[llvm-branch-commits] [llvm] [AMDGPU][CodeGen] Allow remat with multiple users in multiple regions (PR #215598)
Lucas Ramirez via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Tue Aug 11 13:53:40 PDT 2026
https://github.com/lucas-rami updated https://github.com/llvm/llvm-project/pull/215598
>From 88afc7b558dc9dda5ab0c41e1318680b18927e0b Mon Sep 17 00:00:00 2001
From: Lucas Ramirez <lucas.rami at proton.me>
Date: Tue, 11 Aug 2026 13:44:59 +0000
Subject: [PATCH] [AMDGPU][CodeGen] Allow remat with multiple users in multiple
regions
This relaxes one of the constraints on rematerialization candidates in
the scheduler's `PreRARematStage`. The current implementation only allows
rematerializing a register if it has users in a single region. This
allows it when a register has multiple users in multiple regions.
In such cases the register is rematerialized as many times as there are
using regions, just before the first user in each using region. The cost
model for assessing rematerialization opportunities now takes into
account that mutliple new instructions may be created for each candidate.
---
llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp | 163 ++-
llvm/lib/Target/AMDGPU/GCNSchedStrategy.h | 4 +-
.../AMDGPU/gfx-callable-argument-types.ll | 46 +-
.../machine-scheduler-sink-trivial-remats.mir | 369 ++---
...ssert-dead-def-subreg-use-other-subreg.mir | 6 +-
.../AMDGPU/sched_mfma_rewrite_copies.mir | 1201 +++++++++--------
.../AMDGPU/sched_mfma_rewrite_cost.mir | 61 +-
.../AMDGPU/sched_mfma_rewrite_diff_types.mir | 29 +-
8 files changed, 1047 insertions(+), 832 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
index 019c5124ff83b..ba886a7e9d1d0 100644
--- a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
@@ -1542,11 +1542,8 @@ bool PreRARematStage::initGCNSchedStage() {
SmallVector<unsigned> CandidateOrder;
for (unsigned RegIdx = 0, E = Remater.getNumRegs(); RegIdx < E; ++RegIdx) {
const Rematerializer::Reg &CandReg = Remater.getReg(RegIdx);
-
- // All users must be in a single region.
- if (CandReg.Uses.size() != 1)
+ if (!CandReg.hasUsersOutsideDefRegion())
continue;
- const auto [UseRegion, Users] = *CandReg.Uses.begin();
// Rematerialization moves the defining instruction into the region of its
// use, which may sit under different control dependencies (e.g., across a
@@ -1557,40 +1554,57 @@ bool PreRARematStage::initGCNSchedStage() {
[](const MachineInstr *DefMI) { return DefMI->isConvergent(); }))
continue;
- // We further filter the registers that we can rematerialize based on our
- // current tracking capabilities in the stage. Users cannot themselves be
- // marked rematerializable, and no register operand of the defining MI can
- // be marked rematerializable. We also do not rematerialize an instruction
- // if it uses registers that aren't available at its use. This ensures that
- // we are not extending any live range while rematerializing.
- if (llvm::any_of(Users, [&MarkedRegs](const MachineInstr *UserMI) {
- assert(UserMI->getNumOperands() > 0 &&
- "user must have at least one operand");
- const MachineOperand &UseMO = UserMI->getOperand(0);
- return UseMO.isReg() && MarkedRegs.contains(UseMO.getReg());
- }))
- continue;
- MachineInstr *FirstUseMI =
- CandReg.getRegionUseBounds(UseRegion, *DAG.LIS).first;
- assert(FirstUseMI && "there must be a user in the region");
- SlotIndex FirstUseIdx =
- DAG.LIS->getInstructionIndex(*FirstUseMI).getRegSlot(true);
SlotIndex RefIdx =
DAG.LIS->getInstructionIndex(*CandReg.getLastDef()).getRegSlot(true);
+
+ // We further filter the registers that we can rematerialize based on our
+ // current tracking capabilities in the stage. This ensures that we are not
+ // extending any live range while rematerializing, and that rematerializing
+ // any register candidate will not affect other candidates.
+
+ // All register dependencies should not already be considered
+ // rematerializable.
if (llvm::any_of(CandReg.Dependencies, [&](RegisterIdx DepRegIdx) {
- const Rematerializer::Reg &DepReg = Remater.getReg(DepRegIdx);
- Register DepDefReg = DepReg.getDefReg();
- return MarkedRegs.contains(DepDefReg) ||
- !Remater.isRegIdenticalAtUses(DepDefReg, DepReg.Mask, RefIdx,
- {FirstUseIdx});
+ return MarkedRegs.contains(Remater.getReg(DepRegIdx).getDefReg());
}))
continue;
- if (llvm::any_of(Remater.getUnrematableDeps(RegIdx),
- [&](const std::pair<Register, LaneBitmask> &RegAndMask) {
- const auto &[Reg, Mask] = RegAndMask;
- return !Remater.isRegIdenticalAtUses(Reg, Mask, RefIdx,
- {FirstUseIdx});
- }))
+
+ auto IsInvalidUsingRegion =
+ [&](const std::pair<unsigned, Rematerializer::Reg::RegionUsers>
+ &RegionUsers) -> bool {
+ const auto &[UseRegion, Users] = RegionUsers;
+ // Users cannot be rematerializable.
+ if (llvm::any_of(Users, [&MarkedRegs](const MachineInstr *UserMI) {
+ assert(UserMI->getNumOperands() > 0 &&
+ "user must have at least one operand");
+ const MachineOperand &UseMO = UserMI->getOperand(0);
+ return UseMO.isReg() && MarkedRegs.contains(UseMO.getReg());
+ }))
+ return true;
+
+ MachineInstr *FirstUseMI =
+ CandReg.getRegionUseBounds(UseRegion, *DAG.LIS).first;
+ assert(FirstUseMI && "there must be a user in the region");
+ SlotIndex FirstUseIdx =
+ DAG.LIS->getInstructionIndex(*FirstUseMI).getRegSlot(true);
+
+ // All dependencies must be available at the first use in the region.
+ if (llvm::any_of(CandReg.Dependencies, [&](RegisterIdx DepRegIdx) {
+ const Rematerializer::Reg &DepReg = Remater.getReg(DepRegIdx);
+ Register DepDefReg = DepReg.getDefReg();
+ return !Remater.isRegIdenticalAtUses(DepDefReg, DepReg.Mask, RefIdx,
+ {FirstUseIdx});
+ }))
+ return true;
+ return llvm::any_of(
+ Remater.getUnrematableDeps(RegIdx),
+ [&](const std::pair<Register, LaneBitmask> &RegAndMask) {
+ const auto &[Reg, Mask] = RegAndMask;
+ return !Remater.isRegIdenticalAtUses(Reg, Mask, RefIdx,
+ {FirstUseIdx});
+ });
+ };
+ if (any_of(CandReg.Uses, IsInvalidUsingRegion))
continue;
MarkedRegs.insert(CandReg.getDefReg());
@@ -1655,8 +1669,8 @@ bool PreRARematStage::initGCNSchedStage() {
CandidateOrder.pop_back();
#ifdef EXPENSIVE_CHECKS
- // All uses are known to be available / live at the remat point. Thus,
- // the uses should already be live in to the using region.
