Merge pull request #818 from ThePortlandGroup/nv_stage

sscalpone · web-flow · commit 249b986ddbc9 · 2019-10-24T16:28:13.000-07:00
Pull 2019-10-24T15-13 Recent NVIDIA Changes
diff --git a/include/flang/Error/errmsg-in.n b/include/flang/Error/errmsg-in.n
@@ -1510,13 +1510,16 @@ A DO CONCURRENT or FORALL construct or statement may not specify an index name m
 .MS S 1061 "The definition of function return type of $ does not match its declaration type"
 .MS S 1062 "LOCAL_INIT variable does not have an outside variable of the same name - $"
 A DO CONCURRENT variable with LOCAL_INIT locality must have a host variable of the same name.
-.MS S 1200 "OpenMP GPU - [$] is used, it is not implemented yet."
+Starting from 1100, Reserved for OpenMP GPU
+.MS S 1198 "OpenMP GPU - The feature is not implemented yet for the target device"
+.MS W 1199 "OpenMP GPU - \"$\" is ignored for the target device"
+.MS S 1200 "OpenMP GPU - \"$\" 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 1203 "OpenMP GPU - Directive target exit data is used, map type [$] cannot be used. It should be one of 'from', 'release', or 'delete'"
 .MS S 1204 "OpenMP GPU - Directive target data is used, map type [$] cannot be used. It should be one of 'to', 'from', 'tofrom', 'alloc', 'release', or 'delete'"
 .MS S 1205 "OpenMP GPU - Directive target enter data is used, map type [$] cannot be used. It should be one of 'to', or 'alloc'"
-.MS W 1206 "OpenMP GPU - Using [map] clause is used with an array section for $ array is not supported, mapping will be done on entire array."
+.MS W 1206 "OpenMP GPU - Array sections in \"map\" clause is not implemented yet, the entire array will be mapped"
 .MS S 1207 "ERROR STOP stop-code requires either a character or integer expression."
 .MS S 1208 "QUIET requires a logical expression."
 .MS S 1209 "ERROR STOP stop-code integer expression must be an integer of default kind."
diff --git a/include/flang/Error/errmsg.n b/include/flang/Error/errmsg.n
@@ -714,7 +714,10 @@ In integer shift, ISHFT(I, SHIFT), absolute value of SHIFT cannot be bigger than
 The accelerator feature in the PGI compilers is separately licensed;
 that license was not found.
 Please update your licenses to use this feature.
-.MS S 1200 "OpenMP GPU - [$] is used, it is not implemented yet."
+Starting from 1100, Reserved for OpenMP GPU
+.MS S 1198 "OpenMP GPU - The feature is not implemented yet for the target device"
+.MS W 1199 "OpenMP GPU - \"$\" is ignored for the target device"
+.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 [$]"
diff --git a/runtime/flang/ipopcnt.c b/runtime/flang/ipopcnt.c
@@ -26,13 +26,11 @@ __mth_i_ipopcnt(uint32_t u32)
     asm("popcnt %1, %0"
        : "=r"(r32)
        : "r"(u32)
-       :
        );
 #elif   defined(TARGET_LINUX_POWER)
     asm("popcntw    %0, %1"
        : "=r"(r32)
        : "r"(u32)
-       :
        );
 #else
   static const uint32_t u5s = 0x55555555;
diff --git a/runtime/flang/ipopcnti.c b/runtime/flang/ipopcnti.c
@@ -38,13 +38,11 @@ __mth_i_ipopcnti(uint32_t u32, int size)
     asm("popcnt %1, %0"
        : "=r"(r32)
        : "r"(r32)
-       :
        );
 #elif   defined(TARGET_LINUX_POWER)
     asm("popcntw    %0, %1"
        : "=r"(r32)
        : "r"(r32)
-       :
        );
 #else
 
