[flang] Add -f[no-]vectorize flags (llvm#119718)

This patch adds the -fvectorize and -fno-vectorize flags to flang. 

Note that this also changes the behaviour of `flang -fc1` to match that
of `clang -cc1`, which is that vectorization is only enabled in the
presence of the `-vectorize-loops` flag.

Additionally, this patch changes the behaviour of the default
optimisation levels to match clang, such that vectorization only happens
at the same levels as it does there.

This patch is in draft while I write an RFC to discuss the above two
changes.

Author: DavidTruby

clang/include/clang/Driver/Options.td

@@ -3997,11 +3997,15 @@ defm assumptions : BoolFOption<"assumptions",
           "Disable codegen and compile-time checks for C++23's [[assume]] attribute">,
   PosFlag<SetTrue>>;
 
+
+let Visibility = [ClangOption, FlangOption] in {
 def fvectorize : Flag<["-"], "fvectorize">, Group<f_Group>,
   HelpText<"Enable the loop vectorization passes">;
 def fno_vectorize : Flag<["-"], "fno-vectorize">, Group<f_Group>;
 def : Flag<["-"], "ftree-vectorize">, Alias<fvectorize>;
 def : Flag<["-"], "fno-tree-vectorize">, Alias<fno_vectorize>;
+}
+
 def fslp_vectorize : Flag<["-"], "fslp-vectorize">, Group<f_Group>,
   HelpText<"Enable the superword-level parallelism vectorization passes">;
 def fno_slp_vectorize : Flag<["-"], "fno-slp-vectorize">, Group<f_Group>;
@@ -7343,6 +7347,10 @@ def mlink_builtin_bitcode : Separate<["-"], "mlink-builtin-bitcode">,
 def mlink_bitcode_file
     : Separate<["-"], "mlink-bitcode-file">,
       HelpText<"Link the given bitcode file before performing optimizations.">;
+
+def vectorize_loops : Flag<["-"], "vectorize-loops">,
+  HelpText<"Run the Loop vectorization passes">,
+  MarshallingInfoFlag<CodeGenOpts<"VectorizeLoop">>;
 } // let Visibility = [CC1Option, FC1Option]
 
 let Visibility = [CC1Option] in {
@@ -7458,9 +7466,6 @@ defm link_builtin_bitcode_postopt: BoolMOption<"link-builtin-bitcode-postopt",
   PosFlag<SetTrue, [], [ClangOption], "Link builtin bitcodes after the "
   "optimization pipeline">,
   NegFlag<SetFalse, [], [ClangOption]>>;
-def vectorize_loops : Flag<["-"], "vectorize-loops">,
-  HelpText<"Run the Loop vectorization passes">,
-  MarshallingInfoFlag<CodeGenOpts<"VectorizeLoop">>;
 def vectorize_slp : Flag<["-"], "vectorize-slp">,
   HelpText<"Run the SLP vectorization passes">,
   MarshallingInfoFlag<CodeGenOpts<"VectorizeSLP">>;

clang/lib/Driver/ToolChains/Clang.cpp

@@ -511,39 +511,6 @@ static void addCoveragePrefixMapArg(const Driver &D, const ArgList &Args,
   }
 }
 
-/// Vectorize at all optimization levels greater than 1 except for -Oz.
-/// For -Oz the loop vectorizer is disabled, while the slp vectorizer is
-/// enabled.
-static bool shouldEnableVectorizerAtOLevel(const ArgList &Args, bool isSlpVec) {
-  if (Arg *A = Args.getLastArg(options::OPT_O_Group)) {
-    if (A->getOption().matches(options::OPT_O4) ||
-        A->getOption().matches(options::OPT_Ofast))
-      return true;
-
-    if (A->getOption().matches(options::OPT_O0))
-      return false;
-
-    assert(A->getOption().matches(options::OPT_O) && "Must have a -O flag");
-
-    // Vectorize -Os.
-    StringRef S(A->getValue());
-    if (S == "s")
-      return true;
-
-    // Don't vectorize -Oz, unless it's the slp vectorizer.
-    if (S == "z")
-      return isSlpVec;
-
-    unsigned OptLevel = 0;
-    if (S.getAsInteger(10, OptLevel))
-      return false;
-
-    return OptLevel > 1;
-  }
-
-  return false;
-}
-
 /// Add -x lang to \p CmdArgs for \p Input.
 static void addDashXForInput(const ArgList &Args, const InputInfo &Input,
                              ArgStringList &CmdArgs) {

