(cherry-pick) Fixes and Improvements for -flto-partitions option (llvm#1416)

Pierre-vh · web-flow · commit 775d5f3740f7 · 2025-04-10T15:37:49.000+02:00
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
@@ -1458,8 +1458,6 @@ def fhip_emit_relocatable : Flag<["-"], "fhip-emit-relocatable">,
   HelpText<"Compile HIP source to relocatable">;
 def fno_hip_emit_relocatable : Flag<["-"], "fno-hip-emit-relocatable">,
   HelpText<"Do not override toolchain to compile HIP source to relocatable">;
-def flto_partitions_EQ : Joined<["--"], "flto-partitions=">, Group<hip_Group>,
-  HelpText<"Number of partitions to use for parallel full LTO codegen. Use 1 to disable partitioning.">;
 }
 
 // Clang specific/exclusive options for OpenACC.
@@ -3002,6 +3000,8 @@ defm fat_lto_objects : BoolFOption<"fat-lto-objects",
   PosFlag<SetTrue, [], [ClangOption, CC1Option], "Enable">,
   NegFlag<SetFalse, [], [ClangOption, CC1Option], "Disable">,
   BothFlags<[], [ClangOption, CC1Option], " fat LTO object support">>;
+def flto_partitions_EQ : Joined<["-"], "flto-partitions=">, Group<f_Group>,
+  HelpText<"Number of partitions to use for parallel full LTO codegen, ld.lld only.">;
 def fmacro_backtrace_limit_EQ : Joined<["-"], "fmacro-backtrace-limit=">,
   Group<f_Group>, Visibility<[ClangOption, CC1Option, CLOption]>,
   HelpText<"Set the maximum number of entries to print in a macro expansion backtrace (0 = no limit)">,
diff --git a/clang/lib/Driver/ToolChains/AMDGPU.cpp b/clang/lib/Driver/ToolChains/AMDGPU.cpp
@@ -624,19 +624,18 @@ void amdgpu::Linker::ConstructJob(Compilation &C, const JobAction &JA,
   CmdArgs.push_back("--no-undefined");
   CmdArgs.push_back("-shared");
 
-  addLinkerCompressDebugSectionsOption(getToolChain(), Args, CmdArgs);
-  Args.AddAllArgs(CmdArgs, options::OPT_L);
-  getToolChain().AddFilePathLibArgs(Args, CmdArgs);
-  AddLinkerInputs(getToolChain(), Inputs, Args, CmdArgs, JA);
   if (C.getDriver().isUsingLTO()) {
     const bool ThinLTO = (C.getDriver().getLTOMode() == LTOK_Thin);
     addLTOOptions(getToolChain(), Args, CmdArgs, Output, Inputs[0], ThinLTO);
-
-    if (!ThinLTO)
-      addFullLTOPartitionOption(C.getDriver(), Args, CmdArgs);
   } else if (Args.hasArg(options::OPT_mcpu_EQ))
     CmdArgs.push_back(Args.MakeArgString(
         "-plugin-opt=mcpu=" + Args.getLastArgValue(options::OPT_mcpu_EQ)));
+
+  addLinkerCompressDebugSectionsOption(getToolChain(), Args, CmdArgs);
+  Args.AddAllArgs(CmdArgs, options::OPT_L);
+  getToolChain().AddFilePathLibArgs(Args, CmdArgs);
+  AddLinkerInputs(getToolChain(), Inputs, Args, CmdArgs, JA);
+
   CmdArgs.push_back("-o");
   CmdArgs.push_back(Output.getFilename());
   C.addCommand(std::make_unique<Command>(
@@ -645,14 +644,12 @@ void amdgpu::Linker::ConstructJob(Compilation &C, const JobAction &JA,
 }
 
 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;
   }
 
@@ -662,13 +659,8 @@ 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))));
 }
 
 void amdgpu::getAMDGPUTargetFeatures(const Driver &D,
diff --git a/clang/lib/Driver/ToolChains/CommonArgs.cpp b/clang/lib/Driver/ToolChains/CommonArgs.cpp
@@ -863,6 +863,17 @@ void tools::addLTOOptions(const ToolChain &ToolChain, const ArgList &Args,
     // files
     if (IsFatLTO)
       CmdArgs.push_back("--fat-lto-objects");
+
+    if (Args.hasArg(options::OPT_flto_partitions_EQ)) {
+      int Value = 0;
+      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)
+            << Arg->getAsString(Args) << Arg->getValue();
+      }
+      CmdArgs.push_back(Args.MakeArgString("--lto-partitions=" + A));
+    }
   }
 
   const char *PluginOptPrefix = IsOSAIX ? "-bplugin_opt:" : "-plugin-opt=";
diff --git a/clang/lib/Driver/ToolChains/HIPAMD.cpp b/clang/lib/Driver/ToolChains/HIPAMD.cpp
@@ -160,8 +160,6 @@ void AMDGCN::Linker::constructLldCommand(Compilation &C, const JobAction &JA,
 
   addLinkerCompressDebugSectionsOption(TC, Args, LldArgs);
 
