https://github.com/GeorgeWeb updated
https://github.com/llvm/llvm-project/pull/65327:
>From 757383fd9a9783c290fd4957e4e7526b854c9920 Mon Sep 17 00:00:00 2001
From: Georgi Mirazchiyski <georgi.mirazchiy...@codeplay.com>
Date: Tue, 5 Sep 2023 14:21:24 +0100
Subject: [PATCH] [Driver][NVPTX] Add a warning that device debug info does not
work with optimizations
Passing -g to ptxas with any optimizations enabled is not allowed because,
ptxas does not support optimized debugging.
---
clang/lib/Driver/ToolChains/Cuda.cpp | 14 +++++++++++++-
clang/test/Driver/cuda-external-tools.cu | 15 +++++++++++++++
2 files changed, 28 insertions(+), 1 deletion(-)
diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp
b/clang/lib/Driver/ToolChains/Cuda.cpp
index 97472a302715bb..3f76aa49822c26 100644
--- a/clang/lib/Driver/ToolChains/Cuda.cpp
+++ b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -413,13 +413,25 @@ void NVPTX::Assembler::ConstructJob(Compilation &C, const
JobAction &JA,
// TODO: Perhaps we should map host -O2 to ptxas -O3. -O3 is ptxas's
// default, so it may correspond more closely to the spirit of clang -O2.
+ bool noOptimization = A->getOption().matches(options::OPT_O0);
+ // Emit a driver diagnostic as warning if any -O option different from -O0,
+ // is passed since ptxas does not support optimized debugging.
+ if (!noOptimization) {
+ auto &Diags = TC.getDriver().getDiags();
+ unsigned DiagID = Diags.getCustomDiagID(
+ DiagnosticsEngine::Warning,
+ "ptxas cannot emit debug info with optimization "
+ "level ('%0') different than O0.");
+ Diags.Report(DiagID) << A->getAsString(Args);
+ }
+
// -O3 seems like the least-bad option when -Osomething is specified to
// clang but it isn't handled below.
StringRef OOpt = "3";
if (A->getOption().matches(options::OPT_O4) ||
A->getOption().matches(options::OPT_Ofast))
OOpt = "3";
- else if (A->getOption().matches(options::OPT_O0))
+ else if (noOptimization)
OOpt = "0";
else if (A->getOption().matches(options::OPT_O)) {
// -Os, -Oz, and -O(anything else) map to -O2, for lack of better
options.
diff --git a/clang/test/Driver/cuda-external-tools.cu
b/clang/test/Driver/cuda-external-tools.cu
index 1aa87cc09982c6..c49ba51b660ef5 100644
--- a/clang/test/Driver/cuda-external-tools.cu
+++ b/clang/test/Driver/cuda-external-tools.cu
@@ -28,6 +28,17 @@
// RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
// RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,RDC %s
+// Compiling -O{1,2,3,4,fast,s,z} with -g does not pass -g debug info to ptxas.
+// NOTE: This is because ptxas does not support optimized debugging.
+// RUN: %clang -### --target=x86_64-linux-gnu -O3 -g -c %s 2>&1 \
+// RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
+// RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT3-DBG %s
+
+// Compiling -O0 with -g passes -g debug info to ptxas.
+// RUN: %clang -### --target=x86_64-linux-gnu -O0 -g -c %s 2>&1 \
+// RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
+// RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT0-DBG %s
+
// With debugging enabled, ptxas should be run with with no ptxas
optimizations.
// RUN: %clang -### --target=x86_64-linux-gnu --cuda-noopt-device-debug -O2 -g
-c %s 2>&1 \
// RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
@@ -134,6 +145,10 @@
// OPT2-NOT: "-g"
// OPT3-SAME: "-O3"
// OPT3-NOT: "-g"
+// OPT3-DBG-SAME: "-O3" "-lineinfo"
+// OPT3-DBG-NOT: "-g"
+// OPT0-DBG-SAME: "-g" "--dont-merge-basicblocks" "--return-at-end"
+// OPT0-DBG-NOT: "-O0"
// DBG-SAME: "-g" "--dont-merge-basicblocks" "--return-at-end"
// SM35-SAME: "--gpu-name" "sm_35"
// SM35-SAME: "--output-file" "[[CUBINFILE:[^"]*]]"
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
