Skip to content

[flang] Add -f[no-]unroll-loops flag #122906

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Jan 16, 2025
Merged

[flang] Add -f[no-]unroll-loops flag #122906

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
c9b2e58
[flang] Add -f[no-]unroll-loops flag
DavidTruby Jan 8, 2025
d66a7c6
Fix bad rebase and add compiler->frontend forwarding test
DavidTruby Jan 14, 2025
ac03aad
Fix test on x86
DavidTruby Jan 14, 2025
f27462b
Add integration test
DavidTruby Jan 14, 2025
0141e0d
Switch to i64 to simplify tests
DavidTruby Jan 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions clang/include/clang/Driver/Options.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4157,9 +4157,9 @@ def ftrap_function_EQ : Joined<["-"], "ftrap-function=">, Group<f_Group>,
HelpText<"Issue call to specified function rather than a trap instruction">,
MarshallingInfoString<CodeGenOpts<"TrapFuncName">>;
def funroll_loops : Flag<["-"], "funroll-loops">, Group<f_Group>,
HelpText<"Turn on loop unroller">, Visibility<[ClangOption, CC1Option]>;
HelpText<"Turn on loop unroller">, Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>;
def fno_unroll_loops : Flag<["-"], "fno-unroll-loops">, Group<f_Group>,
HelpText<"Turn off loop unroller">, Visibility<[ClangOption, CC1Option]>;
HelpText<"Turn off loop unroller">, Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>;
def ffinite_loops: Flag<["-"], "ffinite-loops">, Group<f_Group>,
HelpText<"Assume all non-trivial loops are finite.">, Visibility<[ClangOption, CC1Option]>;
def fno_finite_loops: Flag<["-"], "fno-finite-loops">, Group<f_Group>,
Expand Down
3 changes: 2 additions & 1 deletion clang/lib/Driver/ToolChains/Flang.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,7 +155,8 @@ void Flang::addCodegenOptions(const ArgList &Args,
options::OPT_flang_deprecated_no_hlfir,
options::OPT_fno_ppc_native_vec_elem_order,
options::OPT_fppc_native_vec_elem_order,
options::OPT_ftime_report, options::OPT_ftime_report_EQ});
options::OPT_ftime_report, options::OPT_ftime_report_EQ,
options::OPT_funroll_loops, options::OPT_fno_unroll_loops});
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can add a forwarding test from the driver to the frontend driver.

}

void Flang::addPicOptions(const ArgList &Args, ArgStringList &CmdArgs) const {
Expand Down
1 change: 1 addition & 0 deletions flang/include/flang/Frontend/CodeGenOptions.def
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,7 @@ 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(LoopVersioning, 1, 0) ///< Enable loop versioning.
CODEGENOPT(UnrollLoops, 1, 0) ///< Enable loop unrolling
CODEGENOPT(AliasAnalysis, 1, 0) ///< Enable alias analysis pass

CODEGENOPT(Underscoring, 1, 1)
Expand Down
4 changes: 4 additions & 0 deletions flang/lib/Frontend/CompilerInvocation.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -246,6 +246,10 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
clang::driver::options::OPT_fno_loop_versioning, false))
opts.LoopVersioning = 1;

opts.UnrollLoops = args.hasFlag(clang::driver::options::OPT_funroll_loops,
clang::driver::options::OPT_fno_unroll_loops,
(opts.OptimizationLevel > 1));

opts.AliasAnalysis = opts.OptimizationLevel > 0;

// -mframe-pointer=none/non-leaf/all option.
Expand Down
2 changes: 2 additions & 0 deletions flang/lib/Frontend/FrontendActions.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1028,6 +1028,8 @@ void CodeGenAction::runOptimizationPipeline(llvm::raw_pwrite_stream &os) {
si.registerCallbacks(pic, &mam);
if (ci.isTimingEnabled())
si.getTimePasses().setOutStream(ci.getTimingStreamLLVM());
pto.LoopUnrolling = opts.UnrollLoops;
pto.LoopInterleaving = opts.UnrollLoops;
llvm::PassBuilder pb(targetMachine, pto, pgoOpt, &pic);

// Attempt to load pass plugins and register their callbacks with PB.
Expand Down
5 changes: 5 additions & 0 deletions flang/test/Driver/funroll-loops.f90
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
! RUN: %flang -### -funroll-loops %s 2>&1 | FileCheck %s -check-prefix UNROLL
! RUN: %flang -### -fno-unroll-loops %s 2>&1 | FileCheck %s -check-prefix NO-UNROLL

