGS NATIVE_COMMAND ${CMAKE_Fortran_FLAGS})
 
-# Fortran files with macros as module names need to be preprocessed. 
+# Fortran files with macros as module names need to be preprocessed.
+# CMake has an internal Fortran parser that parses the module name, but it doesn't
+# consider macros which results in wrong dependencies.
 add_custom_command(
   OUTPUT "${I8_FILES_DIR}/ieee_arithmetic.F95"
-  COMMAND "${CMAKE_C_COMPILER}" -E 
+  COMMAND "${CMAKE_Fortran_COMPILER}" -E -cpp ${SEPARATED_CMAKE_Fortran_FLAGS}
   "${CMAKE_CURRENT_SOURCE_DIR}/ieee_arithmetic.F95" -DDESC_I8 
   > "${I8_FILES_DIR}/ieee_arithmetic.F95"
   COMMENT "Preprocessing ieee_arithmetic.F95"
   VERBATIM
+  DEPENDS flang1 flang2
 )
 
 add_custom_command(
   OUTPUT "${I8_FILES_DIR}/ieee_exceptions.F95"
-  COMMAND "${CMAKE_C_COMPILER}" -E 
+  COMMAND "${CMAKE_Fortran_COMPILER}" -E -cpp ${SEPARATED_CMAKE_Fortran_FLAGS}
   "${CMAKE_CURRENT_SOURCE_DIR}/ieee_exceptions.F95" -DDESC_I8 
   > "${I8_FILES_DIR}/ieee_exceptions.F95"
   COMMENT "Preprocessing ieee_exceptions.F95"
   VERBATIM
+  DEPENDS flang1 flang2
+)
+
+add_custom_command(
+  OUTPUT ${I8_FILES_DIR}/norm2_1.F95 norm2_1.F95
+  COMMAND "${CMAKE_Fortran_COMPILER}" -E -DDESC_I8 -DPGFLANG -cpp ${SEPARATED_CMAKE_Fortran_FLAGS}
+  ${CMAKE_CURRENT_SOURCE_DIR}/norm2.F95 > "${I8_FILES_DIR}/norm2_1.F95"
+  COMMAND "${CMAKE_Fortran_COMPILER}" -E -DPGFLANG -cpp ${SEPARATED_CMAKE_Fortran_FLAGS}
+  ${CMAKE_CURRENT_SOURCE_DIR}/norm2.F95 > "norm2_1.F95"
+  VERBATIM
+  DEPENDS flang1 flang2
 )
 
 # The files lists FTN_INTRINSICS_DESC_DEP and FTN_SUPPORT_DESC_DEP need to be compiled twice (with and without 'DESC_I8' compile definition). So an actual copy is made in a temp file on which this is done.
@@ -453,10 +470,12 @@ FOREACH(file ${FTN_SUPPORT_DESC_DEP})
     LIST(APPEND FTN_SUPPORT_I8 ${I8_FILE})
 ENDFOREACH(file)
 
-LIST(APPEND FTN_SUPPORT_DESC_DEP ieee_arithmetic.F95)
-LIST(APPEND FTN_SUPPORT_DESC_DEP ieee_exceptions.F95)
-LIST(APPEND FTN_SUPPORT_I8 "${I8_FILES_DIR}/ieee_arithmetic.F95")
-LIST(APPEND FTN_SUPPORT_I8 "${I8_FILES_DIR}/ieee_exceptions.F95")
+LIST(APPEND FTN_SUPPORT_DESC_DEP ieee_arithmetic.F95 ieee_exceptions.F95 norm2_1.F95)
+LIST(APPEND FTN_SUPPORT_I8 
+"${I8_FILES_DIR}/ieee_arithmetic.F95"
+"${I8_FILES_DIR}/ieee_exceptions.F95"
+"${I8_FILES_DIR}/norm2_1.F95"
+)
 