+ // All uses are known to be available / live at the remat points. Thus,
+ // the uses should already be live in to the using regions.
for (const MachineInstr *DefMI : Reg.Defs) {
for (const MachineOperand &MO : DefMI->operands()) {
// Exclude the defined register. We are rematerializing all
@@ -1674,16 +1688,17 @@ bool PreRARematStage::initGCNSchedStage() {
if (LI.hasSubRanges() && MO.getSubReg())
LM = DAG.TRI->getSubRegIndexLaneMask(MO.getSubReg());
- const unsigned UseRegion = Reg.Uses.begin()->first;
- LaneBitmask LiveInMask = DAG.LiveIns[UseRegion].at(UseReg);
- LaneBitmask UncoveredLanes = LM & ~(LiveInMask & LM);
- // If this register has lanes not covered by the LiveIns, be sure they
- // do not map to any subrange. ref:
- // machine-scheduler-sink-trivial-remats.mir::omitted_subrange
- if (UncoveredLanes.any()) {
- assert(LI.hasSubRanges());
- for (LiveInterval::SubRange &SR : LI.subranges())
- assert((SR.LaneMask & UncoveredLanes).none());
+ for (const auto &[UseRegion, _] : Reg.Uses) {
+ LaneBitmask LiveInMask = DAG.LiveIns[UseRegion].at(UseReg);
+ LaneBitmask UncoveredLanes = LM & ~(LiveInMask & LM);
+ // If this register has lanes not covered by the LiveIns, be sure
+ // they do not map to any subrange. ref:
+ // machine-scheduler-sink-trivial-remats.mir::omitted_subrange
+ if (UncoveredLanes.any()) {
+ assert(LI.hasSubRanges());
+ for (LiveInterval::SubRange &SR : LI.subranges())
+ assert((SR.LaneMask & UncoveredLanes).none());
+ }
}
}
}
@@ -3066,32 +3081,46 @@ void PreRARematStage::ScoredRemat::init(RegisterIdx RegIdx,
const Rematerializer::Reg &Reg = Remater.getReg(RegIdx);
Register DefReg = Reg.getDefReg();
- assert(Reg.Uses.size() == 1 && "expected users in single region");
- const unsigned UseRegion = Reg.Uses.begin()->first;
// Mark regions in which the rematerializable register is live.
for (unsigned I = 0, E = NumRegions; I != E; ++I) {
- if (DAG.LiveIns[I].contains(DefReg))
+ bool LiveInRegion = false;
+ if (DAG.LiveIns[I].contains(DefReg)) {
LiveIn.set(I);
- if (DAG.RegionLiveOuts.getLiveRegsForRegionIdx(I).contains(DefReg))
+ LiveInRegion = true;
+ }
+ if (DAG.RegionLiveOuts.getLiveRegsForRegionIdx(I).contains(DefReg)) {
LiveOut.set(I);
+ LiveInRegion = true;
+ }
+
+ if (!LiveInRegion)
+ continue;
+ Live.set(I);
// If the register is both unused and live-through in the region, the
// latter's RP is guaranteed to decrease.
- if (!LiveIn[I] || !LiveOut[I] || I == UseRegion)
+ if (!LiveIn[I] || !LiveOut[I] || Reg.Uses.contains(I))
UnpredictableRPSave.set(I);
}
- Live |= LiveIn;
- Live |= LiveOut;
RPSave.inc(DefReg, LaneBitmask::getNone(), Reg.Mask, DAG.MRI);
- // Get frequencies of defining and using regions. A rematerialization from the
- // least frequent region to the most frequent region will yield the greatest
- // in order to penalize rematerializations from or into regions whose
- int64_t DefOrMin = std::max(Freq.Regions[Reg.DefRegion], Freq.MinFreq);
- int64_t UseOrMax = Freq.Regions[UseRegion];
- if (!UseOrMax)
- UseOrMax = Freq.MaxFreq;
+ // Estimate the difference in instruction execution frequency that will
+ // induced by the rematerialization. A rematerialization from the least
+ // frequent region to the most frequent region should yield the greatest
+ // frequency penalty (most negative frequency difference). When the register
+ // has users in its defining region the original instruction(s) won't be
+ // deleted so there is no frequency gain in the defining region.
+ int64_t DefOrMin = Reg.hasUsersInDefRegion()
+ ? 0
+ : std::max(Freq.Regions[Reg.DefRegion], Freq.MinFreq);
+ int64_t UseOrMax = 0;
+ for (const auto &[UseRegion, _] : Reg.Uses) {
+ if (Reg.DefRegion == UseRegion)
+ continue;
+ uint64_t RegionFreq = Freq.Regions[UseRegion];
+ UseOrMax += RegionFreq != 0 ? RegionFreq : Freq.MaxFreq;
+ }
FreqDiff = DefOrMin - UseOrMax;
}
@@ -3131,11 +3160,23 @@ void PreRARematStage::ScoredRemat::update(const BitVector &TargetRegions,
void PreRARematStage::ScoredRemat::rematerialize(
Rematerializer &Remater) const {
const Rematerializer::Reg &Reg = Remater.getReg(RegIdx);
+
+ // All rematerializable dependencies are re-used for all rematerializations.
Rematerializer::DependencyReuseInfo DRI;
for (RegisterIdx DepRegIdx : Reg.Dependencies)
DRI.reuse(DepRegIdx);
- unsigned UseRegion = Reg.Uses.begin()->first;
- Remater.rematerializeToRegion(RegIdx, UseRegion, DRI);
+
+ // Rematerialize once per using region, except in the defining region.
+ SmallVector<unsigned, 8> RematRegions;
+ RematRegions.reserve(Reg.Uses.size());
+ for (const auto &[UseRegion, _] : Reg.Uses) {
+ if (Reg.DefRegion != UseRegion)
+ RematRegions.push_back(UseRegion);
+ }
+ for (const unsigned UseRegion : RematRegions) {
+ Rematerializer::DependencyReuseInfo RegionDRI(DRI);
+ Remater.rematerializeToRegion(RegIdx, UseRegion, RegionDRI);
+ }
}
void PreRARematStage::updateRPTargets(const BitVector &Regions,
diff --git a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.h b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.h
index 2059f4e6479ff..7f4cf8ac23e6b 100644
--- a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.h
+++ b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.h
@@ -551,9 +551,7 @@ class ClusteredLowOccStage : public GCNSchedStage {
/// non-ignorable physical register use.
/// 3 The register has no virtual register use whose live range would be
/// extended by the rematerialization.
-/// 4. The register has a single non-debug user in a different region from its
-/// defining region.
-/// 5. The register is not used by or using another register that is going to be
+/// 4. The register is not used by or using another register that is going to be
/// rematerialized.
class PreRARematStage : public GCNSchedStage {
private:
diff --git a/llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll b/llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll
index 65d3608744000..05fd82b9c9377 100644
--- a/llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll
+++ b/llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll
@@ -5598,23 +5598,21 @@ define amdgpu_gfx void @test_call_external_void_func_v32i8_ret() #0 {
; GFX9-NEXT: s_mov_b32 s34, s33
; GFX9-NEXT: s_mov_b32 s33, s32
; GFX9-NEXT: s_or_saveexec_b64 s[36:37], -1
-; GFX9-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
; GFX9-NEXT: s_mov_b64 exec, s[36:37]
-; GFX9-NEXT: v_writelane_b32 v44, s34, 2
-; GFX9-NEXT: s_addk_i32 s32, 0x800
-; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
-; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
-; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
-; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 ; 4-byte Folded Spill
-; GFX9-NEXT: v_writelane_b32 v44, s30, 0
-; GFX9-NEXT: v_writelane_b32 v44, s31, 1
-; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: v_writelane_b32 v42, s34, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: v_writelane_b32 v42, s30, 0
+; GFX9-NEXT: v_writelane_b32 v42, s31, 1
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 16
; GFX9-NEXT: v_mov_b32_e32 v41, 0
-; GFX9-NEXT: v_mov_b32_e32 v42, 16
-; GFX9-NEXT: v_mov_b32_e32 v43, 0
-; GFX9-NEXT: global_load_dwordx4 v[0:3], v[40:41], off
-; GFX9-NEXT: global_load_dwordx4 v[16:19], v[42:43], off
+; GFX9-NEXT: global_load_dwordx4 v[0:3], v[0:1], off
; GFX9-NEXT: s_mov_b32 s35, external_void_func_v3i8_ret at abs32@hi
+; GFX9-NEXT: global_load_dwordx4 v[16:19], v[40:41], off
; GFX9-NEXT: s_mov_b32 s34, external_void_func_v3i8_ret at abs32@lo
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_lshrrev_b32_e32 v35, 8, v0
@@ -5680,19 +5678,19 @@ define amdgpu_gfx void @test_call_external_void_func_v32i8_ret() #0 {
; GFX9-NEXT: v_lshl_or_b32 v0, v4, 16, v0
; GFX9-NEXT: v_lshl_or_b32 v9, v13, 16, v12
; GFX9-NEXT: v_lshl_or_b32 v8, v10, 16, v8
-; GFX9-NEXT: global_store_dwordx4 v[42:43], v[0:3], off
-; GFX9-NEXT: global_store_dwordx4 v[40:41], v[6:9], off
-; GFX9-NEXT: buffer_load_dword v43, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 0
+; GFX9-NEXT: global_store_dwordx4 v[40:41], v[0:3], off
+; GFX9-NEXT: global_store_dwordx4 v[4:5], v[6:9], off
+; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
; GFX9-NEXT: s_nop 0
-; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
-; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
-; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
-; GFX9-NEXT: v_readlane_b32 s30, v44, 0
-; GFX9-NEXT: v_readlane_b32 s31, v44, 1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GFX9-NEXT: v_readlane_b32 s30, v42, 0
+; GFX9-NEXT: v_readlane_b32 s31, v42, 1
; GFX9-NEXT: s_mov_b32 s32, s33
-; GFX9-NEXT: v_readlane_b32 s34, v44, 2
+; GFX9-NEXT: v_readlane_b32 s34, v42, 2
; GFX9-NEXT: s_or_saveexec_b64 s[36:37], -1
-; GFX9-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
+; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
; GFX9-NEXT: s_mov_b64 exec, s[36:37]
; GFX9-NEXT: s_mov_b32 s33, s34
; GFX9-NEXT: s_waitcnt vmcnt(0)
diff --git a/llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir b/llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir
index 8c9e4a5a26c83..5a741777a7c75 100644
--- a/llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir
+++ b/llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir
@@ -3997,17 +3997,18 @@ body: |
; GFX908-NEXT: [[V_CVT_I32_F64_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 20, implicit $exec, implicit $mode, implicit-def $m0
; GFX908-NEXT: [[V_CVT_I32_F64_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 21, implicit $exec, implicit $mode, implicit-def $m0
; GFX908-NEXT: [[V_CVT_I32_F64_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 22, implicit $exec, implicit $mode, implicit-def $m0
- ; GFX908-NEXT: [[V_CVT_I32_F64_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.1:
; GFX908-NEXT: successors: %bb.2(0x80000000)
; GFX908-NEXT: {{ $}}
- ; GFX908-NEXT: [[V_CVT_I32_F64_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_24]]
+ ; GFX908-NEXT: [[V_CVT_I32_F64_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
+ ; GFX908-NEXT: [[V_CVT_I32_F64_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_23]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_24]]
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.2:
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_23]]
+ ; GFX908-NEXT: [[V_CVT_I32_F64_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_25]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_]], implicit [[V_CVT_I32_F64_e32_1]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_2]], implicit [[V_CVT_I32_F64_e32_3]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_4]], implicit [[V_CVT_I32_F64_e32_5]]
@@ -4049,17 +4050,18 @@ body: |
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 20, implicit $exec, implicit $mode, implicit-def $m0
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 21, implicit $exec, implicit $mode, implicit-def $m0
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 22, implicit $exec, implicit $mode, implicit-def $m0
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.1:
; GFX908-GCNTRACKERS-NEXT: successors: %bb.2(0x80000000)
; GFX908-GCNTRACKERS-NEXT: {{ $}}
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_24]]
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_23]]
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_24]]
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.2:
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_23]]
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_25]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_]], implicit [[V_CVT_I32_F64_e32_1]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_2]], implicit [[V_CVT_I32_F64_e32_3]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_4]], implicit [[V_CVT_I32_F64_e32_5]]
@@ -11112,13 +11114,14 @@ body: |
; GFX908-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
; GFX908-NEXT: S_BRANCH %bb.1
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.1:
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_11]]
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF30]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF29]]
+ ; GFX908-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF31]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF30]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_12]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_5]], implicit [[V_CVT_I32_F32_e32_13]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_6]], implicit [[V_CVT_I32_F32_e32_14]]
@@ -11206,17 +11209,18 @@ body: |
; GFX908-GCNTRACKERS-NEXT: [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 $vgpr8, implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 $vgpr9, implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 $vgpr8, implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 $vgpr9, implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: S_BRANCH %bb.1
