gtbercea created this revision.
This flag "--fopenmp-ptx=" enables the overwriting of the default PTX version
used for GPU offloaded OpenMP target regions: "+ptx42".
Repository:
rL LLVM
https://reviews.llvm.org/D29660
Files:
include/clang/Driver/Options.td
lib/Driver/ToolChains.cpp
test/Driver/openmp-offload.c
Index: test/Driver/openmp-offload.c
===================================================================
--- test/Driver/openmp-offload.c
+++ test/Driver/openmp-offload.c
@@ -627,6 +627,16 @@
/// ###########################################################################
+/// 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-PTX %s
+
+// CHK-PTX: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+// CHK-PTX-NEXT: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+// CHK-PTX-NEXT: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+
+/// ###########################################################################
+
/// Check that CLANG forwards the -v flag to PTXAS.
// RUN: %clang -### -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda
-save-temps -no-canonical-prefixes -v %s 2>&1 \
// RUN: | FileCheck -check-prefix=CHK-VERBOSE %s
Index: lib/Driver/ToolChains.cpp
===================================================================
--- lib/Driver/ToolChains.cpp
+++ lib/Driver/ToolChains.cpp
@@ -4932,8 +4932,18 @@
// than LLVM defaults to. Use PTX4.2 which is the PTX version that
// came with CUDA-7.0.
CC1Args.push_back("-target-feature");
+
+ StringRef PtxVersion = DriverArgs.getLastArgValue(
+ options::OPT_fopenmp_ptx_EQ);
+
CC1Args.push_back("+ptx42");
+ if (DeviceOffloadingKind == Action::OFK_OpenMP &&
+ !PtxVersion.empty()) {
+ // Use PTX version passed to the driver.
+ CC1Args.back() = PtxVersion.data();
+ }
+
if (DeviceOffloadingKind == Action::OFK_OpenMP) {
SmallVector<std::string, 8> LibraryPaths;
if (char *Env = ::getenv("LIBRARY_PATH")) {
Index: include/clang/Driver/Options.td
===================================================================
--- include/clang/Driver/Options.td
+++ include/clang/Driver/Options.td
@@ -490,6 +490,8 @@
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">,
Index: test/Driver/openmp-offload.c
===================================================================
--- test/Driver/openmp-offload.c
+++ test/Driver/openmp-offload.c
@@ -627,6 +627,16 @@
/// ###########################################################################
+/// 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-PTX %s
+
+// CHK-PTX: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+// CHK-PTX-NEXT: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+// CHK-PTX-NEXT: clang{{.*}}.bc" {{.*}}"-target-feature" "+ptx52"
+
+/// ###########################################################################
+
/// Check that CLANG forwards the -v flag to PTXAS.
// RUN: %clang -### -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda -save-temps -no-canonical-prefixes -v %s 2>&1 \
// RUN: | FileCheck -check-prefix=CHK-VERBOSE %s
Index: lib/Driver/ToolChains.cpp
===================================================================
--- lib/Driver/ToolChains.cpp
+++ lib/Driver/ToolChains.cpp
@@ -4932,8 +4932,18 @@
// than LLVM defaults to. Use PTX4.2 which is the PTX version that
// came with CUDA-7.0.
CC1Args.push_back("-target-feature");
+
+ StringRef PtxVersion = DriverArgs.getLastArgValue(
+ options::OPT_fopenmp_ptx_EQ);
+
CC1Args.push_back("+ptx42");
+ if (DeviceOffloadingKind == Action::OFK_OpenMP &&
+ !PtxVersion.empty()) {
+ // Use PTX version passed to the driver.
+ CC1Args.back() = PtxVersion.data();
+ }
+
if (DeviceOffloadingKind == Action::OFK_OpenMP) {
SmallVector<std::string, 8> LibraryPaths;
if (char *Env = ::getenv("LIBRARY_PATH")) {
Index: include/clang/Driver/Options.td
===================================================================
--- include/clang/Driver/Options.td
+++ include/clang/Driver/Options.td
@@ -490,6 +490,8 @@
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">,
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
