[AMDGPU] Disable nested endcf collapse

The assumption is that conditional regions are perfectly nested
and a mask restored at the exit from the inner block will be
completely covered by a mask restored in the outer.

It turns out with our current structurizer this is not always
the case.

Disable the optimization for now, but I want to keep it around
for a while to either try after further structurizer changes or
to move it into control flow lowering where we have more info
and reuse the test.

Differential Revision: https://reviews.llvm.org/D75958

diff --git a/llvm/lib/Target/AMDGPU/SIOptimizeExecMaskingPreRA.cpp b/llvm/lib/Target/AMDGPU/SIOptimizeExecMaskingPreRA.cpp
index 34199d3..e1786ef 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIOptimizeExecMaskingPreRA.cpp
+++ b/llvm/lib/Target/AMDGPU/SIOptimizeExecMaskingPreRA.cpp
@@ -31,6 +31,10 @@ using namespace llvm;
 
 #define DEBUG_TYPE "si-optimize-exec-masking-pre-ra"
 
+static cl::opt<bool>
+RemoveRedundantEndcf("amdgpu-remove-redundant-endcf",
+    cl::init(false), cl::ReallyHidden);
+
 namespace {
 
 class SIOptimizeExecMaskingPreRA : public MachineFunctionPass {
@@ -379,7 +383,13 @@ bool SIOptimizeExecMaskingPreRA::runOnMachineFunction(MachineFunction &MF) {
       continue;
     }
 
+    if (!RemoveRedundantEndcf)
+      continue;
+
     // Try to collapse adjacent endifs.
+    // The assumption is that conditional regions are perfectly nested and
+    // a mask restored at the exit from the inner block will be completely
+    // covered by a mask restored in the outer.
     auto E = MBB.end();
     auto Lead = skipDebugInstructionsForward(MBB.begin(), E);
     if (MBB.succ_size() != 1 || Lead == E || !isEndCF(*Lead, TRI, ST))

diff --git a/llvm/test/CodeGen/AMDGPU/collapse-endcf.ll b/llvm/test/CodeGen/AMDGPU/collapse-endcf.ll
index bb00e67..bc1c364 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/collapse-endcf.ll
+++ b/llvm/test/CodeGen/AMDGPU/collapse-endcf.ll
@@ -1,4 +1,4 @@
-; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,ALL %s
+; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs -amdgpu-remove-redundant-endcf < %s | FileCheck -enable-var-scope -check-prefixes=GCN,ALL %s
 ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs -amdgpu-opt-exec-mask-pre-ra=0 < %s | FileCheck -enable-var-scope -check-prefixes=DISABLED,ALL %s
 
 ; ALL-LABEL: {{^}}simple_nested_if:

diff --git a/llvm/test/CodeGen/AMDGPU/collapse-endcf.mir b/llvm/test/CodeGen/AMDGPU/collapse-endcf.mir
index 1a26a50..3b1358d 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/collapse-endcf.mir
+++ b/llvm/test/CodeGen/AMDGPU/collapse-endcf.mir
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -march=amdgcn -verify-machineinstrs -run-pass=si-optimize-exec-masking-pre-ra %s -o - | FileCheck -check-prefix=GCN %s
+# RUN: llc -march=amdgcn -verify-machineinstrs -run-pass=si-optimize-exec-masking-pre-ra -amdgpu-remove-redundant-endcf %s -o - | FileCheck -check-prefix=GCN %s
 
 # Make sure dbg_value doesn't change codeegn when collapsing end_cf
 ---