[llvm] 03da6e8 - [ModuloSchedule] Do target loop analysis before peeling.
Thomas Raoux via llvm-commits
llvm-commits at lists.llvm.org
Mon Nov 11 09:35:57 PST 2019
Author: Thomas Raoux
Date: 2019-11-11T09:35:39-08:00
New Revision: 03da6e8c00de2320e6b9dacba8f7850faceae319
URL: https://github.com/llvm/llvm-project/commit/03da6e8c00de2320e6b9dacba8f7850faceae319
DIFF: https://github.com/llvm/llvm-project/commit/03da6e8c00de2320e6b9dacba8f7850faceae319.diff
LOG: [ModuloSchedule] Do target loop analysis before peeling.
Simple change to call target hook analyzeLoopForPipelining before
changing the loop. After peeling analyzing the loop may be more
complicated for target that don't have a loop instruction. This doesn't
affect Hexagone and PPC as they have hardware loop instructions.
Differential Revision: https://reviews.llvm.org/D69912
Added:
Modified:
llvm/include/llvm/CodeGen/ModuloSchedule.h
llvm/lib/CodeGen/ModuloSchedule.cpp
Removed:
################################################################################
diff --git a/llvm/include/llvm/CodeGen/ModuloSchedule.h b/llvm/include/llvm/CodeGen/ModuloSchedule.h
index 81a9b63b64ca..f3b2424bd684 100644
--- a/llvm/include/llvm/CodeGen/ModuloSchedule.h
+++ b/llvm/include/llvm/CodeGen/ModuloSchedule.h
@@ -342,6 +342,8 @@ class PeelingModuloScheduleExpander {
MI = CanonicalMIs[MI];
return Schedule.getStage(MI);
}
+ /// Target loop info before kernel peeling.
+ std::unique_ptr<TargetInstrInfo::PipelinerLoopInfo> Info;
};
/// Expander that simply annotates each scheduled instruction with a post-instr
diff --git a/llvm/lib/CodeGen/ModuloSchedule.cpp b/llvm/lib/CodeGen/ModuloSchedule.cpp
index 7ce3c5861801..98de0816f27c 100644
--- a/llvm/lib/CodeGen/ModuloSchedule.cpp
+++ b/llvm/lib/CodeGen/ModuloSchedule.cpp
@@ -1759,10 +1759,6 @@ void PeelingModuloScheduleExpander::rewriteUsesOf(MachineInstr *MI) {
}
void PeelingModuloScheduleExpander::fixupBranches() {
- std::unique_ptr<TargetInstrInfo::PipelinerLoopInfo> Info =
- TII->analyzeLoopForPipelining(BB);
- assert(Info);
-
// Work outwards from the kernel.
bool KernelDisposed = false;
int TC = Schedule.getNumStages() - 1;
@@ -1818,6 +1814,8 @@ void PeelingModuloScheduleExpander::expand() {
BB = Schedule.getLoop()->getTopBlock();
Preheader = Schedule.getLoop()->getLoopPreheader();
LLVM_DEBUG(Schedule.dump());
+ Info = TII->analyzeLoopForPipelining(BB);
+ assert(Info);
rewriteKernel();
peelPrologAndEpilogs();
More information about the llvm-commits
mailing list