maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[AMDGPU] Disable endcf collapse 

There are some functional regressions and I suspect our
scopes are not as perfectly enclosed as I expected.
Disable it for now.

Differential Revision: https://reviews.llvm.org/D76148
This commit is contained in:
Stanislav Mekhanoshin 2020-03-13 10:58:19 -07:00
parent 994c071a1b
commit 32e90cbcd1
4 changed files with 21 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
View File

@ -74,6 +74,10 @@ using namespace llvm;
#define DEBUG_TYPE "si-lower-control-flow" #define DEBUG_TYPE "si-lower-control-flow"
static cl::opt<bool>
RemoveRedundantEndcf("amdgpu-remove-redundant-endcf",
cl::init(false), cl::ReallyHidden);
namespace { namespace {
class SILowerControlFlow : public MachineFunctionPass { class SILowerControlFlow : public MachineFunctionPass {
@ -444,14 +448,17 @@ void SILowerControlFlow::emitEndCf(MachineInstr &MI) {
// If the only instruction immediately following this END_CF is an another // If the only instruction immediately following this END_CF is an another
// END_CF in the only successor we can avoid emitting exec mask restore here. // END_CF in the only successor we can avoid emitting exec mask restore here.
auto Next = skipIgnoreExecInstsTrivialSucc(MBB, std::next(MI.getIterator())); if (RemoveRedundantEndcf) {
if (Next != MBB.end() && (Next->getOpcode() == AMDGPU::SI_END_CF || auto Next =
LoweredEndCf.count(&*Next))) { skipIgnoreExecInstsTrivialSucc(MBB, std::next(MI.getIterator()));
LLVM_DEBUG(dbgs() << "Skip redundant "; MI.dump()); if (Next != MBB.end() && (Next->getOpcode() == AMDGPU::SI_END_CF ||
if (LIS) LoweredEndCf.count(&*Next))) {
LIS->RemoveMachineInstrFromMaps(MI); LLVM_DEBUG(dbgs() << "Skip redundant "; MI.dump());
MI.eraseFromParent(); if (LIS)
return; LIS->RemoveMachineInstrFromMaps(MI);
MI.eraseFromParent();
return;
}
} }
MachineBasicBlock::iterator InsPt = MachineBasicBlock::iterator InsPt =

2
llvm/test/CodeGen/AMDGPU/collapse-endcf.ll
View File

@ -1,4 +1,4 @@
; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs -amdgpu-remove-redundant-endcf < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
; GCN-LABEL: {{^}}simple_nested_if: ; GCN-LABEL: {{^}}simple_nested_if:
; GCN: s_and_saveexec_b64 [[SAVEEXEC:s\[[0-9:]+\]]] ; GCN: s_and_saveexec_b64 [[SAVEEXEC:s\[[0-9:]+\]]]

2
llvm/test/CodeGen/AMDGPU/collapse-endcf.mir
View File

@ -1,5 +1,5 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -march=amdgcn -verify-machineinstrs -run-pass=si-lower-control-flow %s -o - | FileCheck -check-prefix=GCN %s # RUN: llc -march=amdgcn -verify-machineinstrs -run-pass=si-lower-control-flow -amdgpu-remove-redundant-endcf %s -o - | FileCheck -check-prefix=GCN %s
# Make sure dbg_value doesn't change codeegn when collapsing end_cf # Make sure dbg_value doesn't change codeegn when collapsing end_cf
--- ---

6
llvm/test/CodeGen/AMDGPU/mul24-pass-ordering.ll
View File

@ -58,7 +58,7 @@ define void @lsr_order_mul24_1(i32 %arg, i32 %arg1, i32 %arg2, float addrspace(3
; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v5 ; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v5
; GFX9-NEXT: v_cmp_lt_u32_e64 s[4:5], v0, v1 ; GFX9-NEXT: v_cmp_lt_u32_e64 s[4:5], v0, v1
; GFX9-NEXT: s_and_saveexec_b64 s[10:11], s[4:5] ; GFX9-NEXT: s_and_saveexec_b64 s[10:11], s[4:5]
; GFX9-NEXT: s_cbranch_execz BB1_3 ; GFX9-NEXT: s_cbranch_execz BB1_4
; GFX9-NEXT: ; %bb.1: ; %bb19 ; GFX9-NEXT: ; %bb.1: ; %bb19
; GFX9-NEXT: v_cvt_f32_u32_e32 v7, v6 ; GFX9-NEXT: v_cvt_f32_u32_e32 v7, v6
; GFX9-NEXT: v_and_b32_e32 v5, 0xffffff, v6 ; GFX9-NEXT: v_and_b32_e32 v5, 0xffffff, v6
@ -100,7 +100,9 @@ define void @lsr_order_mul24_1(i32 %arg, i32 %arg1, i32 %arg2, float addrspace(3
; GFX9-NEXT: v_add_u32_e32 v3, v3, v6 ; GFX9-NEXT: v_add_u32_e32 v3, v3, v6
; GFX9-NEXT: s_andn2_b64 exec, exec, s[12:13] ; GFX9-NEXT: s_andn2_b64 exec, exec, s[12:13]
; GFX9-NEXT: s_cbranch_execnz BB1_2 ; GFX9-NEXT: s_cbranch_execnz BB1_2
; GFX9-NEXT: BB1_3: ; %Flow3 ; GFX9-NEXT: ; %bb.3: ; %Flow
; GFX9-NEXT: s_or_b64 exec, exec, s[12:13]
; GFX9-NEXT: BB1_4: ; %Flow3
; GFX9-NEXT: s_or_b64 exec, exec, s[10:11] ; GFX9-NEXT: s_or_b64 exec, exec, s[10:11]
; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31] ; GFX9-NEXT: s_setpc_b64 s[30:31]