diff --git a/runtime/flang/ipoppar.c b/runtime/flang/ipoppar.c
@@ -27,14 +27,12 @@ __mth_i_ipoppar(uint32_t u32)
         "\tandl $0x1, %0"
        : "=r"(r32)
        : "r"(u32)
-       :
        );
 #elif   defined(TARGET_LINUX_POWER)
     asm("popcntw    %0, %1\n"
         "\trldicl   %0, %0, 0, 63"
        : "=r"(r32)
        : "r"(u32)
-       :
        );
 #else
   r32 = u32;
diff --git a/runtime/flang/ipoppari.c b/runtime/flang/ipoppari.c
@@ -34,14 +34,12 @@ __mth_i_ipoppari(uint32_t u32, int32_t size)
         "\tandl $0x1, %0"
        : "=r"(r32)
        : "r"(r32)
-       :
        );
 #elif   defined(TARGET_LINUX_POWER)
     asm("popcntw    %0, %1\n"
         "\trldicl   %0, %0, 0, 63"
        : "=r"(r32)
        : "r"(r32)
-       :
        );
 #else
   r32 ^= r32 >> 16;
diff --git a/runtime/flang/kpopcnt.c b/runtime/flang/kpopcnt.c
@@ -26,13 +26,11 @@ __mth_i_kpopcnt(uint64_t u64)
   asm("popcnt %1, %0"
      : "=r"(r64)
      : "r"(u64)
-     :
      );
 #elif   defined(TARGET_LINUX_POWER)
     asm("popcntd    %0, %1"
        : "=r"(r64)
        : "r"(u64)
-       :
        );
 #else
   static const uint64_t u5s = 0x5555555555555555ul;
diff --git a/runtime/flang/kpoppar.c b/runtime/flang/kpoppar.c
@@ -27,14 +27,12 @@ __mth_i_kpoppar(uint64_t u64)
         "\tandq $0x1, %0"
        : "=r"(r64)
        : "r"(u64)
-       :
        );
 #elif   defined(TARGET_LINUX_POWER)
     asm("popcntd    %0, %1\n"
         "\trldicl   %0, %0, 0, 63"
        : "=r"(r64)
        : "r"(u64)
-       :
        );
 #else
   r64 = u64;
diff --git a/test/f90_correct/inc/fs11.mk b/test/f90_correct/inc/fs11.mk
@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2015-2018, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2015-2019, 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.
@@ -18,8 +18,13 @@
 
 # Determine call instruction used
 INSN=call
+LLFLAG=
+OPT=
 ifeq ($(findstring aarch64, $(UNAME)), aarch64)
     INSN=bl
+ifeq ($(findstring O0, $(OPT)), O0)
+    LLFLAG=-Mkeepll
+endif
 endif
 ifeq ($(findstring ppc64le, $(UNAME)), ppc64le)
     INSN=bl
@@ -31,7 +36,7 @@ build:  $(SRC)/fs11.f90
 	-$(RM) fs11.$(EXESUFFIX) core *.d *.mod FOR*.DAT FTN* ftn* fort.*
 	@echo ------------------------------------ building test $@
 	-$(CC) -c $(CFLAGS) $(SRC)/check.c -o check.$(OBJX)
-	-$(FC) $(FFLAGS) $(LDFLAGS) $(SRC)/fs11.f90 -S
+	-$(FC) $(FFLAGS) $(LLFLAG) $(LDFLAGS) $(SRC)/fs11.f90 -S
 	-$(FC) -c $(FFLAGS) $(LDFLAGS) $(SRC)/fs11.f90 -o fs11.$(OBJX)
 	-$(FC) $(FFLAGS) $(LDFLAGS) fs11.$(OBJX) check.$(OBJX) $(LIBS) -o fs11.$(EXESUFFIX)
 
@@ -40,12 +45,24 @@ build:  $(SRC)/fs11.f90
 # This check isn't valid for flang because it allows LLVM to inline.
 run:
 	@echo ------------------------------------ executing test fs11
+ifeq ($(LLFLAG), -Mkeepll)
+	@mmul_calls=`grep -c 'void (.*)\*.*mmul' fs11.ll`; \
+	if [ $$mmul_calls -ne 3 ]; then \
+	  echo "RESULT: FAIL - expected exactly 3 calls to mmul, got $$mmul_calls in fs11.ll" ; \
+	  exit 1; \
+	else \
+	  echo "RESULT: PASS - got exactly 3 calls to mmul in fs11.ll" ; \
+	fi
+else
 ifneq ($(FC), flang)
 	@mmul_calls=`grep -c '$(INSN).*mmul' fs11.s`; \
 	if [ $$mmul_calls -ne 3 ]; then \
-	    echo "RESULT: FAIL - expected exactly 3 calls to mmul, got $$mmul_calls" ; \
+	    echo "RESULT: FAIL - expected exactly 3 calls to mmul, got $$mmul_calls in fs11.s" ; \
 	    exit 1; \
+	else \
+	  echo "RESULT: PASS - got exactly 3 calls to mmul in fs11.s" ; \
 	fi
+endif
 endif
 	fs11.$(EXESUFFIX)
 
