[llvm-branch-commits] [llvm] [AMDGPU] Don't spill an SGPR while SCC is live in frame index lowering (PR #224080)
Domenic Nutile via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Wed Sep 16 11:46:08 PDT 2026
https://github.com/saxlungs updated https://github.com/llvm/llvm-project/pull/224080
>From 4c88d7c115c2a154ee6858f17ceece1fa344be7d Mon Sep 17 00:00:00 2001
From: Domenic Nutile <domenic.nutile at gmail.com>
Date: Wed, 16 Sep 2026 12:12:46 -0400
Subject: [PATCH] [AMDGPU] Don't spill an SGPR while SCC is live in frame index
lowering
When SCC is live into a scalar frame index user, the scaling path avoids
SALU ops that write SCC by computing the address in a VGPR and reading it
back with V_READFIRSTLANE_B32. If the destination of that readfirstlane is
scavenged with spilling allowed, an AMDGPU SGPR spill writes inactive
lanes, so it flips EXEC with S_NOT_B64 and clobbers SCC. Instead, scavenge
that register with AllowSpill=false.
---
llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp | 41 +++-
...to-vmem-scc-clobber-reserved-exec-copy.mir | 207 ++++++++++++++++++
2 files changed, 237 insertions(+), 11 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
index f33c099b845bf3..eb3bf0eeea71ff 100644
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -3437,12 +3437,33 @@ bool SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
llvm::count_if(MI->operands(), [](const MachineOperand &MO) {
return MO.isFI();
}) == 1;
- bool CanUseFrameRegAsScratch = IsSALU && !LiveSCC && FrameReg &&
- IsOnlyFrameIndex &&
- !MI->readsRegister(FrameReg, this) &&
- !MI->modifiesRegister(FrameReg, this);
+ bool CanUseFrameRegAsSGPRScratch = IsSALU && FrameReg &&
+ IsOnlyFrameIndex &&
+ !MI->readsRegister(FrameReg, this) &&
+ !MI->modifiesRegister(FrameReg, this);
+ // ResultReg is a VGPR while SCC is live, so FrameReg cannot stand in for
+ // it there.
+ bool CanUseFrameRegAsScratch = CanUseFrameRegAsSGPRScratch && !LiveSCC;
bool RestoreFrameReg = false;
+
+ // Scavenge the scalar temporary the V_READFIRSTLANE_B32 result lands in.
+ // Spilling an SGPR here would flip EXEC with S_NOT, and that clobbers the
+ // SCC this path exists to preserve, so fall back to FrameReg instead.
+ auto ScavengeSGPRForReadfirstlane =
+ [&](MachineBasicBlock::iterator To) -> Register {
+ if (Register Reg = RS->scavengeRegisterBackwards(
+ AMDGPU::SReg_32_XM0RegClass, To, false, 0,
+ /*AllowSpill=*/false))
+ return Reg;
+ if (CanUseFrameRegAsSGPRScratch) {
+ RestoreFrameReg = true;
+ return FrameReg;
+ }
+ return RS->scavengeRegisterBackwards(AMDGPU::SReg_32_XM0RegClass, To,
+ false, 0);
+ };
+
Register ResultReg;
if (IsCopy) {
ResultReg = MI->getOperand(0).getReg();
@@ -3496,8 +3517,7 @@ bool SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
assert(ResultReg.isPhysical());
NewDest = ResultReg;
} else {
- NewDest = RS->scavengeRegisterBackwards(AMDGPU::SReg_32_XM0RegClass,
- Shift, false, 0);
+ NewDest = ScavengeSGPRForReadfirstlane(Shift);
}
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32), NewDest)
.addReg(TmpResultReg);
@@ -3617,9 +3637,7 @@ bool SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
if (IsCopy) {
NewDest = ResultReg;
} else {
- NewDest = RS->scavengeRegisterBackwards(
- AMDGPU::SReg_32_XM0RegClass, *Add, false, 0,
- /*AllowSpill=*/true);
+ NewDest = ScavengeSGPRForReadfirstlane(*Add);
}
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32),
@@ -3658,8 +3676,9 @@ bool SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
if (Offset) {
int64_t ScaledOffset = -Offset * ST.getWavefrontSize();
- bool SCCLiveAfterMI = MI->definesRegister(AMDGPU::SCC, this) &&
- !MI->registerDefIsDead(AMDGPU::SCC, this);
+ bool SCCLiveAfterMI = MI->definesRegister(AMDGPU::SCC, this)
+ ? !MI->registerDefIsDead(AMDGPU::SCC, this)
+ : LiveSCC;
if (!SCCLiveAfterMI) {
BuildMI(*MBB, InsPt, DL, TII->get(AMDGPU::S_ADD_I32), FrameReg)
.addReg(FrameReg)
diff --git a/llvm/test/CodeGen/AMDGPU/sgpr-spill-to-vmem-scc-clobber-reserved-exec-copy.mir b/llvm/test/CodeGen/AMDGPU/sgpr-spill-to-vmem-scc-clobber-reserved-exec-copy.mir
index 1e232ac8f8a255..185aaa765d67f4 100644
--- a/llvm/test/CodeGen/AMDGPU/sgpr-spill-to-vmem-scc-clobber-reserved-exec-copy.mir
+++ b/llvm/test/CodeGen/AMDGPU/sgpr-spill-to-vmem-scc-clobber-reserved-exec-copy.mir
@@ -244,3 +244,210 @@ body: |
S_ENDPGM 0, implicit $vcc
...
