AMDGPU: VALU carry-in and v_cndmask condition cannot be EXEC
The hardware will only forward EXEC_LO; the high 32 bits will be zero.
Additionally, inline constants do not work. At least,
v_addc_u32_e64 v0, vcc, v0, v1, -1
which could conceivably be used to combine (v0 + v1 + 1) into a single
instruction, acts as if all carry-in bits are zero.
The llvm.amdgcn.ps.live test is adjusted; it would be nice to combine
s_mov_b64 s[0:1], exec
v_cndmask_b32_e64 v0, v1, v2, s[0:1]
into
v_mov_b32 v0, v3
but it's not particularly high priority.
Fixes dEQP-GLES31.functional.shaders.helper_invocation.value.*
llvm-svn: 314522
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 6a751d7..3a125c2 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -3012,15 +3012,18 @@
unsigned DstLo = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
unsigned DstHi = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+ unsigned SrcCondCopy = MRI.createVirtualRegister(&AMDGPU::SReg_64_XEXECRegClass);
+ BuildMI(*BB, MI, DL, TII->get(AMDGPU::COPY), SrcCondCopy)
+ .addReg(SrcCond);
BuildMI(*BB, MI, DL, TII->get(AMDGPU::V_CNDMASK_B32_e64), DstLo)
.addReg(Src0, 0, AMDGPU::sub0)
.addReg(Src1, 0, AMDGPU::sub0)
- .addReg(SrcCond);
+ .addReg(SrcCondCopy);
BuildMI(*BB, MI, DL, TII->get(AMDGPU::V_CNDMASK_B32_e64), DstHi)
.addReg(Src0, 0, AMDGPU::sub1)
.addReg(Src1, 0, AMDGPU::sub1)
- .addReg(SrcCond);
+ .addReg(SrcCondCopy);
BuildMI(*BB, MI, DL, TII->get(AMDGPU::REG_SEQUENCE), Dst)
.addReg(DstLo)