[llvm] r350350 - [AMDGPU] Fix scalar operand folding bug that causes SHOC performance regression.
Alexander Timofeev via llvm-commits
llvm-commits at lists.llvm.org
Thu Jan 3 11:55:32 PST 2019
Author: alex-t
Date: Thu Jan 3 11:55:32 2019
New Revision: 350350
URL: http://llvm.org/viewvc/llvm-project?rev=350350&view=rev
Log:
[AMDGPU] Fix scalar operand folding bug that causes SHOC performance regression.
Detailed description: SIFoldOperands::foldInstOperand iterates over the
operand uses calling the function that changes def-use iteratorson the
way. As a result loop exits immediately when def-use iterator is
changed. Hence, the operand is folded to the very first use instruction
only. This makes VGPR live along the whole basic block and increases
register pressure significantly. The performance drop observed in SHOC
DeviceMemory test is caused by this bug.
Proposed fix: collect uses to separate container for further processing
in another loop.
Testing: make check-llvm
SHOC performance test.
Reviewers: rampitec, ronlieb
Differential Revision: https://reviews.llvm.org/D56161
Modified:
llvm/trunk/lib/Target/AMDGPU/SIFoldOperands.cpp
llvm/trunk/test/CodeGen/AMDGPU/fdiv32-to-rcp-folding.ll
Modified: llvm/trunk/lib/Target/AMDGPU/SIFoldOperands.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/SIFoldOperands.cpp?rev=350350&r1=350349&r2=350350&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIFoldOperands.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/SIFoldOperands.cpp Thu Jan 3 11:55:32 2019
@@ -854,13 +854,17 @@ void SIFoldOperands::foldInstOperand(Mac
}
} else {
// Folding register.
+ SmallVector <MachineRegisterInfo::use_iterator, 4> UsesToProcess;
for (MachineRegisterInfo::use_iterator
Use = MRI->use_begin(Dst.getReg()), E = MRI->use_end();
Use != E; ++Use) {
- MachineInstr *UseMI = Use->getParent();
+ UsesToProcess.push_back(Use);
+ }
+ for (auto U : UsesToProcess) {
+ MachineInstr *UseMI = U->getParent();
- foldOperand(OpToFold, UseMI, Use.getOperandNo(),
- FoldList, CopiesToReplace);
+ foldOperand(OpToFold, UseMI, U.getOperandNo(),
+ FoldList, CopiesToReplace);
}
}
Modified: llvm/trunk/test/CodeGen/AMDGPU/fdiv32-to-rcp-folding.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fdiv32-to-rcp-folding.ll?rev=350350&r1=350349&r2=350350&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fdiv32-to-rcp-folding.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fdiv32-to-rcp-folding.ll Thu Jan 3 11:55:32 2019
@@ -131,10 +131,10 @@ define amdgpu_kernel void @div_v4_1_by_x
; GCN-DENORM-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 1.0, [[S]], vcc
; GCN-DENORM-DAG: v_cmp_gt_f32_e64 vcc, |v{{[0-9]+}}|, [[L]]
; GCN-DENORM-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 1.0, [[S]], vcc
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, v{{[0-9]+}}, -v{{[0-9]+}}
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, v{{[0-9]+}}, -v{{[0-9]+}}
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, v{{[0-9]+}}, -v{{[0-9]+}}
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, v{{[0-9]+}}, -v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -v{{[0-9]+}}
; GCN-DENORM-DAG: v_rcp_f32_e32
; GCN-DENORM-DAG: v_rcp_f32_e32
; GCN-DENORM-DAG: v_rcp_f32_e32
@@ -166,10 +166,10 @@ define amdgpu_kernel void @div_v4_minus_
; GCN-DENORM-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 1.0, [[S]], vcc
; GCN-DENORM-DAG: v_cmp_gt_f32_e64 vcc, |v{{[0-9]+}}|, [[L]]
; GCN-DENORM-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 1.0, [[S]], vcc
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -s{{[0-9]+}}, v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -s{{[0-9]+}}, v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -s{{[0-9]+}}, v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -s{{[0-9]+}}, v{{[0-9]+}}
; GCN-DENORM-DAG: v_rcp_f32_e32
; GCN-DENORM-DAG: v_rcp_f32_e32
; GCN-DENORM-DAG: v_rcp_f32_e32
@@ -246,7 +246,7 @@ define amdgpu_kernel void @div_v4_minus_
; GCN-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 1.0, [[S]], vcc
; GCN-DENORM-DAG: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, v{{[0-9]+}}, -v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -v{{[0-9]+}}
; GCN-DENORM-DAG: v_rcp_f32_e32 [[RCP1:v[0-9]+]], v{{[0-9]+}}
; GCN-DENORM-DAG: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, [[RCP1]]
; GCN-DENORM-DAG: v_rcp_f32_e32 [[RCP2:v[0-9]+]], v{{[0-9]+}}
@@ -288,7 +288,7 @@ define amdgpu_kernel void @div_v4_c_by_x
; GCN-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 1.0, [[S]], vcc
; GCN-DENORM-DAG: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
-; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}
+; GCN-DENORM-DAG: v_mul_f32_e64 v{{[0-9]+}}, -s{{[0-9]+}}, v{{[0-9]+}}
; GCN-DENORM-DAG: v_rcp_f32_e32 [[RCP1:v[0-9]+]], v{{[0-9]+}}
; GCN-DENORM-DAG: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, [[RCP1]]
; GCN-DENORM-DAG: v_rcp_f32_e32 [[RCP2:v[0-9]+]], v{{[0-9]+}}
More information about the llvm-commits
mailing list