 add_flang_library(flang_static
   ${FTN_INTRINSICS_DESC_INDEP}
@@ -488,6 +507,16 @@ add_flang_library(flang_shared
   ${SHARED_SOURCES}
   )
 set_property(TARGET flang_shared PROPERTY OUTPUT_NAME flang)
+#
+# Seralize the building of flang_shared and flang_static to eliminate
+# conflicts with the same module files from the shared and static builds
+# being created/recreated in the common directory
+# ${CMAKE_Fortran_MODULE_DIRECTORY}.
+#
+# Note: building of each library is still parallelized.
+#
+add_dependencies(flang_shared flang_static)
+
 target_link_libraries(flang_shared flangrti_shared)
 # Resolve symbols against libm and librt
 if (NOT MSVC)
@@ -542,6 +571,13 @@ set_property(
  PG_PIC
  )
 
+set_source_files_properties(
+  ${I8_FILES_DIR}/red_norm2_stride1.c
+  red_norm2_stride1.c
+  PROPERTIES
+  COMPILE_FLAGS "-ffast-math"
+  )
+
 set_target_properties(flang_static flang_shared
   PROPERTIES
   ARCHIVE_OUTPUT_DIRECTORY ${FLANG_RTE_LIB_DIR}

runtime/flang/allo.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1995-2018, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 1995-2019, NVIDIA CORPORATION.  All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -1190,14 +1190,17 @@ ENTF90(PTR_SRC_ALLOC03A, ptr_src_alloc03a)(F90_Desc *sd, __INT_T *nelem,
   src_len = ENTF90(GET_OBJECT_SIZE, get_object_size)(sd);
   if (sd && sd->tag == __DESC && sd->lsize > 1)
     src_len *= sd->lsize;
+  else if (nelem && *nelem > 1) {
+    src_len *= *nelem;
+  }
   max_len = (len && nelem) ? (*len * *nelem) : 0;
   if (max_len < src_len)
     max_len = src_len;
 
   if (ISPRESENT(stat) && firsttime && *firsttime)
     *stat = 0;
 
-  (void)I8(__alloc04)(*nelem, (dtype)*kind, (size_t)max_len, stat, pointer,
+  (void)I8(__alloc04)(1, (dtype)*kind, (size_t)max_len, stat, pointer,
                       offset, 0, 0, LOCAL_MODE ? __fort_malloc_without_abort
                                                : __fort_gmalloc_without_abort,
                       0, CADR(errmsg), CLEN(errmsg));
@@ -1225,16 +1228,19 @@ ENTF90(PTR_SRC_CALLOC03A, ptr_src_calloc03a)(F90_Desc *sd, __INT_T *nelem,
   __INT_T src_len, max_len;
 
   src_len = ENTF90(GET_OBJECT_SIZE, get_object_size)(sd);
-  if (sd && sd->tag == __DESC && sd->lsize > 1)
+  if (sd && sd->tag == __DESC && sd->lsize > 1) {
     src_len *= sd->lsize;
+  } else if (nelem && *nelem > 1) {
+    src_len *= *nelem; 
+  }
   max_len = (len && nelem) ? (*len * *nelem) : 0;
   if (max_len < src_len)
     max_len = src_len;
 
   if (ISPRESENT(stat) && firsttime && *firsttime)
     *stat = 0;
 
-  (void)I8(__alloc04)(*nelem, (dtype)*kind, (size_t)max_len, stat, pointer,
+  (void)I8(__alloc04)(1, (dtype)*kind, (size_t)max_len, stat, pointer,
                       offset, 0, 0, LOCAL_MODE ? __fort_calloc_without_abort
                                                : __fort_gcalloc_without_abort,
                       0, CADR(errmsg), CLEN(errmsg));
@@ -1264,14 +1270,17 @@ ENTF90(PTR_SRC_ALLOC04A, ptr_src_alloc04a)(F90_Desc *sd, __NELEM_T *nelem,
   src_len = ENTF90(GET_OBJECT_SIZE, get_object_size)(sd);
   if (sd && sd->tag == __DESC && sd->lsize > 1)
     src_len *= sd->lsize;
+  else if (nelem && *nelem > 1) {
+    src_len *= *nelem;
+  }
   max_len = (len && nelem) ? (*len * *nelem) : 0;
   if (max_len < src_len)
     max_len = src_len;
 
   if (ISPRESENT(stat) && firsttime && *firsttime)
     *stat = 0;
 
