[llvm-branch-commits] [llvm] [AMDGPU] Replace getRepeatRate with getBlockingCycles (PR #239855)

Jeffrey Byrnes via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Fri Nov 6 16:55:08 PST 2026


https://github.com/jrbyrnes updated https://github.com/llvm/llvm-project/pull/239855

>From e5b40e7ea6bae38bc075f60078403ec750c5330e Mon Sep 17 00:00:00 2001
From: Jeffrey Byrnes <Jeffrey.Byrnes at amd.com>
Date: Fri, 6 Nov 2026 15:42:55 -0800
Subject: [PATCH 1/2] [AMDGPU] Replace getRepeatRate with getBlockingCycles

Change-Id: I5598664f64fc010bc1b8fed0c3a1b53477ef208b
---
 .../lib/Target/AMDGPU/GCNHazardRecognizer.cpp |  16 +--
 llvm/lib/Target/AMDGPU/SIInstrInfo.cpp        |  18 ---
 .../CodeGen/AMDGPU/GlobalISel/fptrunc.bf16.ll |  10 +-
 llvm/test/CodeGen/AMDGPU/GlobalISel/mul.ll    |  53 ++++----
 llvm/test/CodeGen/AMDGPU/addrspacecast-gas.ll |   4 +-
 llvm/test/CodeGen/AMDGPU/bf16-conversions.ll  |   8 +-
 llvm/test/CodeGen/AMDGPU/bf16.ll              |  89 +++++++-------
 .../CodeGen/AMDGPU/calling-conventions.ll     |   5 +-
 llvm/test/CodeGen/AMDGPU/clmul.ll             |   3 +-
 llvm/test/CodeGen/AMDGPU/coexec-scheduler.ll  |  26 ++--
 .../test/CodeGen/AMDGPU/flat-saddr-atomics.ll |  12 +-
 .../AMDGPU/float-to-arbitrary-fp-fp8-hw.ll    |  29 +++--
 .../AMDGPU/float-to-arbitrary-fp-widen.ll     |   2 +-
 .../CodeGen/AMDGPU/global-atomicrmw-fadd.ll   |   4 +-
 llvm/test/CodeGen/AMDGPU/global-load-xcnt.ll  |  46 +++----
 .../CodeGen/AMDGPU/integer-mad-patterns.ll    |  47 ++++----
 .../AMDGPU/llvm.amdgcn.cvt.scale.pk.ll        |  36 +++---
 .../CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll  |   2 +-
 llvm/test/CodeGen/AMDGPU/lshl-add-u64.ll      |  10 +-
 llvm/test/CodeGen/AMDGPU/mad_64_32.ll         |  12 +-
 llvm/test/CodeGen/AMDGPU/mul.ll               |   6 +-
 llvm/test/CodeGen/AMDGPU/packed-fp32.ll       |   6 +-
 .../AMDGPU/reassoc-mul-add-1-to-mad.ll        |  27 +++--
 llvm/test/CodeGen/AMDGPU/shl.v2i64.ll         | 114 ++++++++++++------
 llvm/test/CodeGen/AMDGPU/slp-int-to-fp.ll     |  45 ++++---
 .../CodeGen/AMDGPU/strict_fptrunc_bf16.ll     |  18 +--
 26 files changed, 326 insertions(+), 322 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
index b91ef681cf6722..ccc34018fdc52b 100644
--- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
@@ -279,7 +279,7 @@ void GCNHazardRecognizer::updateTRANSState(const MachineInstr &MI) {
 void GCNHazardRecognizer::updateMultiCycleVALUState(const MachineInstr &MI) {
   if (!hasCoExecWindowModel())
     return;
-  // Multi-cycle VALU (CVT, etc.) blocks subsequent VALU for repeat rate cycles.
+  // Multi-cycle VALU (CVT, etc.) blocks subsequent VALU for BlockingCycles cycles.
   if (!SIInstrInfo::isVALU(MI, /*AllowLDSDMA=*/true))
     return;
 
@@ -288,13 +288,13 @@ void GCNHazardRecognizer::updateMultiCycleVALUState(const MachineInstr &MI) {
       SIInstrInfo::isMFMA(MI) || SIInstrInfo::isTRANS(MI))
     return;
 