+
+---
+# SCC is live *into* the scalar frame index user: S_CMP_EQ_U32 produces it and
+# S_CSELECT_B32 reads it. With no SGPR free to hold the V_READFIRSTLANE_B32
+# result, scavenging one would emergency-spill, and an SGPR spill flips EXEC
+# with S_NOT_B64, clobbering the very SCC this path preserves. Scale FrameReg in
+# place instead and restore it after MI with S_MUL_I32, which does not write SCC.
+name: scc_live_into_scalar_frame_index_no_free_sgpr
+tracksRegLiveness: true
+stack:
+ - { id: 0, size: 64, alignment: 4 }
+ - { id: 1, size: 64, alignment: 4 }
+machineFunctionInfo:
+ isEntryFunction: false
+ scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3'
+ frameOffsetReg: '$sgpr33'
+ stackPtrOffsetReg: '$sgpr32'
+ sgprForEXECCopy: '$sgpr100_sgpr101'
+ occupancy: 2
+body: |
+ ; CHECK-LABEL: name: scc_live_into_scalar_frame_index_no_free_sgpr
+ ; CHECK: bb.0:
+ ; CHECK-NEXT: successors: %bb.1(0x80000000)
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_def_aspace_cfa $sgpr32, 0, 6
+ ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_register_pair $pc_reg, $sgpr30, 32, $sgpr31, 32
+ ; CHECK-NEXT: S_BRANCH %bb.1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.1:
+ ; CHECK-NEXT: successors: %bb.2(0x80000000)
+ ; CHECK-NEXT: liveins: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, $sgpr96_sgpr97_sgpr98_sgpr99, $vcc
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_CMP_EQ_U32 renamable $sgpr96, 0, implicit-def $scc
+ ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 6, $sgpr32, implicit $exec
+ ; CHECK-NEXT: $sgpr32 = V_READFIRSTLANE_B32 $vgpr0, implicit $exec
+ ; CHECK-NEXT: renamable $vcc_lo = S_CSELECT_B32 $sgpr32, renamable $sgpr97, implicit $scc
+ ; CHECK-NEXT: $sgpr32 = S_MUL_I32 $sgpr32, 64
+ ; CHECK-NEXT: S_NOP 0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, implicit killed $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, implicit killed $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, implicit killed $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, implicit killed $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, implicit killed $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, implicit killed $sgpr96_sgpr97_sgpr98_sgpr99
+ ; CHECK-NEXT: S_BRANCH %bb.2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.2:
+ ; CHECK-NEXT: liveins: $vcc
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_ENDPGM 0, implicit $vcc
+ bb.0:
+ S_BRANCH %bb.1
+
+ bb.1:
+ liveins: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, $sgpr96_sgpr97_sgpr98_sgpr99, $vcc
+
+ S_CMP_EQ_U32 renamable $sgpr96, 0, implicit-def $scc
+ renamable $vcc_lo = S_CSELECT_B32 %stack.0, renamable $sgpr97, implicit $scc
+ S_NOP 0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, implicit killed $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, implicit killed $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, implicit killed $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, implicit killed $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, implicit killed $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, implicit killed $sgpr96_sgpr97_sgpr98_sgpr99
+ S_BRANCH %bb.2
+
+ bb.2:
+ liveins: $vcc
+
+ S_ENDPGM 0, implicit $vcc
+...
+
+---
+# As above, but with a non-zero offset and with SCC live *across* the frame index
+# user: S_CBRANCH_SCC1 still needs it after S_CSELECT_B32 has read it. Restoring
+# FrameReg therefore has to fold the offset back without losing SCC, so S_MUL_I32
+# is followed by the S_ADDC_U32/S_BITCMP1_B32/S_BITSET0_B32 sequence that
+# smuggles SCC through bit 0, which the scaling has just cleared.