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.1:
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_9]]
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_28]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF30]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_29]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF29]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_29]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF31]]
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_30]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF30]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_10]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_11]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_12]]
@@ -11702,6 +11706,8 @@ body: |
; GFX908-NEXT: [[DEF27:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
@@ -11713,19 +11719,19 @@ body: |
; GFX908-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
; GFX908-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
; GFX908-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
; GFX908-NEXT: S_BRANCH %bb.1
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.1:
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]]
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_30]], implicit [[V_CVT_I32_F32_e32_25]], implicit [[DEF30]], implicit [[DEF25]]
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_31]], implicit [[V_CVT_I32_F32_e32_26]], implicit [[DEF31]], implicit [[DEF26]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF22]], implicit [[DEF27]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF23]], implicit [[DEF28]]
+ ; GFX908-NEXT: [[DEF32:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF32]], implicit [[DEF27]]
+ ; GFX908-NEXT: [[DEF33:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF33]], implicit [[DEF28]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF24]], implicit [[DEF29]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF25]], implicit [[DEF30]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_26]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF26]], implicit [[DEF31]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_7]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_8]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_9]]
@@ -11799,6 +11805,8 @@ body: |
; GFX908-GCNTRACKERS-NEXT: [[DEF27:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
@@ -11810,19 +11818,19 @@ body: |
; GFX908-GCNTRACKERS-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
; GFX908-GCNTRACKERS-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: S_BRANCH %bb.1
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.1:
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]]
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_30]], implicit [[V_CVT_I32_F32_e32_25]], implicit [[DEF30]], implicit [[DEF25]]
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_31]], implicit [[V_CVT_I32_F32_e32_26]], implicit [[DEF31]], implicit [[DEF26]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF22]], implicit [[DEF27]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF23]], implicit [[DEF28]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF32:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF32]], implicit [[DEF27]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF33:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF33]], implicit [[DEF28]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF24]], implicit [[DEF29]]
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF25]], implicit [[DEF30]]
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_26]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF26]], implicit [[DEF31]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_7]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_8]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_9]]
@@ -12005,18 +12013,20 @@ body: |
; GFX908-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
; GFX908-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
+ ; GFX908-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
; GFX908-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
- ; GFX908-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
- ; GFX908-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
- ; GFX908-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
; GFX908-NEXT: S_BRANCH %bb.1
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.1:
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]], implicit [[DEF22]], implicit [[DEF27]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF23]], implicit [[DEF28]]
+ ; GFX908-NEXT: [[DEF32:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]], implicit [[DEF32]], implicit [[DEF27]]
+ ; GFX908-NEXT: [[DEF33:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF33]], implicit [[DEF28]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF24]], implicit [[DEF29]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF25]], implicit [[DEF30]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF26]], implicit [[DEF31]]
@@ -12102,18 +12112,20 @@ body: |
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
+ ; GFX908-GCNTRACKERS-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
- ; GFX908-GCNTRACKERS-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
- ; GFX908-GCNTRACKERS-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
; GFX908-GCNTRACKERS-NEXT: S_BRANCH %bb.1
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.1:
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]], implicit [[DEF22]], implicit [[DEF27]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF23]], implicit [[DEF28]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF32:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]], implicit [[DEF32]], implicit [[DEF27]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF33:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF33]], implicit [[DEF28]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF24]], implicit [[DEF29]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF25]], implicit [[DEF30]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF26]], implicit [[DEF31]]
@@ -12288,10 +12300,12 @@ body: |
; GFX908-NEXT: [[V_CVT_I32_F32_e32_18:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF20]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[V_CVT_I32_F32_e32_19:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF21]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub0, implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub2, implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF26:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
@@ -12299,26 +12313,28 @@ body: |
; GFX908-NEXT: [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub0, implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub2, implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
; GFX908-NEXT: S_BRANCH %bb.1
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.1:
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_26]], implicit [[DEF]].sub0, implicit [[DEF]].sub2
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_31]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF31]], implicit [[DEF27]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF24]], implicit [[DEF28]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF25]], implicit [[DEF29]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF26]], implicit [[DEF30]]
+ ; GFX908-NEXT: undef [[DEF32:%[0-9]+]].sub0:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[DEF32:%[0-9]+]].sub2:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_20]], implicit [[V_CVT_I32_F32_e32_21]], implicit [[DEF32]].sub0, implicit [[DEF32]].sub2
+ ; GFX908-NEXT: undef [[DEF32:%[0-9]+]].sub1:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: dead undef [[DEF32:%[0-9]+]].sub3:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF24]], implicit [[DEF28]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF25]], implicit [[DEF29]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_26]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF26]], implicit [[DEF30]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_27]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF27]], implicit [[DEF31]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_7]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_8]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_9]]
@@ -12328,8 +12344,8 @@ body: |
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_13]], implicit [[V_CVT_I32_F32_e32_14]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_15]], implicit [[V_CVT_I32_F32_e32_16]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_17]], implicit [[V_CVT_I32_F32_e32_18]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_19]], implicit [[V_CVT_I32_F32_e32_20]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_21]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_19]], implicit [[V_CVT_I32_F32_e32_22]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]]
; GFX908-NEXT: S_ENDPGM 0
;
; GFX908-GCNTRACKERS-LABEL: name: remat_virtual_vgpr_subreg_occ_6
@@ -12391,34 +12407,37 @@ body: |
; GFX908-GCNTRACKERS-NEXT: [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF26:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF27:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]].sub0, implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]].sub2, implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]].sub2, implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
- ; GFX908-GCNTRACKERS-NEXT: [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
; GFX908-GCNTRACKERS-NEXT: [[DEF22:%[0-9]+]].sub1:vreg_512 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF22:%[0-9]+]].sub3:vreg_512 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: dead [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: dead [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
- ; GFX908-GCNTRACKERS-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: S_BRANCH %bb.1
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.1:
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_24]], implicit [[DEF22]].sub0, implicit [[DEF22]].sub2
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_29]], implicit [[V_CVT_I32_F32_e32_25]], implicit [[DEF28]], implicit [[DEF24]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_30]], implicit [[V_CVT_I32_F32_e32_26]], implicit [[DEF30]], implicit [[DEF25]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_31]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF31]], implicit [[DEF26]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF23]], implicit [[DEF27]]
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF22]].sub0, implicit [[DEF22]].sub2
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF32:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF32]], implicit [[DEF27]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF33:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF33]], implicit [[DEF28]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF34:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF34]], implicit [[DEF29]]
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_26]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF26]], implicit [[DEF30]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_7]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_8]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_9]]
@@ -12592,10 +12611,12 @@ body: |
; GFX908-NEXT: [[V_CVT_I32_F32_e32_18:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF20]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[V_CVT_I32_F32_e32_19:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF21]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub0, implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub2, implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF26:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
@@ -12604,12 +12625,10 @@ body: |
; GFX908-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub0, implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub2, implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
; GFX908-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
@@ -12617,12 +12636,17 @@ body: |
; GFX908-NEXT: S_BRANCH %bb.1
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.1:
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF]].sub1, implicit [[DEF]].sub3
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]], implicit [[DEF24]], implicit [[DEF28]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF25]], implicit [[DEF29]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF26]], implicit [[DEF30]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF27]], implicit [[DEF31]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_26]], implicit [[V_CVT_I32_F32_e32_31]]