diff --git a/tools/flang1/flang1exe/semsmp.c b/tools/flang1/flang1exe/semsmp.c
@@ -1921,6 +1921,11 @@ semsmp(int rednum, SST *top)
     clause_errchk((BT_TARGET | BT_SIMD), "OMP TARGET SIMD");
     mp_create_bscope(0);
     DI_BTARGET(sem.doif_depth) = emit_btarget(A_MP_TARGET);
+#if defined(OMP_OFFLOAD_LLVM) || defined(OMP_OFFLOAD_PGI)
+    if(flg.omptarget)
+      A_COMBINEDTYPEP(DI_BTARGET(sem.doif_depth),
+                      get_omp_combined_mode(BT_TARGET | BT_SIMD));
+#endif
     par_push_scope(TRUE);
     begin_parallel_clause(sem.doif_depth);
     SST_ASTP(LHS, 0);
@@ -10220,6 +10225,9 @@ get_omp_combined_mode(BIGINT64 type)
   combined_type = BT_TARGET | BT_PARDO | BT_SIMD;
   if ((type & combined_type) == combined_type)
     return mode_target_parallel_for_simd;
+  combined_type = BT_TARGET | BT_SIMD;
+  if ((type & combined_type) == combined_type)
+    return mode_target_simd;
   if ((type & BT_TARGET))
     return mode_target;
   return mode_none_target;
diff --git a/tools/flang2/docs/xflag.n b/tools/flang2/docs/xflag.n
@@ -3863,6 +3863,8 @@ Allow some formatted print statements in CUDA Fortran, EXPERIMENTAL
 Don't allow statements between the DO loops of a cuf kernels do construct
 .XB 0x40000:
 reserved
+.XB 0x80000:
+reserved
 
 .XF "138:"
 vect prefetch limit
diff --git a/tools/flang2/flang2exe/exp_rte.cpp b/tools/flang2/flang2exe/exp_rte.cpp
@@ -3277,6 +3277,7 @@ exp_call(ILM_OP opc, ILM *ilmp, int curilm)
   int retdesc;
   int struct_tmp;
   int chain_pointer_arg = 0;
+  int result_arg = 0;
 
   nargs = ILM_OPND(ilmp, 1); /* # args */
   func_addr = 0;
@@ -3632,6 +3633,8 @@ exp_call(ILM_OP opc, ILM *ilmp, int curilm)
       ilm1 = IILM_OPND(ilm1, 1);
     }
     gargili = ILM_RESULT(ilm1);
+    if (!result_arg)
+      result_arg = gargili;
     ilmlnk = (ILM *)(ilmb.ilm_base + ilm1);
     /* ilmlnk is ith argument */
     switch (argopc = ILM_OPC(ilmlnk)) {
@@ -4157,8 +4160,17 @@ exp_call(ILM_OP opc, ILM *ilmp, int curilm)
         /* already processed the procedure descriptor argument in this case */
         continue;
       }
-      ilix = ad4ili(IL_GARG, garg_ili[gi].ilix, gargl, garg_ili[gi].dtype,
-                    garg_ili[gi].val_flag);
+
+      // IM_CALL falls in the default case and a BIND(C) func returning struct
+      // should use IL_GARGRET.
+      if (opc == IM_CALL && CSTRUCTRETG(exp_call_sym) &&
+          garg_ili[gi].ilix == result_arg) {
+        ilix = ad4ili(IL_GARGRET, garg_ili[gi].ilix, gargl, garg_ili[gi].dtype,
+                      garg_ili[gi].val_flag);
+      } else {
+        ilix = ad4ili(IL_GARG, garg_ili[gi].ilix, gargl, garg_ili[gi].dtype,
+                      garg_ili[gi].val_flag);
+      }
       gargl = ilix;
     }
     if (gfch_addr) {
diff --git a/tools/flang2/flang2exe/expsmp.cpp b/tools/flang2/flang2exe/expsmp.cpp
@@ -959,7 +959,7 @@ exp_smp(ILM_OP opc, ILM *ilmp, int curilm)
   static SPTR single_thread;
   static SPTR in_single;
   static SPTR targetfunc_sptr = SPTR_NULL, targetdevice_func_sptr = SPTR_NULL;
-  int target_mode = 0;
+  static int target_mode = 0;
   SPTR nlower, nupper, nstride;
 #if defined(OMP_OFFLOAD_LLVM) || defined(OMP_OFFLOAD_PGI)
   static int target_ili_num_threads = 0;
diff --git a/tools/flang2/flang2exe/ll_structure.h b/tools/flang2/flang2exe/ll_structure.h
@@ -386,6 +386,15 @@ ll_feature_debug_info_ver80(const LL_IRFeatures *feature)
   return feature->version >= LL_Version_8_0;
 }
 