! UNROLL: "-funroll-loops"
! NO-UNROLL: "-fno-unroll-loops"
43 changes: 43 additions & 0 deletions flang/test/HLFIR/unroll-loops.fir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,43 @@
// RUN: %flang_fc1 -emit-llvm -O1 -funroll-loops -mllvm -force-vector-width=2 -o- %s | FileCheck %s --check-prefixes=CHECK,UNROLL
// RUN: %flang_fc1 -emit-llvm -O2 -mllvm -force-vector-width=2 -o- %s | FileCheck %s --check-prefixes=CHECK,UNROLL
// RUN: %flang_fc1 -emit-llvm -O1 -fno-unroll-loops -mllvm -force-vector-width=2 -o- %s | FileCheck %s --check-prefixes=CHECK,NO-UNROLL
// RUN: %flang_fc1 -emit-llvm -O1 -mllvm -force-vector-width=2 -o- %s | FileCheck %s --check-prefixes=CHECK,NO-UNROLL

// CHECK-LABEL: @unroll
// CHECK-SAME: (ptr nocapture writeonly %[[ARG0:.*]])
func.func @unroll(%arg0: !fir.ref<!fir.array<1000xf64>> {fir.bindc_name = "a"}) {
%scope = fir.dummy_scope : !fir.dscope
%c1000 = arith.constant 1000 : index
%shape = fir.shape %c1000 : (index) -> !fir.shape<1>
%a:2 = hlfir.declare %arg0(%shape) dummy_scope %scope {uniq_name = "unrollEa"} : (!fir.ref<!fir.array<1000xf64>>, !fir.shape<1>, !fir.dscope) -> (!fir.ref<!fir.array<1000xf64>>, !fir.ref<!fir.array<1000xf64>>)
%c1 = arith.constant 1 : index
fir.do_loop %arg1 = %c1 to %c1000 step %c1 {
// CHECK: br label %[[BLK:.*]]
// CHECK: [[BLK]]:
// CHECK-NEXT: %[[IND:.*]] = phi i64 [ 0, %{{.*}} ], [ %[[NIV:.*]], %[[BLK]] ]
// CHECK-NEXT: %[[VIND:.*]] = phi <2 x i64> [ <i64 1, i64 2>, %{{.*}} ], [ %[[NVIND:.*]], %[[BLK]] ]

// NO-UNROLL-NEXT: %[[IV_D:.*]] = uitofp nneg <2 x i64> %[[VIND]] to <2 x double>
// NO-UNROLL-NEXT: %[[GEP:.*]] = getelementptr double, ptr %[[ARG0]], i64 %[[IND]]
// NO-UNROLL-NEXT: store <2 x double> %[[IV_D]], ptr %[[GEP]]
// NO-UNROLL-NEXT: %[[NIV:.*]] = add nuw i64 %{{.*}}, 2
// NO-UNROLL-NEXT: %[[NVIND]] = add <2 x i64> %[[VIND]], splat (i64 2)

// UNROLL-NEXT: %[[VIND1:.*]] = add <2 x i64> %[[VIND]], splat (i64 2)
// UNROLL-NEXT: %[[IV0_D:.*]] = uitofp nneg <2 x i64> %[[VIND]] to <2 x double>
// UNROLL-NEXT: %[[IV1_D:.*]] = uitofp nneg <2 x i64> %[[VIND1]] to <2 x double>
// UNROLL-NEXT: %[[GEP0:.*]] = getelementptr double, ptr %[[ARG0]], i64 %[[IND]]
// UNROLL-NEXT: %[[GEP1:.*]] = getelementptr i8, ptr %[[GEP0]], i64 16
// UNROLL-NEXT: store <2 x double> %[[IV0_D]], ptr %[[GEP0]]
// UNROLL-NEXT: store <2 x double> %[[IV1_D]], ptr %[[GEP1]]
// UNROLL-NEXT: %[[NIV:.*]] = add nuw i64 %[[IND]], 4
// UNROLL-NEXT: %[[NVIND:.*]] = add <2 x i64> %[[VIND]], splat (i64 4)

