[Clang] Make --lto-partitions only default for HIP

[jhuber6]

Summary:
The default behavior for LTO on other targets does not specify the
number of LTO partitions. Recent changes made this default to 8 on
AMDGPU which had some issues with the libc project. The option to
disable this is HIP only so I think for now we should restrict this just
to HIP.

I'm definitely on board with getting some more parallelism here, but I
think it should probably be restricted to just offloading languages. The
new driver goes through the --target=amdgcn-amd-amdhsa for its output,
which means we'd need to forward the default somehow.

[llvmbot]

@llvm/pr-subscribers-libc
@llvm/pr-subscribers-clang

@llvm/pr-subscribers-clang-driver

Author: Joseph Huber (jhuber6)

Changes

Summary:
The default behavior for LTO on other targets does not specify the
number of LTO partitions. Recent changes made this default to 8 on
AMDGPU which had some issues with the libc project. The option to
disable this is HIP only so I think for now we should restrict this just
to HIP.

I'm definitely on board with getting some more parallelism here, but I
think it should probably be restricted to just offloading languages. The
new driver goes through the --target=amdgcn-amd-amdhsa for its output,
which means we'd need to forward the default somehow.

---

Full diff: https://github.com/llvm/llvm-project/pull/133164.diff

3 Files Affected:

• (modified) clang/lib/Driver/ToolChains/AMDGPU.cpp (+8-14)
• (modified) clang/test/Driver/amdgpu-toolchain.c (+1-15)
• (modified) libc/startup/gpu/CMakeLists.txt (+2-8)

diff --git a/clang/lib/Driver/ToolChains/AMDGPU.cpp b/clang/lib/Driver/ToolChains/AMDGPU.cpp
index e919f4e941f47..c754cae2f96b2 100644
--- a/clang/lib/Driver/ToolChains/AMDGPU.cpp
+++ b/clang/lib/Driver/ToolChains/AMDGPU.cpp
@@ -625,6 +625,7 @@ void amdgpu::Linker::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back("-shared");
   }
 
+  llvm::errs() << JA.getOffloadingDeviceKind() << "\n";
   addLinkerCompressDebugSectionsOption(getToolChain(), Args, CmdArgs);
   Args.AddAllArgs(CmdArgs, options::OPT_L);
   getToolChain().AddFilePathLibArgs(Args, CmdArgs);
@@ -633,7 +634,7 @@ void amdgpu::Linker::ConstructJob(Compilation &C, const JobAction &JA,
     const bool ThinLTO = (C.getDriver().getLTOMode() == LTOK_Thin);
     addLTOOptions(getToolChain(), Args, CmdArgs, Output, Inputs[0], ThinLTO);
 
-    if (!ThinLTO)
+    if (!ThinLTO && JA.getOffloadingDeviceKind() == Action::OFK_HIP)
       addFullLTOPartitionOption(C.getDriver(), Args, CmdArgs);
   } else if (Args.hasArg(options::OPT_mcpu_EQ)) {
     CmdArgs.push_back(Args.MakeArgString(
@@ -712,14 +713,12 @@ void amdgpu::getAMDGPUTargetFeatures(const Driver &D,
 }
 
 static unsigned getFullLTOPartitions(const Driver &D, const ArgList &Args) {
-  const Arg *A = Args.getLastArg(options::OPT_flto_partitions_EQ);
-  // In the absence of an option, use 8 as the default.
-  if (!A)
-    return 8;
   int Value = 0;
-  if (StringRef(A->getValue()).getAsInteger(10, Value) || (Value < 1)) {
+  StringRef A = Args.getLastArgValue(options::OPT_flto_partitions_EQ, "8");
+  if (A.getAsInteger(10, Value) || (Value < 1)) {
+    Arg *Arg = Args.getLastArg(options::OPT_flto_partitions_EQ);
     D.Diag(diag::err_drv_invalid_int_value)
-        << A->getAsString(Args) << A->getValue();
+        << Arg->getAsString(Args) << Arg->getValue();
     return 1;
   }
 
@@ -729,13 +728,8 @@ static unsigned getFullLTOPartitions(const Driver &D, const ArgList &Args) {
 void amdgpu::addFullLTOPartitionOption(const Driver &D,
                                        const llvm::opt::ArgList &Args,
                                        llvm::opt::ArgStringList &CmdArgs) {
-  // TODO: Should this be restricted to fgpu-rdc only ? Currently we'll
-  //       also do it for non gpu-rdc LTO
-
-  if (unsigned NumParts = getFullLTOPartitions(D, Args); NumParts > 1) {
-    CmdArgs.push_back(
-        Args.MakeArgString("--lto-partitions=" + Twine(NumParts)));
-  }
+  CmdArgs.push_back(Args.MakeArgString("--lto-partitions=" +
+                                       Twine(getFullLTOPartitions(D, Args))));
 }
 
