RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-02-15 09:51:00 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

AMDGPU: Don't fold copies to physregs

Browse Source 
In a future patch, this will help cleanup m0 handling.

The register coalescer handles copies from a register that
materializes an immediate, but doesn't handle move immediates
itself. The virtual register uses will often be allocated to the same
register, so there end up being no real copy.

llvm-svn: 374257
This commit is contained in:
Matt Arsenault 2019-10-09 22:51:42 +00:00
parent 116a22fd06
commit 983c9e76c0
2 changed files with 10 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
lib/Target/AMDGPU/SIFoldOperands.cpp
View File

@ -581,13 +581,17 @@ void SIFoldOperands::foldOperand(
if (FoldingImmLike && UseMI->isCopy()) {
Register DestReg = UseMI->getOperand(0).getReg();
const TargetRegisterClass *DestRC = Register::isVirtualRegister(DestReg)
? MRI->getRegClass(DestReg)
: TRI->getPhysRegClass(DestReg);
// Don't fold into a copy to a physical register. Doing so would interfere
// with the register coalescer's logic which would avoid redundant
// initalizations.
if (DestReg.isPhysical())
return;
const TargetRegisterClass *DestRC = MRI->getRegClass(DestReg);
Register SrcReg = UseMI->getOperand(1).getReg();
if (Register::isVirtualRegister(DestReg) &&
Register::isVirtualRegister(SrcReg)) {
if (SrcReg.isVirtual()) { // XXX - This can be an assert?
const TargetRegisterClass * SrcRC = MRI->getRegClass(SrcReg);
if (TRI->isSGPRClass(SrcRC) && TRI->hasVectorRegisters(DestRC)) {
MachineRegisterInfo::use_iterator NextUse;

2
test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll
View File

@ -14,8 +14,8 @@ define amdgpu_kernel void @kernel_background_evaluate(float addrspace(5)* %kg, <
; GCN-NEXT: s_mov_b64 s[0:1], s[36:37]
; GCN-NEXT: v_mov_b32_e32 v1, 0x2000
; GCN-NEXT: v_mov_b32_e32 v2, 0x4000
; GCN-NEXT: s_mov_b64 s[2:3], s[38:39]
; GCN-NEXT: v_mov_b32_e32 v3, 0
; GCN-NEXT: s_mov_b64 s[2:3], s[38:39]
; GCN-NEXT: v_mov_b32_e32 v4, 0x400000
; GCN-NEXT: s_add_u32 s32, s33, 0xc0000
; GCN-NEXT: v_add_nc_u32_e64 v32, 4, 0x4000