-  amdgpu::addFullLTOPartitionOption(D, Args, LldArgs);
-
   // Given that host and device linking happen in separate processes, the device
   // linker doesn't always have the visibility as to which device symbols are
   // needed by a program, especially for the device symbol dependencies that are
@@ -355,6 +353,10 @@ HIPAMDToolChain::TranslateArgs(const llvm::opt::DerivedArgList &Args,
     checkTargetID(*DAL);
   }
 
+  if (!Args.hasArg(options::OPT_flto_partitions_EQ))
+    DAL->AddJoinedArg(nullptr, Opts.getOption(options::OPT_flto_partitions_EQ),
+                      "8");
+
   return DAL;
 }
 
diff --git a/clang/test/Driver/amdgpu-toolchain.c b/clang/test/Driver/amdgpu-toolchain.c
@@ -20,26 +20,13 @@
 
 // RUN: %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx906 -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=gfx906"{{.*}}"--lto-partitions={{[0-9]+}}"
+// LTO: clang{{.*}}"-flto=full"{{.*}}"-fconvergent-functions"
+// LTO: ld.lld{{.*}}"-plugin-opt=mcpu=gfx906"{{.*}}"{{.*}}
+
 // RUN: %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx906 -nogpulib \
 // RUN:   -L. -fconvergent-functions %s 2>&1 | FileCheck -check-prefix=MCPU %s
-// MCPU: ld.lld{{.*}}"-L."{{.*}}"-plugin-opt=mcpu=gfx906"
+// MCPU: ld.lld{{.*}}"-plugin-opt=mcpu=gfx906"{{.*}}
 
 // RUN: %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx906 -nogpulib \
 // RUN:   -fuse-ld=ld %s 2>&1 | FileCheck -check-prefixes=LD %s
 // LD: ld.lld
-
-// 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/clang/test/Driver/hip-toolchain-rdc-flto-partitions.hip b/clang/test/Driver/hip-toolchain-rdc-flto-partitions.hip
@@ -1,5 +1,5 @@
 // RUN: %clang -### --target=x86_64-linux-gnu \
-// RUN:   -x hip --cuda-gpu-arch=gfx803 --flto-partitions=42 \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 -flto-partitions=42 \
 // RUN:   --no-offload-new-driver --emit-static-lib -nogpulib \
 // RUN:   -fuse-ld=lld -B%S/Inputs/lld -fgpu-rdc -nogpuinc \
 // RUN:   %S/Inputs/hip_multiple_inputs/a.cu \
@@ -10,26 +10,26 @@
 // FIXED-PARTS-NOT: ".*opt"
 // FIXED-PARTS-NOT: ".*llc"
 // FIXED-PARTS: [[LLD: ".*lld.*"]] {{.*}} "-plugin-opt=-amdgpu-internalize-symbols"
-// FIXED-PARTS-SAME: "-plugin-opt=mcpu=gfx803"
 // FIXED-PARTS-SAME: "--lto-partitions=42"
+// FIXED-PARTS-SAME: "-plugin-opt=mcpu=gfx803"
 // FIXED-PARTS-SAME: "-o" "{{.*out}}" "{{.*bc}}"
 
 // RUN: not %clang -### --target=x86_64-linux-gnu \
-// RUN:   -x hip --cuda-gpu-arch=gfx803 --flto-partitions=a \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 -flto-partitions=a \
 // RUN:   --no-offload-new-driver --emit-static-lib -nogpulib \
 // RUN:   -fuse-ld=lld -B%S/Inputs/lld -fgpu-rdc -nogpuinc \
 // RUN:   %S/Inputs/hip_multiple_inputs/a.cu \
 // RUN:   %S/Inputs/hip_multiple_inputs/b.hip \
 // RUN: 2>&1 | FileCheck %s --check-prefix=LTO_PARTS_INV0
 
-// LTO_PARTS_INV0: clang: error: invalid integral value 'a' in '--flto-partitions=a'
+// LTO_PARTS_INV0: clang: error: invalid integral value 'a' in '-flto-partitions=a'
 