// CHECK-NEXT: %[[EXIT:.*]] = icmp eq i64 %[[NIV]], 1000
// CHECK-NEXT: br i1 %[[EXIT]], label %{{.*}}, label %[[BLK]]
%iv = fir.convert %arg1 : (index) -> f64
%ai = hlfir.designate %a#0 (%arg1) : (!fir.ref<!fir.array<1000xf64>>, index) -> !fir.ref<f64>
hlfir.assign %iv to %ai : f64, !fir.ref<f64>
}
return
}
37 changes: 37 additions & 0 deletions flang/test/Integration/unroll-loops.f90
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
! RUN: %flang_fc1 -emit-llvm -O1 -funroll-loops -mllvm -force-vector-width=2 -o- %s | FileCheck %s --check-prefixes=CHECK,UNROLL
! RUN: %flang_fc1 -emit-llvm -O2 -mllvm -force-vector-width=2 -o- %s | FileCheck %s --check-prefixes=CHECK,UNROLL
! RUN: %flang_fc1 -emit-llvm -O1 -fno-unroll-loops -mllvm -force-vector-width=2 -o- %s | FileCheck %s --check-prefixes=CHECK,NO-UNROLL
! RUN: %flang_fc1 -emit-llvm -O1 -mllvm -force-vector-width=2 -o- %s | FileCheck %s --check-prefixes=CHECK,NO-UNROLL

! CHECK-LABEL: @unroll
! CHECK-SAME: (ptr nocapture writeonly %[[ARG0:.*]])
subroutine unroll(a)
real(kind=8), intent(out) :: a(1000)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can this be an integer array to avoid the uitofp ?

integer(kind=8) :: i
! CHECK: br label %[[BLK:.*]]
! CHECK: [[BLK]]:
! CHECK-NEXT: %[[IND:.*]] = phi i64 [ 0, %{{.*}} ], [ %[[NIV:.*]], %[[BLK]] ]
! CHECK-NEXT: %[[VIND:.*]] = phi <2 x i64> [ <i64 1, i64 2>, %{{.*}} ], [ %[[NVIND:.*]], %[[BLK]] ]
!
! NO-UNROLL-NEXT: %[[IV_D:.*]] = uitofp nneg <2 x i64> %[[VIND]] to <2 x double>
! NO-UNROLL-NEXT: %[[GEP:.*]] = getelementptr double, ptr %[[ARG0]], i64 %[[IND]]
! NO-UNROLL-NEXT: store <2 x double> %[[IV_D]], ptr %[[GEP]]
! NO-UNROLL-NEXT: %[[NIV:.*]] = add nuw i64 %{{.*}}, 2
! NO-UNROLL-NEXT: %[[NVIND]] = add <2 x i64> %[[VIND]], splat (i64 2)
Copy link
Contributor

@kiranchandramohan kiranchandramohan Jan 14, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Unroll should ideally check for the branch back to the body. And nounroll should probably check that there is no such branch?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

-funroll-loops doesn't mean that the loop will be fully unrolled, just that some unrolling can occur. So there's a branch back to the body of the loop in both cases.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For small loops, it will fully unroll if the body is small.

The issue with checking for two iterations is that this could be due to interleaving during vectorisation as well.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I actually did try to only have (e.g.) 2 iterations of the loop; for some reason the full loop is still emitted, at least at -O1. What I ended up with here was essentially copied from the clang -funroll-loops tests but converted to fortran.

!
! UNROLL-NEXT: %[[VIND1:.*]] = add <2 x i64> %[[VIND]], splat (i64 2)
! UNROLL-NEXT: %[[IV0_D:.*]] = uitofp nneg <2 x i64> %[[VIND]] to <2 x double>
! UNROLL-NEXT: %[[IV1_D:.*]] = uitofp nneg <2 x i64> %[[VIND1]] to <2 x double>
! UNROLL-NEXT: %[[GEP0:.*]] = getelementptr double, ptr %[[ARG0]], i64 %[[IND]]
! UNROLL-NEXT: %[[GEP1:.*]] = getelementptr i8, ptr %[[GEP0]], i64 16
! UNROLL-NEXT: store <2 x double> %[[IV0_D]], ptr %[[GEP0]]
! UNROLL-NEXT: store <2 x double> %[[IV1_D]], ptr %[[GEP1]]
! UNROLL-NEXT: %[[NIV:.*]] = add nuw i64 %[[IND]], 4
! UNROLL-NEXT: %[[NVIND:.*]] = add <2 x i64> %[[VIND]], splat (i64 4)
!
! CHECK-NEXT: %[[EXIT:.*]] = icmp eq i64 %[[NIV]], 1000
! CHECK-NEXT: br i1 %[[EXIT]], label %{{.*}}, label %[[BLK]]
do i=1,1000
a(i) = i
end do
end subroutine
Loading