+/**
+   \brief Version 9.0 debug metadata
+ */
+INLINE static bool
+ll_feature_debug_info_ver90(const LL_IRFeatures *feature)
+{
+  return feature->version >= LL_Version_9_0;
+}
+
 /**
    \brief Version 9.0 onwards uses 3 field syntax for constructors
    and destructors
@@ -484,6 +493,7 @@ ll_feature_no_file_in_namespace(const LL_IRFeatures *feature)
 #define ll_feature_debug_info_ver38(f) ((f)->version >= LL_Version_3_8)
 #define ll_feature_debug_info_ver70(f) ((f)->version >= LL_Version_7_0)
 #define ll_feature_debug_info_ver80(f) ((f)->version >= LL_Version_8_0)
+#define ll_feature_debug_info_ver90(f) ((f)->version >= LL_Version_9_0)
 #define ll_feature_three_argument_ctor_and_dtor(f) \
   ((f)->version >= LL_Version_9_0)
 #define ll_feature_use_distinct_metadata(f) ((f)->version >= LL_Version_3_8)
diff --git a/tools/flang2/flang2exe/ll_write.cpp b/tools/flang2/flang2exe/ll_write.cpp
@@ -1052,6 +1052,28 @@ static const MDTemplate Tmpl_DISubprogram[] = {
   { "scopeLine",                UnsignedField }
 };
 
+static const MDTemplate Tmpl_DISubprogram_90[] = {
+  { "DISubprogram", TF, 18 },
+  { "tag",                      DWTagField, FlgHidden },
+  { "file",                     NodeField },
+  { "scope",                    NodeField },
+  { "name",                     StringField },
+  { "displayName",              StringField, FlgHidden },
+  { "linkageName",              StringField },
+  { "line",                     UnsignedField },
+  { "type",                     NodeField },
+  { "virtuality",               DWVirtualityField },
+  { "virtualIndex",             UnsignedField },
+  { "containingType",           NodeField },
+  { "flags",                    UnsignedField }, /* TBD: DIFlag... */
+  { "spFlags",                  UnsignedField }, /* TBD: DISPFlag... */
+  { "function",                 ValueField, FlgHidden },
+  { "templateParams",           NodeField },
+  { "declaration",              NodeField },
+  { "unit",                     NodeField },
+  { "scopeLine",                UnsignedField }
+};
+
 static const MDTemplate Tmpl_DISubprogram_70[] = {
   { "DISubprogram", TF, 20 },
   { "tag",                      DWTagField, FlgHidden },
@@ -2009,6 +2031,12 @@ emitDISubprogram(FILE *out, LLVMModuleRef mod, const LL_MDNode *mdnode,
                  unsigned mdi)
 {
   if (!ll_feature_debug_info_pre34(&mod->ir)) {
+    if (ll_feature_debug_info_ver90(&mod->ir)) {
+      // 9.0, 'isLocal:', 'isDefinition:', and 'isOptimized:' removed
+      // and 'spFlags:' added
+      emitTmpl(out, mod, mdnode, mdi, Tmpl_DISubprogram_90);
+      return;
+    }
     if (ll_feature_debug_info_ver70(&mod->ir)) {
       // 7.0, 'variables:' was removed
       emitTmpl(out, mod, mdnode, mdi, Tmpl_DISubprogram_70);
diff --git a/tools/flang2/flang2exe/lldebug.cpp b/tools/flang2/flang2exe/lldebug.cpp
diff --git a/tools/flang2/flang2exe/tgtutil.cpp b/tools/flang2/flang2exe/tgtutil.cpp
diff --git a/tools/shared/llmputil.c b/tools/shared/llmputil.c
diff --git a/tools/shared/llmputil.h b/tools/shared/llmputil.h