+ ; GFX908-NEXT: undef [[DEF32:%[0-9]+]].sub1:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[DEF32:%[0-9]+]].sub3:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_20]], implicit [[V_CVT_I32_F32_e32_21]], implicit [[DEF32]].sub1, implicit [[DEF32]].sub3
+ ; GFX908-NEXT: [[DEF33:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_6]], implicit [[DEF33]], implicit [[DEF28]]
+ ; GFX908-NEXT: undef [[DEF32:%[0-9]+]].sub0:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: dead undef [[DEF32:%[0-9]+]].sub2:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF25]], implicit [[DEF29]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF26]], implicit [[DEF30]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_26]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF27]], implicit [[DEF31]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_27]], implicit [[V_CVT_I32_F32_e32_31]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_7]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_8]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_9]]
@@ -12632,8 +12656,8 @@ body: |
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_13]], implicit [[V_CVT_I32_F32_e32_14]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_15]], implicit [[V_CVT_I32_F32_e32_16]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_17]], implicit [[V_CVT_I32_F32_e32_18]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_19]], implicit [[V_CVT_I32_F32_e32_20]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_21]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_19]], implicit [[V_CVT_I32_F32_e32_22]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]]
; GFX908-NEXT: S_ENDPGM 0
;
; GFX908-GCNTRACKERS-LABEL: name: remat_virtual_vgpr_subreg_uses_not_available
@@ -12705,20 +12729,21 @@ body: |
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
; GFX908-GCNTRACKERS-NEXT: dead [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
; GFX908-GCNTRACKERS-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: undef [[DEF:%[0-9]+]].sub1:vreg_512 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF:%[0-9]+]].sub3:vreg_512 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: S_BRANCH %bb.1
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.1:
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_7]], implicit [[DEF]].sub1, implicit [[DEF]].sub3
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_8]], implicit [[DEF23]], implicit [[DEF27]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF32:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_8]], implicit [[DEF32]], implicit [[DEF27]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF24]], implicit [[DEF28]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF25]], implicit [[DEF29]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_26]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF26]], implicit [[DEF30]]
@@ -12891,18 +12916,20 @@ body: |
; GFX908-NEXT: [[V_CVT_I32_F32_e32_15:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF17]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[V_CVT_I32_F32_e32_16:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF18]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_17:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub2, implicit $exec, implicit $mode
+ ; GFX908-NEXT: undef [[V_CVT_I32_F32_e32_1:%[0-9]+]].sub0:vreg_64 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub0, implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_17:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF19]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_18:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF19]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_18:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF20]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_19:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF20]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_19:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF21]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF21]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
; GFX908-NEXT: [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF26:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF27:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
@@ -12910,35 +12937,40 @@ body: |
; GFX908-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub2, implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
- ; GFX908-NEXT: undef [[V_CVT_I32_F32_e32_1:%[0-9]+]].sub0:vreg_64 = nofpexcept V_CVT_I32_F32_e32 [[DEF]].sub0, implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
+ ; GFX908-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
; GFX908-NEXT: S_BRANCH %bb.1
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.1:
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_7]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_8]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]].sub0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[DEF]].sub0, implicit [[DEF]].sub2
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
- ; GFX908-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_27]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF1]], implicit [[DEF28]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_27]], implicit [[V_CVT_I32_F32_e32_24]], implicit [[DEF25]], implicit [[DEF29]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_29]], implicit [[V_CVT_I32_F32_e32_25]], implicit [[DEF26]], implicit [[DEF30]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_30]], implicit [[V_CVT_I32_F32_e32_26]], implicit [[DEF27]], implicit [[DEF31]]
+ ; GFX908-NEXT: undef [[DEF32:%[0-9]+]].sub0:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[DEF32:%[0-9]+]].sub2:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]].sub0, implicit [[V_CVT_I32_F32_e32_17]], implicit [[DEF32]].sub0, implicit [[DEF32]].sub2
+ ; GFX908-NEXT: undef [[DEF32:%[0-9]+]].sub1:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: dead undef [[DEF32:%[0-9]+]].sub3:vreg_512 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[DEF33:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF33]], implicit [[DEF28]]
+ ; GFX908-NEXT: [[DEF34:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF34]], implicit [[DEF29]]
+ ; GFX908-NEXT: [[DEF35:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF35]], implicit [[DEF30]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_26]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF27]], implicit [[DEF31]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_9]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_5]], implicit [[V_CVT_I32_F32_e32_10]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_6]], implicit [[V_CVT_I32_F32_e32_11]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_12]], implicit [[V_CVT_I32_F32_e32_13]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_14]], implicit [[V_CVT_I32_F32_e32_15]]
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_14]], implicit [[V_CVT_I32_F32_e32_15]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_16]], implicit [[V_CVT_I32_F32_e32_17]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_18]], implicit [[V_CVT_I32_F32_e32_19]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_20]], implicit [[V_CVT_I32_F32_e32_21]]
- ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_22]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_16]], implicit [[V_CVT_I32_F32_e32_18]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_19]], implicit [[V_CVT_I32_F32_e32_20]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_21]], implicit [[V_CVT_I32_F32_e32_22]]
+ ; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]]
; GFX908-NEXT: S_ENDPGM 0
;
; GFX908-GCNTRACKERS-LABEL: name: remat_virtual_vgpr_subreg_def
@@ -12949,13 +12981,14 @@ body: |
; GFX908-GCNTRACKERS-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr0_sgpr1
; GFX908-GCNTRACKERS-NEXT: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
; GFX908-GCNTRACKERS-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 52, 0 :: (dereferenceable invariant load (s64), align 4, addrspace 4)
+ ; GFX908-GCNTRACKERS-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
; GFX908-GCNTRACKERS-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF2:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF3:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF4:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF5:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: [[DEF6:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: dead [[DEF6:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF7:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF8:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF9:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
@@ -12968,65 +13001,67 @@ body: |
; GFX908-GCNTRACKERS-NEXT: [[DEF16:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: dead [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF18]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: dead [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF19]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
+ ; GFX908-GCNTRACKERS-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
; GFX908-GCNTRACKERS-NEXT: undef [[V_CVT_I32_F32_e32_1:%[0-9]+]].sub1:vreg_64 = nofpexcept V_CVT_I32_F32_e32 [[DEF]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_2:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF1]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_3:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF2]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_4:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF3]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_5:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF4]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_6:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF5]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_7:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF6]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_8:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF7]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_9:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF8]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_10:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF9]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_11:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF10]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_12:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF11]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_13:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF12]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_14:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF13]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_15:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF14]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_16:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF15]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_17:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF16]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_18:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF17]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_19:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF19]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF20]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF21]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_7:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF7]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_8:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF8]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_9:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF9]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_10:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF10]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_11:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF11]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_12:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF12]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_13:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF13]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_14:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF14]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_15:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF15]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_16:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF16]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_17:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF17]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_18:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF18]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_19:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF20]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF21]], implicit $exec, implicit $mode