+name: scc_live_into_scalar_frame_index_no_free_sgpr_nonzero_offset
+tracksRegLiveness: true
+stack:
+ - { id: 0, size: 64, alignment: 4 }
+ - { id: 1, size: 64, alignment: 4 }
+machineFunctionInfo:
+ isEntryFunction: false
+ scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3'
+ frameOffsetReg: '$sgpr33'
+ stackPtrOffsetReg: '$sgpr32'
+ sgprForEXECCopy: '$sgpr100_sgpr101'
+ occupancy: 2
+body: |
+ ; CHECK-LABEL: name: scc_live_into_scalar_frame_index_no_free_sgpr_nonzero_offset
+ ; CHECK: bb.0:
+ ; CHECK-NEXT: successors: %bb.1(0x80000000)
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_def_aspace_cfa $sgpr32, 0, 6
+ ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_register_pair $pc_reg, $sgpr30, 32, $sgpr31, 32
+ ; CHECK-NEXT: S_BRANCH %bb.1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.1:
+ ; CHECK-NEXT: successors: %bb.2(0x80000000)
+ ; CHECK-NEXT: liveins: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, $sgpr96_sgpr97_sgpr98_sgpr99, $vcc
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_CMP_EQ_U32 renamable $sgpr96, 0, implicit-def $scc
+ ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 6, $sgpr32, implicit $exec
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 64, killed $vgpr0, implicit $exec
+ ; CHECK-NEXT: $sgpr32 = V_READFIRSTLANE_B32 $vgpr0, implicit $exec
+ ; CHECK-NEXT: renamable $vcc_lo = S_CSELECT_B32 $sgpr32, renamable $sgpr97, implicit $scc
+ ; CHECK-NEXT: $sgpr32 = S_MUL_I32 $sgpr32, 64
+ ; CHECK-NEXT: $sgpr32 = S_ADDC_U32 $sgpr32, -4096, implicit-def $scc, implicit $scc
+ ; CHECK-NEXT: S_BITCMP1_B32 $sgpr32, 0, implicit-def $scc
+ ; CHECK-NEXT: $sgpr32 = S_BITSET0_B32 0, $sgpr32
+ ; CHECK-NEXT: S_NOP 0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, implicit killed $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, implicit killed $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, implicit killed $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, implicit killed $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, implicit killed $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, implicit killed $sgpr96_sgpr97_sgpr98_sgpr99
+ ; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
+ ; CHECK-NEXT: S_BRANCH %bb.2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.2:
+ ; CHECK-NEXT: liveins: $vcc
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_ENDPGM 0, implicit $vcc
+ bb.0:
+ S_BRANCH %bb.1
+
+ bb.1:
+ liveins: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, $sgpr96_sgpr97_sgpr98_sgpr99, $vcc
+
+ S_CMP_EQ_U32 renamable $sgpr96, 0, implicit-def $scc
+ renamable $vcc_lo = S_CSELECT_B32 %stack.1, renamable $sgpr97, implicit $scc
+ S_NOP 0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, implicit killed $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, implicit killed $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, implicit killed $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, implicit killed $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, implicit killed $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, implicit killed $sgpr96_sgpr97_sgpr98_sgpr99
+ S_CBRANCH_SCC1 %bb.2, implicit $scc
+ S_BRANCH %bb.2
+
+ bb.2:
+ liveins: $vcc
+
+ S_ENDPGM 0, implicit $vcc
+...
+
+---
+# The single-frame-index restriction still applies once SCC is live into the
+# user, so the first frame index falls back to the scavenge. That scavenge
+# emergency-spills and clobbers SCC -- a remaining gap the frame register
+# fallback cannot cover while a second frame index is present. What this pins is
+# that FrameReg is never left scaled underneath it.