 /// AMDGPU Toolchain
diff --git a/clang/test/Driver/amdgpu-toolchain.c b/clang/test/Driver/amdgpu-toolchain.c
index 6617108e59fcf..20656f1fadeb7 100644
--- a/clang/test/Driver/amdgpu-toolchain.c
+++ b/clang/test/Driver/amdgpu-toolchain.c
@@ -21,7 +21,7 @@
 // RUN: %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx90a:xnack+:sramecc- -nogpulib \
 // RUN:   -L. -flto -fconvergent-functions %s 2>&1 | FileCheck -check-prefix=LTO %s
 // LTO: clang{{.*}} "-flto=full"{{.*}}"-fconvergent-functions"
-// LTO: ld.lld{{.*}}"-L."{{.*}}"-plugin-opt=mcpu=gfx90a"{{.*}}"--lto-partitions={{[0-9]+}}"{{.*}}"-plugin-opt=-mattr=-sramecc,+xnack"
+// LTO: ld.lld{{.*}}"-L."{{.*}}"-plugin-opt=mcpu=gfx90a"{{.*}}"-plugin-opt=-mattr=-sramecc,+xnack"
 
 // RUN: %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx90a:xnack+:sramecc- -nogpulib \
 // RUN:   -L. -fconvergent-functions %s 2>&1 | FileCheck -check-prefix=MCPU %s
@@ -38,17 +38,3 @@
 // RUN: %clang -target amdgcn-amd-amdhsa -march=gfx90a -stdlib -startfiles \
 // RUN:   -nogpulib -nogpuinc -### %s 2>&1 | FileCheck -check-prefix=STARTUP %s
 // STARTUP: ld.lld{{.*}}"-lc" "-lm" "{{.*}}crt1.o"
-
-// Check --flto-partitions
-
-// RUN: %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx90a -nogpulib \
-// RUN:   -L. -flto --flto-partitions=42 %s 2>&1 | FileCheck -check-prefix=LTO_PARTS %s
-// LTO_PARTS: ld.lld{{.*}}"-L."{{.*}}"-plugin-opt=mcpu=gfx90a"{{.*}}"--lto-partitions=42"
-
-// RUN: not %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx90a -nogpulib \
-// RUN:   -L. -flto --flto-partitions=a %s 2>&1 | FileCheck -check-prefix=LTO_PARTS_INV0 %s
-// LTO_PARTS_INV0: clang: error: invalid integral value 'a' in '--flto-partitions=a'
-
-// RUN: not %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx90a -nogpulib \
-// RUN:   -L. -flto --flto-partitions=0 %s 2>&1 | FileCheck -check-prefix=LTO_PARTS_INV1 %s
-// LTO_PARTS_INV1: clang: error: invalid integral value '0' in '--flto-partitions=0'
diff --git a/libc/startup/gpu/CMakeLists.txt b/libc/startup/gpu/CMakeLists.txt
index a30e6f07a66a3..fa326ef46a9d1 100644
--- a/libc/startup/gpu/CMakeLists.txt
+++ b/libc/startup/gpu/CMakeLists.txt
@@ -33,14 +33,8 @@ function(add_startup_object name)
     set_target_properties(${fq_target_name}.exe PROPERTIES
       RUNTIME_OUTPUT_DIRECTORY ${LIBC_LIBRARY_DIR}
       RUNTIME_OUTPUT_NAME ${name}.o)
-    # FIXME: A bug in the AMDGPU LTO pass is incorrectly removing the kernels.
-    if(LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
-      target_link_options(${fq_target_name}.exe PRIVATE
-                          "-r" "-nostdlib" "-flto" "-Wl,--lto-emit-llvm")
-    else()
-      target_link_options(${fq_target_name}.exe PRIVATE
-                          "-r" "-nostdlib" "-Wl,--lto-emit-llvm")
-    endif()
+    target_link_options(${fq_target_name}.exe PRIVATE
+                        "-r" "-nostdlib" "-flto" "-Wl,--lto-emit-llvm")
   endif()
 endfunction()
 

[shiltian]

unrelated?

[jhuber6]

Related, this works around what this patch disables.

[shiltian]

can you move these checks to a hip file then?

[jhuber6]

I'll clean up the option later as well. --flto with double dashes is wrong.