-  (void)I8(__alloc04)(*nelem, (dtype)*kind, (size_t)max_len, stat, pointer,
+  (void)I8(__alloc04)(1, (dtype)*kind, (size_t)max_len, stat, pointer,
                       offset, 0, 0, LOCAL_MODE ? __fort_malloc_without_abort
                                                : __fort_gmalloc_without_abort,
                       *align, CADR(errmsg), CLEN(errmsg));
@@ -1309,6 +1318,8 @@ ENTF90(PTR_SRC_CALLOC04A, ptr_src_calloc04a)
                sd->kind > 0 && sd->kind <= __NTYPES) {
       src_len = sd->len;
     }
+  } else if (nelem && *nelem > 1) {
+    src_len *= *nelem; 
   }
   max_len = (len && nelem) ? (*len * *nelem) : 0;
   if (max_len < src_len)
@@ -1317,7 +1328,7 @@ ENTF90(PTR_SRC_CALLOC04A, ptr_src_calloc04a)
   if (ISPRESENT(stat) && firsttime && *firsttime)
     *stat = 0;
 
-  (void)I8(__alloc04)(*nelem, (dtype)*kind, (size_t)max_len, stat, pointer,
+  (void)I8(__alloc04)(1, (dtype)*kind, (size_t)max_len, stat, pointer,
                       offset, 0, 0, LOCAL_MODE ? __fort_calloc_without_abort
                                                : __fort_gcalloc_without_abort,
                       *align, CADR(errmsg), CLEN(errmsg));

runtime/flang/dbug.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1995-2018, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 1995-2019, NVIDIA CORPORATION.  All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -362,7 +362,8 @@ void ENTFTN(SHOW, show)(void *b, F90_Desc *d)
   }
 }
 
-void ENTF90(SHOW_, show_)(void *b, F90_Desc *d)
+#if (defined(DESC_I8) && defined(__PGLLVM__)) || (!defined(DESC_I8) && !defined(__PGLLVM__))
+void ENTF90COMN(SHOW_, show_)(void *b, F90_Desc *d)
 {
   DECL_HDR_PTRS(t);
   DECL_DIM_PTRS(td);
@@ -410,6 +411,7 @@ void ENTF90(SHOW_, show_)(void *b, F90_Desc *d)
     }
   }
 }
+#endif /* (defined(DESC_I8) && defined(__PGLLVM__)) || (!defined(DESC_I8) && !defined(__PGLLVM__)) */
 
 void I8(__fort_describe)(char *b, F90_Desc *d)
 {

runtime/flang/directives.h

@@ -52,6 +52,7 @@
 #define ELF_OBJ(s) .type ENT(s), @object
 #define ELF_SIZE(s) .size ENT(s), .- ENT(s)
 #endif
+#define ELF_HIDDEN(s)
 #define AS_VER .version "01.01"
 #define I1 %rcx
 #define I1W %ecx
@@ -87,6 +88,7 @@
 #define ELF_FUNC(s) .type ENT(s), @function
 #define ELF_OBJ(s) .type ENT(s), @object
 #define ELF_SIZE(s) .size ENT(s), .- ENT(s)
+#define ELF_HIDDEN(s)	.globl	s ;	.hidden	s
 #define AS_VER .version "01.01"
 #define I1 %rdi
 #define I1W %edi
@@ -110,6 +112,7 @@
 #define ELF_FUNC(s)
 #define ELF_OBJ(s)
 #define ELF_SIZE(s)
+#define ELF_HIDDEN(s)
 #define AS_VER
 #define I1 %rdi
 #define I1W %edi