; GFX908-GCNTRACKERS-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: undef [[DEF23:%[0-9]+]].sub0:vreg_512 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: undef [[DEF24:%[0-9]+]].sub0:vreg_512 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF26:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF27:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: undef [[V_CVT_I32_F32_e32_1:%[0-9]+]].sub0:vreg_64 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]].sub0, implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
- ; GFX908-GCNTRACKERS-NEXT: dead [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
- ; GFX908-GCNTRACKERS-NEXT: [[DEF23:%[0-9]+]].sub2:vreg_512 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[DEF31:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: undef [[V_CVT_I32_F32_e32_1:%[0-9]+]].sub0:vreg_64 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]].sub0, implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF24:%[0-9]+]].sub2:vreg_512 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: S_BRANCH %bb.1
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.1:
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_7]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_8]]
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]].sub2, implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]].sub0, implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF23]].sub0, implicit [[DEF23]].sub2
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF31]], implicit $exec, implicit $mode
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_28]], implicit [[V_CVT_I32_F32_e32_23]], implicit [[DEF]], implicit [[DEF24]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_28]], implicit [[V_CVT_I32_F32_e32_24]], implicit [[DEF29]], implicit [[DEF25]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_29]], implicit [[V_CVT_I32_F32_e32_25]], implicit [[DEF30]], implicit [[DEF26]]
- ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_30]], implicit [[V_CVT_I32_F32_e32_26]], implicit [[DEF31]], implicit [[DEF27]]
+ ; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]].sub2, implicit $exec, implicit $mode
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]].sub0, implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF24]].sub0, implicit [[DEF24]].sub2
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF32:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF33:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_26]], implicit [[DEF32]], implicit [[DEF28]]
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_23]], implicit [[V_CVT_I32_F32_e32_27]], implicit [[DEF33]], implicit [[DEF29]]
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF34:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_24]], implicit [[V_CVT_I32_F32_e32_28]], implicit [[DEF34]], implicit [[DEF30]]
+ ; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_25]], implicit [[V_CVT_I32_F32_e32_29]], implicit [[DEF27]], implicit [[DEF31]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_9]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_5]], implicit [[V_CVT_I32_F32_e32_10]]
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F32_e32_6]], implicit [[V_CVT_I32_F32_e32_11]]
@@ -13195,7 +13230,7 @@ body: |
; GFX908-NEXT: [[V_CVT_I32_F64_e32_41:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 51, implicit $exec, implicit $mode, implicit-def $m0
; GFX908-NEXT: [[V_CVT_I32_F64_e32_42:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 52, implicit $exec, implicit $mode, implicit-def $m0
; GFX908-NEXT: undef [[V_RCP_F32_e32_:%[0-9]+]].sub0:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_RCP_F32_e32 [[DEF]].sub0, implicit $mode, implicit $exec
- ; GFX908-NEXT: [[DEF2:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[V_PK_MUL_F32_:%[0-9]+]]:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_PK_MUL_F32 0, [[V_RCP_F32_e32_]], 8, [[DEF]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.1:
; GFX908-NEXT: successors: %bb.2(0x80000000)
@@ -13212,10 +13247,10 @@ body: |
; GFX908-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_9]], implicit [[V_CVT_I32_F64_e32_19]], implicit [[V_CVT_I32_F64_e32_29]], implicit [[V_CVT_I32_F64_e32_39]]
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: bb.2:
- ; GFX908-NEXT: dead [[DEF3:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; GFX908-NEXT: undef [[V_FMA_F32_e64_:%[0-9]+]].sub0:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_FMA_F32_e64 0, [[DEF2]].sub1, 0, [[V_RCP_F32_e32_]].sub0, 0, [[DEF2]].sub0, 0, 0, implicit $mode, implicit $exec
+ ; GFX908-NEXT: dead [[DEF2:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; GFX908-NEXT: [[DEF3:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; GFX908-NEXT: undef [[V_FMA_F32_e64_:%[0-9]+]].sub0:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_FMA_F32_e64 0, [[DEF3]].sub1, 0, [[V_RCP_F32_e32_]].sub0, 0, [[DEF3]].sub0, 0, 0, implicit $mode, implicit $exec
; GFX908-NEXT: %temp2:vreg_64_align2 = IMPLICIT_DEF
- ; GFX908-NEXT: [[V_PK_MUL_F32_:%[0-9]+]]:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_PK_MUL_F32 0, [[V_RCP_F32_e32_]], 8, [[DEF]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX908-NEXT: dead [[V_PK_FMA_F32_:%[0-9]+]]:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_PK_FMA_F32 0, [[V_FMA_F32_e64_]], 8, %temp2, 11, [[V_PK_MUL_F32_]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX908-NEXT: S_NOP 0, implicit [[DEF]], implicit [[V_RCP_F32_e32_]]
; GFX908-NEXT: S_ENDPGM 0
@@ -13271,7 +13306,7 @@ body: |
; GFX908-GCNTRACKERS-NEXT: dead [[DEF1:%[0-9]+]]:vreg_64 = IMPLICIT_DEF
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_41:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 51, implicit $exec, implicit $mode, implicit-def $m0
; GFX908-GCNTRACKERS-NEXT: [[V_CVT_I32_F64_e32_42:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 52, implicit $exec, implicit $mode, implicit-def $m0
- ; GFX908-GCNTRACKERS-NEXT: [[DEF2:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[V_PK_MUL_F32_:%[0-9]+]]:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_PK_MUL_F32 0, [[V_RCP_F32_e32_]], 8, [[DEF]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.1:
; GFX908-GCNTRACKERS-NEXT: successors: %bb.2(0x80000000)
@@ -13288,10 +13323,10 @@ body: |
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[V_CVT_I32_F64_e32_9]], implicit [[V_CVT_I32_F64_e32_19]], implicit [[V_CVT_I32_F64_e32_29]], implicit [[V_CVT_I32_F64_e32_39]]
; GFX908-GCNTRACKERS-NEXT: {{ $}}
; GFX908-GCNTRACKERS-NEXT: bb.2:
- ; GFX908-GCNTRACKERS-NEXT: dead [[DEF3:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: undef [[V_FMA_F32_e64_:%[0-9]+]].sub0:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_FMA_F32_e64 0, [[DEF2]].sub1, 0, [[V_RCP_F32_e32_]].sub0, 0, [[DEF2]].sub0, 0, 0, implicit $mode, implicit $exec
+ ; GFX908-GCNTRACKERS-NEXT: dead [[DEF2:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: [[DEF3:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; GFX908-GCNTRACKERS-NEXT: undef [[V_FMA_F32_e64_:%[0-9]+]].sub0:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_FMA_F32_e64 0, [[DEF3]].sub1, 0, [[V_RCP_F32_e32_]].sub0, 0, [[DEF3]].sub0, 0, 0, implicit $mode, implicit $exec
; GFX908-GCNTRACKERS-NEXT: %temp2:vreg_64_align2 = IMPLICIT_DEF
- ; GFX908-GCNTRACKERS-NEXT: [[V_PK_MUL_F32_:%[0-9]+]]:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_PK_MUL_F32 0, [[V_RCP_F32_e32_]], 8, [[DEF]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX908-GCNTRACKERS-NEXT: dead [[V_PK_FMA_F32_:%[0-9]+]]:vreg_64_align2 = nnan ninf nsz arcp contract afn reassoc nofpexcept V_PK_FMA_F32 0, [[V_FMA_F32_e64_]], 8, %temp2, 11, [[V_PK_MUL_F32_]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX908-GCNTRACKERS-NEXT: S_NOP 0, implicit [[DEF]], implicit [[V_RCP_F32_e32_]]
; GFX908-GCNTRACKERS-NEXT: S_ENDPGM 0
diff --git a/llvm/test/CodeGen/AMDGPU/sched-assert-dead-def-subreg-use-other-subreg.mir b/llvm/test/CodeGen/AMDGPU/sched-assert-dead-def-subreg-use-other-subreg.mir
index d511b47a4f864..dc515af825b4e 100644
--- a/llvm/test/CodeGen/AMDGPU/sched-assert-dead-def-subreg-use-other-subreg.mir
+++ b/llvm/test/CodeGen/AMDGPU/sched-assert-dead-def-subreg-use-other-subreg.mir
@@ -24,7 +24,7 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: undef [[V_MOV_B32_e32_:%[0-9]+]].sub3:vreg_512 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
- ; CHECK-NEXT: [[V_MOV_B32_e32_2:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 0, [[V_MOV_B32_e32_1]], implicit $exec
; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_512 = COPY [[V_MOV_B32_e32_]]
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
@@ -33,17 +33,17 @@ body: |
; CHECK-NEXT: BUFFER_STORE_DWORD_OFFEN [[V_MOV_B32_e32_]].sub3, undef %5:vgpr_32, $sgpr24_sgpr25_sgpr26_sgpr27, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32), align 8, addrspace 5)
; CHECK-NEXT: dead [[DS_READ_B32_gfx9_:%[0-9]+]]:vgpr_32 = DS_READ_B32_gfx9 undef %7:vgpr_32, 0, 0, implicit $exec
; CHECK-NEXT: dead [[DS_READ_B64_gfx9_:%[0-9]+]]:vreg_64 = DS_READ_B64_gfx9 [[V_MOV_B32_e32_1]], 0, 0, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 0, [[V_MOV_B32_e32_1]], implicit $exec
; CHECK-NEXT: dead [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128 = DS_READ_B128_gfx9 [[V_ADD_U32_e32_]], 0, 0, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; CHECK-NEXT: undef [[COPY2:%[0-9]+]].sub1:vreg_512 = COPY [[COPY]].sub1
; CHECK-NEXT: INLINEASM &"", sideeffect attdialect, regdef:VGPR_32, def dead [[COPY1]], regdef:VGPR_32, def dead [[COPY]].sub1, reguse:VGPR_32, [[COPY1]], reguse:VGPR_32, [[COPY]].sub1
; CHECK-NEXT: [[COPY2:%[0-9]+]].sub0:vreg_512 = COPY [[COPY]].sub0
; CHECK-NEXT: [[COPY2:%[0-9]+]].sub3:vreg_512 = COPY [[COPY]].sub3
+ ; CHECK-NEXT: [[V_MOV_B32_e32_2:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+ ; CHECK-NEXT: dead [[V_ADD_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_CO_U32_e32 4, [[V_MOV_B32_e32_2]], implicit-def dead $vcc, implicit $exec
; CHECK-NEXT: [[COPY2:%[0-9]+]].sub2:vreg_512 = COPY undef [[V_MOV_B32_e32_1]]
; CHECK-NEXT: [[COPY2:%[0-9]+]].sub5:vreg_512 = COPY undef [[V_MOV_B32_e32_1]]
; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_512 = COPY [[COPY2]]
- ; CHECK-NEXT: dead [[V_ADD_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_CO_U32_e32 4, [[V_MOV_B32_e32_2]], implicit-def dead $vcc, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.1
bb.0:
liveins: $sgpr6_sgpr7
diff --git a/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_copies.mir b/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_copies.mir
index a4a94089d2f7b..d4150df4fbe4b 100644
--- a/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_copies.mir
+++ b/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_copies.mir
@@ -248,6 +248,7 @@ body: |
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
@@ -255,17 +256,18 @@ body: |
; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF18]], implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF19]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -352,16 +354,17 @@ body: |
; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF18]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -439,37 +442,39 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub0, [[DEF13]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub1, [[DEF14]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF19]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -553,38 +558,40 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub0, [[DEF13]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub1, [[DEF14]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF12]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF19]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -673,10 +680,11 @@ body: |
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.3, implicit killed $scc
; CHECK-NEXT: {{ $}}
@@ -686,17 +694,17 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
@@ -704,10 +712,11 @@ body: |
; CHECK-NEXT: KILL [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5]]
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub0, [[DEF21]], implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]]
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF22]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -815,16 +824,17 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