clang/lib/Driver/ToolChains/CommonArgs.cpp

@@ -3133,3 +3133,36 @@ void tools::renderCommonIntegerOverflowOptions(const ArgList &Args,
   if (use_fwrapv_pointer)
     CmdArgs.push_back("-fwrapv-pointer");
 }
+
+/// Vectorize at all optimization levels greater than 1 except for -Oz.
+/// For -Oz the loop vectorizer is disabled, while the slp vectorizer is
+/// enabled.
+bool tools::shouldEnableVectorizerAtOLevel(const ArgList &Args, bool isSlpVec) {
+  if (Arg *A = Args.getLastArg(options::OPT_O_Group)) {
+    if (A->getOption().matches(options::OPT_O4) ||
+        A->getOption().matches(options::OPT_Ofast))
+      return true;
+
+    if (A->getOption().matches(options::OPT_O0))
+      return false;
+
+    assert(A->getOption().matches(options::OPT_O) && "Must have a -O flag");
+
+    // Vectorize -Os.
+    StringRef S(A->getValue());
+    if (S == "s")
+      return true;
+
+    // Don't vectorize -Oz, unless it's the slp vectorizer.
+    if (S == "z")
+      return isSlpVec;
+
+    unsigned OptLevel = 0;
+    if (S.getAsInteger(10, OptLevel))
+      return false;
+
+    return OptLevel > 1;
+  }
+
+  return false;
+}

clang/lib/Driver/ToolChains/CommonArgs.h

@@ -265,6 +265,8 @@ bool shouldRecordCommandLine(const ToolChain &TC,
 void renderCommonIntegerOverflowOptions(const llvm::opt::ArgList &Args,
                                         llvm::opt::ArgStringList &CmdArgs);
 
+bool shouldEnableVectorizerAtOLevel(const llvm::opt::ArgList &Args,
+                                    bool isSlpVec);
 } // end namespace tools
 } // end namespace driver
 } // end namespace clang

clang/lib/Driver/ToolChains/Flang.cpp

@@ -149,6 +149,16 @@ void Flang::addCodegenOptions(const ArgList &Args,
       !stackArrays->getOption().matches(options::OPT_fno_stack_arrays))
     CmdArgs.push_back("-fstack-arrays");
 
+  // Enable vectorization per default according to the optimization level
+  // selected. For optimization levels that want vectorization we use the alias
+  // option to simplify the hasFlag logic.
+  bool enableVec = shouldEnableVectorizerAtOLevel(Args, false);
+  OptSpecifier vectorizeAliasOption =
+      enableVec ? options::OPT_O_Group : options::OPT_fvectorize;
+  if (Args.hasFlag(options::OPT_fvectorize, vectorizeAliasOption,
+                   options::OPT_fno_vectorize, enableVec))
+    CmdArgs.push_back("-vectorize-loops");
+
   if (shouldLoopVersion(Args))
     CmdArgs.push_back("-fversion-loops-for-stride");
 

flang/include/flang/Frontend/CodeGenOptions.def

@@ -31,6 +31,7 @@ CODEGENOPT(PrepareForFullLTO , 1, 0) ///< Set when -flto is enabled on the
 CODEGENOPT(PrepareForThinLTO , 1, 0) ///< Set when -flto=thin is enabled on the
                                      ///< compile step.
 CODEGENOPT(StackArrays, 1, 0) ///< -fstack-arrays (enable the stack-arrays pass)
+CODEGENOPT(VectorizeLoop, 1, 0) ///< Enable loop vectorization.
 CODEGENOPT(LoopVersioning, 1, 0) ///< Enable loop versioning.
 CODEGENOPT(UnrollLoops, 1, 0) ///< Enable loop unrolling
 CODEGENOPT(AliasAnalysis, 1, 0) ///< Enable alias analysis pass