 // RUN: not %clang -### --target=x86_64-linux-gnu \
-// RUN:   -x hip --cuda-gpu-arch=gfx803 --flto-partitions=0 \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 -flto-partitions=0 \
 // RUN:   --no-offload-new-driver --emit-static-lib -nogpulib \
 // RUN:   -fuse-ld=lld -B%S/Inputs/lld -fgpu-rdc -nogpuinc \
 // RUN:   %S/Inputs/hip_multiple_inputs/a.cu \
 // RUN:   %S/Inputs/hip_multiple_inputs/b.hip \
 // RUN: 2>&1 | FileCheck %s --check-prefix=LTO_PARTS_INV1
 
-// LTO_PARTS_INV1: clang: error: invalid integral value '0' in '--flto-partitions=0'
+// LTO_PARTS_INV1: clang: error: invalid integral value '0' in '-flto-partitions=0'
diff --git a/clang/test/Driver/hip-toolchain-rdc-static-lib.hip b/clang/test/Driver/hip-toolchain-rdc-static-lib.hip
@@ -48,8 +48,8 @@
 // CHECK-NOT: ".*opt"
 // CHECK-NOT: ".*llc"
 // CHECK: [[LLD: ".*lld.*"]] {{.*}} "-plugin-opt=-amdgpu-internalize-symbols"
-// CHECK-SAME: "-plugin-opt=mcpu=gfx803"
 // CHECK-SAME: "--lto-partitions={{[0-9]+}}"
+// CHECK-SAME: "-plugin-opt=mcpu=gfx803"
 // CHECK-SAME: "-o" "[[IMG_DEV1:.*out]]" [[A_BC1]] [[B_BC1]]
 
 // generate image for device side path on gfx900
@@ -77,8 +77,8 @@
 // CHECK-NOT: ".*opt"
 // CHECK-NOT: ".*llc"
 // CHECK: [[LLD]] {{.*}} "-plugin-opt=-amdgpu-internalize-symbols"
-// CHECK-SAME: "-plugin-opt=mcpu=gfx900"
 // CHECK-SAME: "--lto-partitions={{[0-9]+}}"
+// CHECK-SAME: "-plugin-opt=mcpu=gfx900"
 // CHECK-SAME: "--whole-archive"
 // CHECK-SAME: "-o" "[[IMG_DEV2:.*out]]" [[A_BC2]] [[B_BC2]]
 // CHECK-SAME: "--no-whole-archive"
diff --git a/clang/test/Driver/hip-toolchain-rdc.hip b/clang/test/Driver/hip-toolchain-rdc.hip
@@ -146,8 +146,8 @@
 // CHECK-NOT: ".*opt"
 // CHECK-NOT: ".*llc"
 // CHECK: {{".*lld.*"}} {{.*}} "-plugin-opt=-amdgpu-internalize-symbols"
-// CHECK-SAME: "-plugin-opt=mcpu=gfx900"
 // CHECK-SAME: "--lto-partitions={{[0-9]+}}"
+// CHECK-SAME: "-plugin-opt=mcpu=gfx900"
 // CHECK-SAME: "-o" "[[IMG_DEV2:.*.out]]" [[A_BC2]] [[B_BC2]]
 
 // combine images generated into hip fat binary object
@@ -161,3 +161,21 @@
 // output the executable
 // LNX: [[LD:".*ld.*"]] {{.*}}"-o" "a.out" {{.*}} [[A_OBJ_HOST]] [[B_OBJ_HOST]] [[OBJBUNDLE]]
 // MSVC: [[LD:".*lld-link.*"]] {{.*}}"-out:a.exe" {{.*}} [[A_OBJ_HOST]] [[B_OBJ_HOST]] [[OBJBUNDLE]]
+
+// Check -flto-partitions
+
+// RUN: %clang -### -fgpu-rdc --offload-arch=gfx90a -nogpulib -nogpuinc --no-offload-new-driver \
+// RUN:   -L. -foffload-lto %s 2>&1 | FileCheck -check-prefix=LTO_DEFAULT %s
+// LTO_DEFAULT: lld{{.*}}"--lto-partitions=8"
+
+// RUN: %clang -### -fgpu-rdc --offload-arch=gfx90a -nogpulib -nogpuinc --no-offload-new-driver \
+// RUN:   -L. -foffload-lto -flto-partitions=42 %s 2>&1 | FileCheck -check-prefix=LTO_PARTS %s
+// LTO_PARTS: lld{{.*}}"--lto-partitions=42"
+
+// RUN: not %clang -### -fgpu-rdc --offload-arch=gfx90a -nogpulib -nogpuinc --no-offload-new-driver \
+// RUN:   -L. -foffload-lto -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 -### -fgpu-rdc --offload-arch=gfx90a -nogpulib -nogpuinc --no-offload-new-driver \
+// RUN:   -L. -foffload-lto -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'
