[PATCH] D29660: [OpenMP] Add flag for overwriting default PTX version for OpenMP targets

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

gtbercea updated this revision to Diff 95126.
gtbercea added a comment.

Fix.


Repository:
  rL LLVM

https://reviews.llvm.org/D29660

Files:
  include/clang/Driver/Options.td
  lib/Driver/ToolChains/Cuda.cpp
  test/Driver/openmp-offload.c


Index: test/Driver/openmp-offload.c
===================================================================
--- test/Driver/openmp-offload.c
+++ test/Driver/openmp-offload.c
@@ -634,6 +634,16 @@
 
 /// ###########################################################################
 
+/// Check PTXAS is passed the compute capability passed to the driver.
+// RUN:   %clang -### -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda --fopenmp-ptx=+ptx52 -save-temps -no-canonical-prefixes %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-PTXAS-VERSION %s
+
+// CHK-PTXAS-VERSION: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+// CHK-PTXAS-VERSION-NEXT: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+// CHK-PTXAS-VERSION-NEXT: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+
+/// ###########################################################################
+
 /// Check -fopenmp-is-device is also passed when generating the *.i and *.s intermediate files.
 // RUN:   %clang -### -fopenmp=libomp -fopenmp-targets=powerpc64le-ibm-linux-gnu -save-temps -no-canonical-prefixes %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-FOPENMP-IS-DEVICE %s
Index: lib/Driver/ToolChains/Cuda.cpp
===================================================================
--- lib/Driver/ToolChains/Cuda.cpp
+++ lib/Driver/ToolChains/Cuda.cpp
@@ -500,7 +500,12 @@
   // than LLVM defaults to. Use PTX4.2 which is the PTX version that
   // came with CUDA-7.0.
   CC1Args.push_back("-target-feature");
-  CC1Args.push_back("+ptx42");
+
+  if (DeviceOffloadingKind == Action::OFK_OpenMP)
+    CC1Args.push_back(
+        DriverArgs.getLastArgValue(options::OPT_fopenmp_ptx_EQ, "+ptx42").data());
+  else
+    CC1Args.push_back("+ptx42");
 
   if (DeviceOffloadingKind == Action::OFK_OpenMP) {
     SmallVector<std::string, 8> LibraryPaths;
Index: include/clang/Driver/Options.td
===================================================================
--- include/clang/Driver/Options.td
+++ include/clang/Driver/Options.td
@@ -529,6 +529,8 @@
   HelpText<"CUDA installation path">;
 def fopenmp_cuda_gpu_arch_EQ : Joined<["--"], "fopenmp-cuda-gpu-arch=">, Flags<[DriverOption]>,
   HelpText<"Pass a single CUDA GPU architecture (default sm_20) to be used by OpenMP device offloading.">;
+def fopenmp_ptx_EQ : Joined<["--"], "fopenmp-ptx=">, Flags<[DriverOption]>,
+  HelpText<"Pass a PTX version +ptxXX, default +ptx42 (for PTX version 4.2) used by OpenMP device offloading.">;
 def ptxas_path_EQ : Joined<["--"], "ptxas-path=">, Group<i_Group>,
   HelpText<"Path to ptxas (used for compiling CUDA code)">;
 def fcuda_flush_denormals_to_zero : Flag<["-"], "fcuda-flush-denormals-to-zero">,


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D29660.95126.patch
Type: text/x-patch
Size: 2664 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20170413/c66c2811/attachment.bin>