@@ -833,10 +843,11 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_3:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_3:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub0, [[DEF21]], implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF22]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -942,13 +953,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -960,19 +973,20 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.6(0x80000000)
@@ -980,10 +994,11 @@ body: |
; CHECK-NEXT: KILL [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5]]
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF23]], implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]]
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF24]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -1099,13 +1114,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -1117,19 +1134,20 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.6(0x80000000)
@@ -1138,10 +1156,11 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_1]].sub0, implicit $exec
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_3:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_3:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF23]], implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF24]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -1248,18 +1267,19 @@ body: |
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.4(0x40000000), %bb.3(0x40000000)
@@ -1270,17 +1290,19 @@ body: |
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.5
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF19]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -1374,13 +1396,14 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.4(0x40000000), %bb.3(0x40000000)
@@ -1391,21 +1414,23 @@ body: |
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF18]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.5
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF19]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF19]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]]
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -1496,32 +1521,34 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub0, [[DEF13]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub1, [[DEF14]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_ADD_U32_e32_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.6(0x40000000), %bb.5(0x40000000)
@@ -1532,19 +1559,21 @@ body: |
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF19]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF19]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.7
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF20]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF20]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF21]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -1641,32 +1670,34 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub0, [[DEF13]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub1, [[DEF14]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_ADD_U32_e32_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.6(0x40000000), %bb.5(0x40000000)
@@ -1677,21 +1708,23 @@ body: |
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF19]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF19]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.7
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF20]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF20]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF12]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]]
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF21]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -1791,10 +1824,11 @@ body: |
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.4, implicit killed $scc
; CHECK-NEXT: {{ $}}
@@ -1804,23 +1838,23 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_6:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_7:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_8:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_9:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_10:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_11:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_6:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_7:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_8:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_9:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_10:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_11:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
@@ -1836,17 +1870,19 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: dead undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: dead undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub1, [[DEF21]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.5
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.6(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_3:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_3:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub0, [[DEF22]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_3]]
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF23]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -1971,16 +2007,17 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
@@ -1996,21 +2033,23 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.6(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub1, [[DEF21]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub0, [[DEF21]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.6
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.6(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub0, [[DEF22]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]].sub1, [[DEF22]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
; CHECK-NEXT: [[V_ADD_U32_e32_3:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF23]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -2127,13 +2166,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -2145,19 +2186,20 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.6(0x80000000)
@@ -2173,19 +2215,21 @@ body: |
; CHECK-NEXT: bb.7:
; CHECK-NEXT: successors: %bb.9(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF23]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF23]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.9
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.8:
; CHECK-NEXT: successors: %bb.9(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF24]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF24]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.9:
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]]
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF25]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -2312,13 +2356,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -2330,13 +2376,14 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x40000000), %bb.6(0x40000000)
@@ -2347,21 +2394,23 @@ body: |
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF23]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF23]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.8
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF24]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF24]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.8:
; CHECK-NEXT: [[V_ADD_U32_e32_3:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_1]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF25]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -2481,16 +2530,17 @@ body: |
; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
@@ -2498,10 +2548,11 @@ body: |
; CHECK-NEXT: KILL [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5]]
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF18]], [[DS_READ_B128_gfx9_]], 0, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DS_READ_B128_gfx9_]]
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF19]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DS_READ_B128_gfx9_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -2603,16 +2654,17 @@ body: |
; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
@@ -2622,8 +2674,8 @@ body: |
; CHECK-NEXT: bb.3:
; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[DS_READ_B128_gfx9_]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DS_READ_B128_gfx9_]], [[V_ADD_U32_e32_]]
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF18]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DS_READ_B128_gfx9_]], [[V_ADD_U32_e32_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -2713,37 +2765,39 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF13]], 0, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF14]], 128, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DS_READ_B128_gfx9_]]
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF19]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[DS_READ_B128_gfx9_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -2828,38 +2882,41 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF13]], 0, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF14]], 128, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF17]].sub0, [[DEF18]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF19]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 0, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -2960,16 +3017,18 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF21]], [[DS_READ_B128_gfx9_1]], 0, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_1]]
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF22]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -3069,17 +3128,19 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]], 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]], 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF21]], [[DS_READ_B128_gfx9_]], 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF21]], [[DS_READ_B128_gfx9_1]], 128, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_1]]
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF22]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -3176,13 +3237,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF11]], 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF17]], 256, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF11]], 512, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF18]], 512, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -3194,19 +3257,21 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_1]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DS_READ_B128_gfx9_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]], 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF23]], [[DS_READ_B128_gfx9_]], 128, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]]
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF24]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -3312,13 +3377,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF11]], 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF17]], 256, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF11]], 512, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF18]], 512, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -3330,20 +3397,22 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_1]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DS_READ_B128_gfx9_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]], 128, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]], 384, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF23]], [[DS_READ_B128_gfx9_1]], 128, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF23]], [[DS_READ_B128_gfx9_]], 384, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]]
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF24]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -3447,13 +3516,14 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.4(0x40000000), %bb.3(0x40000000)
@@ -3464,17 +3534,19 @@ body: |
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF18]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 0, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.5
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF19]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 128, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -3567,13 +3639,14 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF15]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF17]].sub0, [[DEF14]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.4(0x40000000), %bb.3(0x40000000)
@@ -3584,21 +3657,24 @@ body: |
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF18]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF18]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 256, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.5