flang/lib/Frontend/CompilerInvocation.cpp

@@ -23,6 +23,7 @@
 #include "clang/Basic/AllDiagnostics.h"
 #include "clang/Basic/DiagnosticDriver.h"
 #include "clang/Basic/DiagnosticOptions.h"
+#include "clang/Driver/Driver.h"
 #include "clang/Driver/DriverDiagnostic.h"
 #include "clang/Driver/OptionUtils.h"
 #include "clang/Driver/Options.h"
@@ -242,6 +243,9 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
                    clang::driver::options::OPT_fno_stack_arrays, false))
     opts.StackArrays = 1;
 
+  if (args.getLastArg(clang::driver::options::OPT_vectorize_loops))
+    opts.VectorizeLoop = 1;
+
   if (args.hasFlag(clang::driver::options::OPT_floop_versioning,
                    clang::driver::options::OPT_fno_loop_versioning, false))
     opts.LoopVersioning = 1;

flang/lib/Frontend/FrontendActions.cpp

@@ -1037,6 +1037,8 @@ void CodeGenAction::runOptimizationPipeline(llvm::raw_pwrite_stream &os) {
     si.getTimePasses().setOutStream(ci.getTimingStreamLLVM());
   pto.LoopUnrolling = opts.UnrollLoops;
   pto.LoopInterleaving = opts.UnrollLoops;
+  pto.LoopVectorization = opts.VectorizeLoop;
+
   llvm::PassBuilder pb(targetMachine, pto, pgoOpt, &pic);
 
   // Attempt to load pass plugins and register their callbacks with PB.

flang/test/Driver/optimization-remark.f90

@@ -5,33 +5,33 @@
 ! DEFINE: %{output} = -emit-llvm -flang-deprecated-no-hlfir -o /dev/null 2>&1
 
 ! Check fc1 can handle -Rpass
-! RUN: %flang_fc1 %s -O1 -Rpass %{output} 2>&1 | FileCheck %s --check-prefix=REMARKS
+! RUN: %flang_fc1 %s -O1 -vectorize-loops -Rpass %{output} 2>&1 | FileCheck %s --check-prefix=REMARKS
 
 ! Check that we can override -Rpass= with -Rno-pass.
-! RUN: %flang_fc1 %s -O1 -Rpass -Rno-pass %{output} 2>&1 | FileCheck %s --allow-empty --check-prefix=NO-REMARKS
+! RUN: %flang_fc1 %s -O1 -vectorize-loops -Rpass -Rno-pass %{output} 2>&1 | FileCheck %s --allow-empty --check-prefix=NO-REMARKS
 
 ! Check -Rno-pass, -Rno-pass-analysis, -Rno-pass-missed nothing emitted
-! RUN: %flang %s -O1 -Rno-pass -S %{output} 2>&1 | FileCheck %s --allow-empty --check-prefix=NO-REMARKS
-! RUN: %flang %s -O1 -Rno-pass-missed -S %{output} 2>&1 | FileCheck %s --allow-empty --check-prefix=NO-REMARKS
-! RUN: %flang %s -O1 -Rno-pass-analysis -S %{output} 2>&1 | FileCheck %s --allow-empty --check-prefix=NO-REMARKS
+! RUN: %flang %s -O2 -Rno-pass -S %{output} 2>&1 | FileCheck %s --allow-empty --check-prefix=NO-REMARKS
+! RUN: %flang %s -O2 -Rno-pass-missed -S %{output} 2>&1 | FileCheck %s --allow-empty --check-prefix=NO-REMARKS
+! RUN: %flang %s -O2 -Rno-pass-analysis -S %{output} 2>&1 | FileCheck %s --allow-empty --check-prefix=NO-REMARKS
 