-  unsigned RepeatRate = TII.getRepeatRate(MI);
-  if (RepeatRate > 1) {
+  unsigned BlockingCycles = TII.getBlockingCycles(MI);
+  if (BlockingCycles > 1) {
     // bumpCycle's AdvanceCycle decrements once before the next pick's
     // hazard check (same convention as CyclesUntilTRANS), so to expose
-    // RepeatRate-1 cycles of shadow we must seed with RepeatRate.
-    CyclesUntilVALU = RepeatRate;
-    LLVM_DEBUG(dbgs() << "    Multi-cycle VALU: repeat=" << RepeatRate
+    // BlockingCycles-1 cycles of shadow we must seed with BlockingCycles.
+    CyclesUntilVALU = BlockingCycles;
+    LLVM_DEBUG(dbgs() << "    Multi-cycle VALU: BlockingCycles=" << BlockingCycles
                       << ", CyclesUntilVALU=" << CyclesUntilVALU << "\n");
   }
 }
@@ -315,7 +315,7 @@ unsigned GCNHazardRecognizer::checkTRANSHazard(const MachineInstr &MI) const {
 
   if (SIInstrInfo::isVALU(MI, /*AllowLDSDMA=*/true) &&
       !SIInstrInfo::isWMMA(MI) && !SIInstrInfo::isSWMMAC(MI) &&
-      TII.getRepeatRate(MI) > 1)
+      TII.getBlockingCycles(MI) > 1)
     return CyclesUntilTRANS;
 
   return 0;
@@ -327,7 +327,7 @@ GCNHazardRecognizer::checkMultiCycleVALUHazard(const MachineInstr &MI) const {
     return 0;
 
   // Multi-cycle VALU blocks anything on the VALU pipe - VALU, WMMA, SWMMAC,
-  // and TRANS - for RepeatRate-1 cycles. Only off-pipe instructions (MEM,
+  // and TRANS - for BlockingCycles-1 cycles. Only off-pipe instructions (MEM,
   // SALU, control) can fill the shadow.
   if (!SIInstrInfo::isVALU(MI, /*AllowLDSDMA=*/true) &&
       !SIInstrInfo::isWMMA(MI) && !SIInstrInfo::isSWMMAC(MI) &&
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index 7895c1c8a6a8fe..fe4957a9a5218f 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -11974,24 +11974,6 @@ void SIInstrInfo::enforceOperandRCAlignment(MachineInstr &MI,
   MI.addOperand(MachineOperand::CreateReg(NewVR, false, true));
 }
 