+name: two_frame_indices_scc_live_into
+tracksRegLiveness: true
+stack:
+ - { id: 0, size: 64, alignment: 4 }
+ - { id: 1, size: 64, alignment: 4 }
+machineFunctionInfo:
+ isEntryFunction: false
+ scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3'
+ frameOffsetReg: '$sgpr33'
+ stackPtrOffsetReg: '$sgpr32'
+ sgprForEXECCopy: '$sgpr100_sgpr101'
+ occupancy: 2
+body: |
+ ; CHECK-LABEL: name: two_frame_indices_scc_live_into
+ ; CHECK: bb.0:
+ ; CHECK-NEXT: successors: %bb.1(0x80000000)
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_def_aspace_cfa $sgpr32, 0, 6
+ ; CHECK-NEXT: frame-setup CFI_INSTRUCTION llvm_register_pair $pc_reg, $sgpr30, 32, $sgpr31, 32
+ ; CHECK-NEXT: S_BRANCH %bb.1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.1:
+ ; CHECK-NEXT: successors: %bb.2(0x80000000)
+ ; CHECK-NEXT: liveins: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, $sgpr96_sgpr97_sgpr98_sgpr99, $vcc
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_CMP_EQ_U32 renamable $sgpr96, 0, implicit-def $scc
+ ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 6, $sgpr32, implicit $exec
+ ; CHECK-NEXT: $exec = S_NOT_B64 $exec, implicit-def dead $scc, implicit-def $vgpr1
+ ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, implicit $exec :: ("amdgpu-thread-private" store (s32) into %stack.2, addrspace 5)
+ ; CHECK-NEXT: $vgpr1 = SI_SPILL_S32_TO_VGPR killed $sgpr4, 0, undef $vgpr1
+ ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, implicit $exec :: ("amdgpu-thread-private" store (s32) into %stack.2, addrspace 5)
+ ; CHECK-NEXT: $exec = S_NOT_B64 $exec, implicit-def dead $scc
+ ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, implicit $exec :: ("amdgpu-thread-private" store (s32) into %stack.2, addrspace 5)
+ ; CHECK-NEXT: $exec = S_NOT_B64 $exec, implicit-def dead $scc
+ ; CHECK-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, implicit $exec :: ("amdgpu-thread-private" load (s32) from %stack.2, addrspace 5)
+ ; CHECK-NEXT: $exec = S_NOT_B64 $exec, implicit-def dead $scc, implicit killed $vgpr1
+ ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 64, killed $vgpr0, implicit $exec
+ ; CHECK-NEXT: $sgpr4 = V_READFIRSTLANE_B32 $vgpr0, implicit $exec
+ ; CHECK-NEXT: $vgpr2 = V_LSHRREV_B32_e64 6, $sgpr32, implicit $exec
+ ; CHECK-NEXT: $sgpr32 = V_READFIRSTLANE_B32 $vgpr2, implicit $exec
+ ; CHECK-NEXT: renamable $vcc_lo = S_CSELECT_B32 killed $sgpr4, $sgpr32, implicit $scc
+ ; CHECK-NEXT: $sgpr32 = S_MUL_I32 $sgpr32, 64
+ ; CHECK-NEXT: $exec = S_NOT_B64 $exec, implicit-def dead $scc, implicit-def $vgpr0
+ ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, implicit $exec :: ("amdgpu-thread-private" store (s32) into %stack.2, addrspace 5)
+ ; CHECK-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, implicit $exec :: ("amdgpu-thread-private" load (s32) from %stack.2, addrspace 5)
+ ; CHECK-NEXT: $exec = S_NOT_B64 $exec, implicit-def dead $scc
+ ; CHECK-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, implicit $exec :: ("amdgpu-thread-private" load (s32) from %stack.2, addrspace 5)
+ ; CHECK-NEXT: $exec = S_NOT_B64 $exec, implicit-def dead $scc
+ ; CHECK-NEXT: $sgpr4 = SI_RESTORE_S32_FROM_VGPR killed $vgpr0, 0
+ ; CHECK-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, implicit $exec :: ("amdgpu-thread-private" load (s32) from %stack.2, addrspace 5)
+ ; CHECK-NEXT: $exec = S_NOT_B64 $exec, implicit-def dead $scc, implicit killed $vgpr0
+ ; CHECK-NEXT: S_NOP 0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, implicit killed $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, implicit killed $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, implicit killed $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, implicit killed $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, implicit killed $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, implicit killed $sgpr96_sgpr97_sgpr98_sgpr99
+ ; CHECK-NEXT: S_BRANCH %bb.2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.2:
+ ; CHECK-NEXT: liveins: $vcc
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: S_ENDPGM 0, implicit $vcc
+ bb.0:
+ S_BRANCH %bb.1
+
+ bb.1:
+ liveins: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, $sgpr96_sgpr97_sgpr98_sgpr99, $vcc
+
+ S_CMP_EQ_U32 renamable $sgpr96, 0, implicit-def $scc
+ renamable $vcc_lo = S_CSELECT_B32 %stack.1, %stack.0, implicit $scc
+ S_NOP 0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, implicit killed $sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31_sgpr32_sgpr33_sgpr34_sgpr35, implicit killed $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, implicit killed $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, implicit killed $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83, implicit killed $sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95, implicit killed $sgpr96_sgpr97_sgpr98_sgpr99
+ S_BRANCH %bb.2
+
+ bb.2:
+ liveins: $vcc
+
+ S_ENDPGM 0, implicit $vcc
+...
More information about the llvm-branch-commits
mailing list