 ! Check valid -Rpass regex
-! RUN: %flang %s -O1 -Rpass=loop -S %{output} 2>&1 | FileCheck %s --check-prefix=PASS-REGEX-LOOP-ONLY
+! RUN: %flang %s -O2 -Rpass=loop -S %{output} 2>&1 | FileCheck %s --check-prefix=PASS-REGEX-LOOP-ONLY
 
 ! Check valid -Rpass-missed regex
-! RUN: %flang %s -O1 -Rpass-missed=loop -S %{output} 2>&1 | FileCheck %s --check-prefix=MISSED-REGEX-LOOP-ONLY
+! RUN: %flang %s -O2 -Rpass-missed=loop -S %{output} 2>&1 | FileCheck %s --check-prefix=MISSED-REGEX-LOOP-ONLY
 
 ! Check valid -Rpass-analysis regex
-! RUN: %flang %s -O1 -Rpass-analysis=loop -S %{output} 2>&1 | FileCheck %s --check-prefix=ANALYSIS-REGEX-LOOP-ONLY
+! RUN: %flang %s -O2 -Rpass-analysis=loop -S %{output} 2>&1 | FileCheck %s --check-prefix=ANALYSIS-REGEX-LOOP-ONLY
 
 ! Check full -Rpass message is emitted
-! RUN: %flang %s -O1 -Rpass -S %{output} 2>&1 | FileCheck %s --check-prefix=PASS
+! RUN: %flang %s -O2 -Rpass -S %{output} 2>&1 | FileCheck %s --check-prefix=PASS
 
 ! Check full -Rpass-missed message is emitted
-! RUN: %flang %s -O1 -Rpass-missed -S %{output} 2>&1 | FileCheck %s --check-prefix=MISSED
+! RUN: %flang %s -O2 -Rpass-missed -S %{output} 2>&1 | FileCheck %s --check-prefix=MISSED
 
 ! Check full -Rpass-analysis message is emitted
-! RUN: %flang %s -O1 -Rpass-analysis -S -o /dev/null 2>&1 | FileCheck %s --check-prefix=ANALYSIS
+! RUN: %flang %s -O2 -Rpass-analysis -S -o /dev/null 2>&1 | FileCheck %s --check-prefix=ANALYSIS
 
 ! REMARKS: remark:
 ! NO-REMARKS-NOT: remark:

flang/test/Integration/unroll-loops.f90

@@ -1,8 +1,8 @@
 ! FIXME: https://github.com/llvm/llvm-project/issues/123668
 !
 ! DEFINE: %{triple} =
-! DEFINE: %{check-unroll} = %flang_fc1 -emit-llvm -O1 -funroll-loops -mllvm -force-vector-width=2 -triple %{triple} -o- %s | FileCheck %s --check-prefixes=CHECK,UNROLL
-! DEFINE: %{check-nounroll} = %flang_fc1 -emit-llvm -O1 -mllvm -force-vector-width=2 -triple %{triple} -o- %s | FileCheck %s --check-prefixes=CHECK,NO-UNROLL
+! DEFINE: %{check-unroll} = %flang_fc1 -emit-llvm -O1 -vectorize-loops -funroll-loops -mllvm -force-vector-width=2 -triple %{triple} -o- %s | FileCheck %s --check-prefixes=CHECK,UNROLL
+! DEFINE: %{check-nounroll} = %flang_fc1 -emit-llvm -O1 -vectorize-loops -mllvm -force-vector-width=2 -triple %{triple} -o- %s | FileCheck %s --check-prefixes=CHECK,NO-UNROLL
 !
 ! REDEFINE: %{triple} = aarch64-unknown-linux-gnu
 ! RUN: %if aarch64-registered-target %{ %{check-unroll} %}