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF19]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF19]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 256, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF20]], [[DS_READ_B128_gfx9_]], 0, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF21]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -3690,32 +3766,34 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF13]], 0, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF14]], 256, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.6(0x40000000), %bb.5(0x40000000)
@@ -3726,19 +3804,21 @@ body: |
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF19]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF19]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 128, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.7
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF20]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF20]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 128, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF21]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -3836,32 +3916,34 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF13]], 0, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF14]], 256, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.6(0x40000000), %bb.5(0x40000000)
@@ -3872,21 +3954,24 @@ body: |
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF19]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF19]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 128, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.7
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF20]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF20]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, 128, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF12]], [[DS_READ_B128_gfx9_]].sub0, 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF21]], [[DS_READ_B128_gfx9_]].sub0, 256, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF22]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -4000,10 +4085,11 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.4(0x40000000), %bb.3(0x40000000)
@@ -4014,17 +4100,19 @@ body: |
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF21]], [[DS_READ_B128_gfx9_1]], 0, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.5
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]], 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF22]], [[DS_READ_B128_gfx9_1]], 256, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_1]]
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF23]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -4134,10 +4222,11 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.4(0x40000000), %bb.3(0x40000000)
@@ -4148,21 +4237,24 @@ body: |
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]].sub0, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]].sub1, 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF21]], [[DS_READ_B128_gfx9_1]].sub0, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF21]], [[DS_READ_B128_gfx9_1]].sub1, 256, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.5
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]].sub1, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]].sub0, 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF22]], [[DS_READ_B128_gfx9_1]].sub1, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF22]], [[DS_READ_B128_gfx9_1]].sub0, 256, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF23]], [[DS_READ_B128_gfx9_]], 0, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_1]]
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF24]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -4272,13 +4364,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF11]], 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF17]], 256, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF11]], 512, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF18]], 512, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -4290,13 +4384,14 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_1]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DS_READ_B128_gfx9_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x40000000), %bb.6(0x40000000)
@@ -4307,19 +4402,21 @@ body: |
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]].sub0, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]].sub1, 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF23]], [[DS_READ_B128_gfx9_]].sub0, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF23]], [[DS_READ_B128_gfx9_]].sub1, 128, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.8
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]].sub1, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]].sub0, 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF24]], [[DS_READ_B128_gfx9_]].sub1, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF24]], [[DS_READ_B128_gfx9_]].sub0, 128, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.8:
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]]
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF25]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -4437,13 +4534,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF11]], 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF17]], 256, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF11]], 512, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF18]], 512, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -4455,13 +4554,14 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DS_READ_B128_gfx9_1]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF11]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DS_READ_B128_gfx9_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x40000000), %bb.6(0x40000000)
@@ -4472,21 +4572,24 @@ body: |
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]].sub0, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]].sub1, 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF23]], [[DS_READ_B128_gfx9_]].sub0, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF23]], [[DS_READ_B128_gfx9_]].sub1, 128, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.8
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]].sub1, 0, 0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_]].sub0, 128, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF24]], [[DS_READ_B128_gfx9_]].sub1, 0, 0, implicit $exec
+ ; CHECK-NEXT: DS_WRITE_B32_gfx9 [[DEF24]], [[DS_READ_B128_gfx9_]].sub0, 128, 0, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.8:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF11]], [[DS_READ_B128_gfx9_1]], 256, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF25]], [[DS_READ_B128_gfx9_1]], 256, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]]
+ ; CHECK-NEXT: [[DEF26:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF26]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF12]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[DS_READ_B128_gfx9_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -4596,6 +4699,7 @@ body: |
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
@@ -4603,17 +4707,18 @@ body: |
; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_ADD_U32_e32_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF17]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF18]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 0, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF19]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -4704,13 +4809,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF18]], 0, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -4722,13 +4829,14 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF14]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF15]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[DEF16]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x40000000), %bb.6(0x40000000)
@@ -4739,20 +4847,22 @@ body: |
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF12]], [[V_ADD_U32_e32_]], 0, 0, implicit $exec
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF23]], [[V_ADD_U32_e32_]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: S_BRANCH %bb.8
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[DEF20:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[DEF20:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[DEF24:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub0, [[DEF25]], implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_ADD_U32_e32_]].sub1, [[DEF25]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.8:
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF21]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[DEF20]]
+ ; CHECK-NEXT: [[DEF26:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF26]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_]], [[DEF24]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -4875,16 +4985,17 @@ body: |
; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF14]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF15]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_4:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_5:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[DEF16]], 4, 4, [[DEF19]].sub0, [[DEF20]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
@@ -4893,10 +5004,11 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_]].sub0, implicit $exec
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF12]], [[V_ADD_U32_e32_1]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF21]], [[V_ADD_U32_e32_1]], 0, 0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]]
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF22]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[DEF15]], [[DEF16]], [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -4990,32 +5102,34 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF12]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DS_READ_B128_gfx9_:%[0-9]+]]:vreg_128_align2 = DS_READ_B128_gfx9 [[DEF13]], 0, 0, implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[DS_READ_B128_gfx9_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[DS_READ_B128_gfx9_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF12]].sub1, [[DEF14]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[DS_READ_B128_gfx9_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF18]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.6(0x40000000), %bb.5(0x40000000)
@@ -5026,20 +5140,23 @@ body: |
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub1, [[DEF19]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF19]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.7
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.7(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 0, 0, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: DS_WRITE_B128_gfx9 [[DEF20]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], 0, 0, implicit $exec
; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
- ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DS_READ_B128_gfx9_]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_1]]
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DS_READ_B128_gfx9_]].sub0, [[DEF21]], implicit $exec
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF22]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -5135,49 +5252,55 @@ body: |
; CHECK-NEXT: S_NOP 0, implicit-def %13
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF11:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: dead [[DEF17:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: dead [[DEF14:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: $scc = IMPLICIT_DEF
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.2, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.4(0x40000000), %bb.3(0x40000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF16]].sub0, [[DEF15]], implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF11]], [[DEF12]], [[V_ADD_U32_e32_]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_ADD_U32_e32_]], 4, 4, [[DEF18]].sub0, [[DEF19]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF18]].sub0, [[DEF19]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF18]].sub0, [[DEF19]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: $scc = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF11]], [[DEF12]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF16]], [[DEF17]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF18]].sub0, [[DEF19]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.4, implicit killed $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.4(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF11]], [[DEF12]], [[DEF18]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF11]], [[DEF12]], [[DEF18]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF11]], [[DEF12]], [[DEF18]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF11]], [[DEF12]], [[DEF18]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF20]], [[DEF21]], [[DEF15]], 4, 4, [[DEF22]].sub0, [[DEF23]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF20]], [[DEF21]], [[DEF15]], 4, 4, [[DEF22]].sub0, [[DEF23]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF20]], [[DEF21]], [[DEF15]], 4, 4, [[DEF22]].sub0, [[DEF23]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF20]], [[DEF21]], [[DEF15]], 4, 4, [[DEF22]].sub0, [[DEF23]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.4:
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF18]].sub0, [[DEF15]], implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF15]].sub0, [[DEF24]], implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF19]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF13]], [[DEF16]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[DEF18]], [[V_ADD_U32_e32_1]]
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF25]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[DEF15]], [[V_ADD_U32_e32_1]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -5273,28 +5396,29 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF13:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
; CHECK-NEXT: undef [[AV_MOV_:%[0-9]+]].sub0:vreg_128_align2 = AV_MOV_B32_IMM_PSEUDO 0, implicit $exec
; CHECK-NEXT: [[AV_MOV_:%[0-9]+]].sub1:vreg_128_align2 = AV_MOV_B32_IMM_PSEUDO 0, implicit $exec
; CHECK-NEXT: [[AV_MOV_:%[0-9]+]].sub2:vreg_128_align2 = AV_MOV_B32_IMM_PSEUDO 0, implicit $exec
; CHECK-NEXT: [[AV_MOV_:%[0-9]+]].sub3:vreg_128_align2 = AV_MOV_B32_IMM_PSEUDO 0, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[AV_MOV_]], 4, 4, [[DEF15]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF15]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF15]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF15]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF12]], [[DEF13]], [[AV_MOV_]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF16]], implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF16]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_]]
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -5373,26 +5497,27 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF13:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
; CHECK-NEXT: undef [[AV_MOV_:%[0-9]+]].sub0_sub1:vreg_128_align2 = AV_MOV_B64_IMM_PSEUDO 0, implicit $exec
; CHECK-NEXT: [[AV_MOV_:%[0-9]+]].sub2_sub3:vreg_128_align2 = AV_MOV_B64_IMM_PSEUDO 0, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[AV_MOV_]], 4, 4, [[DEF15]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF15]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF15]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF15]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF12]], [[DEF13]], [[AV_MOV_]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF12]], [[DEF13]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], 4, 4, [[DEF14]].sub0, [[DEF15]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]].sub0, [[DEF16]], implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF16]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_]]
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_1]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_2]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_3]], [[V_ADD_U32_e32_]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -5469,28 +5594,29 @@ body: |
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
; CHECK-NEXT: SCHED_BARRIER 0
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF13:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF12:%[0-9]+]]:vreg_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[DEF13:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF14:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF15:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF14]], [[DEF15]], [[DEF13]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF15:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF16:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF13]], [[DEF14]], [[DEF12]], 4, 4, [[DEF15]].sub0, [[DEF16]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[COPY:%[0-9]+]]:areg_128_align2 = COPY [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]]
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_:%[0-9]+]]:areg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64 [[DEF14]], [[DEF15]], [[COPY]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_1:%[0-9]+]]:areg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
- ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_2:%[0-9]+]]:areg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64 [[DEF14]], [[DEF15]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_1]], 4, 4, [[DEF16]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_:%[0-9]+]]:areg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64 [[DEF13]], [[DEF14]], [[COPY]], 4, 4, [[DEF15]].sub0, [[DEF16]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_1:%[0-9]+]]:areg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64 [[DEF13]], [[DEF14]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_]], 4, 4, [[DEF15]].sub0, [[DEF16]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_2:%[0-9]+]]:areg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64 [[DEF13]], [[DEF14]], [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_1]], 4, 4, [[DEF15]].sub0, [[DEF16]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64_]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_128_align2 = COPY [[V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_e64_2]]
; CHECK-NEXT: [[COPY2:%[0-9]+]]:vreg_128_align2 = COPY [[COPY1]]
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF17]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[V_ADD_U32_e32_]], [[COPY2]]
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF18]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF12]], [[V_ADD_U32_e32_]], [[COPY2]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -5578,7 +5704,8 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_64_align2 = V_ADD_U32_e32 [[AV_MOV_]].sub0, [[DEF10]], implicit $exec
+ ; CHECK-NEXT: [[AV_MOV_1:%[0-9]+]]:vreg_64_align2 = AV_MOV_B64_IMM_PSEUDO 0, implicit $exec
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_:%[0-9]+]].sub0:vreg_64_align2 = V_ADD_U32_e32 [[AV_MOV_1]].sub0, [[DEF10]], implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_64_align2 = COPY [[V_MFMA_F64_4X4X4F64_e64_3]]
; CHECK-NEXT: [[COPY2:%[0-9]+]]:vreg_64_align2 = COPY [[COPY1]]
; CHECK-NEXT: SCHED_BARRIER 0
diff --git a/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_cost.mir b/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_cost.mir
index 32375684cbf99..ae333b8f03e47 100644
--- a/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_cost.mir
+++ b/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_cost.mir
@@ -68,13 +68,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -86,10 +88,11 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x40000000), %bb.6(0x40000000)
@@ -100,8 +103,9 @@ body: |
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: dead undef [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF23]], implicit $exec
+ ; CHECK-NEXT: dead undef [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF23]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.8
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
@@ -113,20 +117,22 @@ body: |
; CHECK-NEXT: bb.8:
; CHECK-NEXT: successors: %bb.9(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_3:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: dead undef [[V_ADD_U32_e32_3:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_3:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF24]], implicit $exec
+ ; CHECK-NEXT: dead undef [[V_ADD_U32_e32_3:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF24]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.9:
; CHECK-NEXT: successors: %bb.10(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: undef [[V_ADD_U32_e32_4:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF14]].sub0, implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_4:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub2, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_ADD_U32_e32_4:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub2, [[DEF25]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.10:
; CHECK-NEXT: [[V_ADD_U32_e32_5:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF14]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[V_ADD_U32_e32_4]], [[V_ADD_U32_e32_5]]
+ ; CHECK-NEXT: [[DEF26:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF26]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[V_ADD_U32_e32_4]], [[V_ADD_U32_e32_5]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
@@ -252,13 +258,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -270,10 +278,11 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF14:%[0-9]+]]:vreg_128_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x40000000), %bb.6(0x40000000)
@@ -284,21 +293,23 @@ body: |
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF23]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF23]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.8
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF24]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF24]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.8:
; CHECK-NEXT: [[V_ADD_U32_e32_3:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_1]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF25]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
diff --git a/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_diff_types.mir b/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_diff_types.mir
index 7528b855e23e3..fd5697df63bcf 100644
--- a/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_diff_types.mir
+++ b/llvm/test/CodeGen/AMDGPU/sched_mfma_rewrite_diff_types.mir
@@ -57,13 +57,15 @@ body: |
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub0, [[DEF17]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_1:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF13]].sub1, [[DEF18]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
@@ -75,10 +77,11 @@ body: |
; CHECK-NEXT: bb.4:
; CHECK-NEXT: successors: %bb.5(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: [[DEF17:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF18:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF19:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
- ; CHECK-NEXT: [[DEF10:%[0-9]+]].sub0_sub1_sub2_sub3:vreg_1024_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF17]], [[DEF18]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF19]].sub0, [[DEF12]], 0, 0, implicit $mode, implicit $exec
+ ; CHECK-NEXT: [[DEF19:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF20:%[0-9]+]]:av_128_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF21:%[0-9]+]]:vreg_64_align2 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF10:%[0-9]+]].sub0_sub1_sub2_sub3:vreg_1024_align2 = contract nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 [[DEF19]], [[DEF20]], [[V_ADD_U32_e32_1]], 4, 4, [[DEF21]].sub0, [[DEF22]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.5:
; CHECK-NEXT: successors: %bb.7(0x40000000), %bb.6(0x40000000)
@@ -89,21 +92,23 @@ body: |
; CHECK-NEXT: bb.6:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF23]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF23]], implicit $exec
; CHECK-NEXT: S_BRANCH %bb.8
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.7:
; CHECK-NEXT: successors: %bb.8(0x80000000)
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF12]], implicit $exec
- ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF12]], implicit $exec
+ ; CHECK-NEXT: [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: undef [[V_ADD_U32_e32_2:%[0-9]+]].sub0:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub0, [[DEF24]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]].sub1:vreg_128_align2 = V_ADD_U32_e32 [[DEF14]].sub1, [[DEF24]], implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.8:
; CHECK-NEXT: [[V_ADD_U32_e32_3:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF13]].sub1, [[V_ADD_U32_e32_1]].sub0, implicit $exec
; CHECK-NEXT: SCHED_BARRIER 0
- ; CHECK-NEXT: [[DEF20:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
- ; CHECK-NEXT: KILL [[DEF20]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
+ ; CHECK-NEXT: [[DEF25:%[0-9]+]]:vreg_1024 = IMPLICIT_DEF
+ ; CHECK-NEXT: KILL [[DEF25]], [[DEF]], [[DEF1]], [[DEF2]], [[DEF3]], [[DEF4]], [[DEF5]], [[DEF6]], [[DEF7]], [[DEF8]], [[DEF9]], [[DEF10]], [[DEF11]], [[DEF13]], [[DEF14]], [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]]
; CHECK-NEXT: S_NOP 0, implicit %12, implicit %13
; CHECK-NEXT: S_ENDPGM 0
bb.0:
More information about the llvm-branch-commits
mailing list