-unsigned SIInstrInfo::getRepeatRate(const MachineInstr &MI) const {
-  if (!SchedModel.hasInstrSchedModel())
-    return 0;
-
-  // The repeat rate is the throughput-limiting resource occupancy: the largest
-  // number of cycles any written processor resource is held.
-  const MCSchedClassDesc *SCDesc = SchedModel.resolveSchedClass(&MI);
-  unsigned RepeatRate = 0;
-  for (TargetSchedModel::ProcResIter
-           PI = SchedModel.getWriteProcResBegin(SCDesc),
-           PE = SchedModel.getWriteProcResEnd(SCDesc);
-       PI != PE; ++PI) {
-    RepeatRate = std::max(RepeatRate, (unsigned)PI->ReleaseAtCycle);
-  }
-
-  return RepeatRate;
-}
-
 bool SIInstrInfo::isGlobalMemoryObject(const MachineInstr *MI) const {
   if (isIGLP(*MI))
     return false;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/fptrunc.bf16.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/fptrunc.bf16.ll
index 1c4a9b928f75bb..64888d85c6d775 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/fptrunc.bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/fptrunc.bf16.ll
@@ -221,28 +221,28 @@ define amdgpu_ps bfloat @fptrunc_f64_to_bf16_s(double inreg %a) {
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[0:1], v2
 ; GFX1250-NEXT:    v_cmp_nlg_f64_e32 vcc_lo, s[0:1], v[0:1]
-; GFX1250-NEXT:    v_cmp_ngt_f64_e64 s0, |s[0:1]|, |v[0:1]|
-; GFX1250-NEXT:    v_readfirstlane_b32 s1, v2
 ; GFX1250-NEXT:    s_cmp_lg_u32 vcc_lo, 0
+; GFX1250-NEXT:    v_cmp_ngt_f64_e64 s0, |s[0:1]|, |v[0:1]|
 ; GFX1250-NEXT:    s_cselect_b32 s2, 1, 0
-; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GFX1250-NEXT:    v_readfirstlane_b32 s1, v2
 ; GFX1250-NEXT:    s_or_b32 s2, s2, s1
 ; GFX1250-NEXT:    s_cmp_lg_u32 s0, 0
 ; GFX1250-NEXT:    s_cselect_b32 s0, -1, 0
 ; GFX1250-NEXT:    s_and_b32 s2, s2, 1
 ; GFX1250-NEXT:    s_or_b32 s0, s0, 1
+; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
 ; GFX1250-NEXT:    s_add_co_i32 s0, s1, s0
 ; GFX1250-NEXT:    s_cmp_lg_u32 s2, 0
 ; GFX1250-NEXT:    s_cselect_b32 s0, s1, s0
-; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
 ; GFX1250-NEXT:    s_bfe_u32 s1, s0, 0x10010
 ; GFX1250-NEXT:    s_or_b32 s2, s0, 0x400000
 ; GFX1250-NEXT:    s_add_co_i32 s1, s1, s0
+; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
 ; GFX1250-NEXT:    s_addk_co_i32 s1, 0x7fff
 ; GFX1250-NEXT:    s_cmp_u_f32 s0, 0
 ; GFX1250-NEXT:    s_cselect_b32 s0, s2, s1
-; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX1250-NEXT:    s_lshr_b32 s0, s0, 16
+; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1250-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX1250-NEXT:    ; return to shader part epilog
   %result = fptrunc double %a to bfloat
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/mul.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/mul.ll
index e0bcc5da36b58d..ee4af399d247a8 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/mul.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/mul.ll
@@ -3602,70 +3602,65 @@ define i256 @v_mul_i256(i256 %num, i256 %den) {
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    v_dual_mov_b32 v16, v0 :: v_dual_mov_b32 v17, v1
 ; GFX1250-NEXT:    v_mul_lo_u32 v29, v5, v10
-; GFX1250-NEXT:    v_mul_lo_u32 v30, v4, v11
-; GFX1250-NEXT:    v_mul_lo_u32 v31, v3, v12
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[0:1], v16, v14, 0
+; GFX1250-NEXT:    v_mul_lo_u32 v30, v4, v11
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[18:19], v17, v13, v[0:1]
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[0:1], v16, v12, 0
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[20:21], v2, v12, v[18:19]
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[18:19], s0, v17, v11, v[0:1]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1250-NEXT:    v_cndmask_b32_e64 v22, 0, 1, s0
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[0:1], v3, v11, v[20:21]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[20:21], vcc_lo, v2, v10, v[18:19]
-; GFX1250-NEXT:    v_add_co_ci_u32_e64 v22, null, 0, v22, vcc_lo
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[18:19], v4, v10, v[0:1]
+; GFX1250-NEXT:    v_cndmask_b32_e64 v22, 0, 1, s0
+; GFX1250-NEXT:    v_mul_lo_u32 v31, v3, v12
+; GFX1250-NEXT:    v_add_co_ci_u32_e64 v22, null, 0, v22, vcc_lo
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[0:1], vcc_lo, v3, v9, v[20:21]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1250-NEXT:    v_add_co_ci_u32_e64 v26, null, 0, v22, vcc_lo
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[20:21], v5, v9, v[18:19]
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[18:19], v16, v10, 0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_add_co_ci_u32_e64 v26, null, 0, v22, vcc_lo
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[22:23], vcc_lo, v4, v8, v[0:1]
-; GFX1250-NEXT:    v_add_co_ci_u32_e64 v28, null, 0, v26, vcc_lo
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[24:25], v6, v8, v[20:21]
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[0:1], s0, v17, v9, v[18:19]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_dual_mov_b32 v18, v23 :: v_dual_mov_b32 v19, v24
+; GFX1250-NEXT:    v_add_co_ci_u32_e64 v28, null, 0, v26, vcc_lo
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[20:21], vcc_lo, v2, v8, v[0:1]
 ; GFX1250-NEXT:    v_mul_lo_u32 v24, v6, v9
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v6, 0, 1, s0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[0:1], s0, v16, v13, v[18:19]
-; GFX1250-NEXT:    v_mov_b32_e32 v19, v22
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_dual_mov_b32 v18, v21 :: v_dual_mov_b32 v19, v22
 ; GFX1250-NEXT:    v_add_co_ci_u32_e64 v6, null, 0, v6, vcc_lo
-; GFX1250-NEXT:    v_mov_b32_e32 v18, v21
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[26:27], vcc_lo, v17, v12, v[0:1]
-; GFX1250-NEXT:    v_mad_nc_u64_u32 v[0:1], v16, v8, 0
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[22:23], s2, v16, v11, v[18:19]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1250-NEXT:    v_cndmask_b32_e64 v21, 0, 1, s2
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[18:19], s1, v2, v11, v[26:27]
-; GFX1250-NEXT:    v_mov_b32_e32 v11, v20
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_mad_nc_u64_u32 v[0:1], v16, v8, 0
+; GFX1250-NEXT:    v_cndmask_b32_e64 v21, 0, 1, s2
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[26:27], s2, v17, v10, v[22:23]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[22:23], s3, v3, v10, v[18:19]
-; GFX1250-NEXT:    v_mov_b32_e32 v10, v1
+; GFX1250-NEXT:    v_dual_mov_b32 v10, v1 :: v_dual_mov_b32 v11, v20
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_add_co_ci_u32_e64 v1, null, 0, v21, s2
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[18:19], s2, v2, v9, v[26:27]
 ; GFX1250-NEXT:    v_mul_lo_u32 v2, v2, v13
+; GFX1250-NEXT:    v_mad_co_u64_u32 v[12:13], s4, v4, v9, v[22:23]
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[20:21], s5, v16, v9, v[10:11]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX1250-NEXT:    v_add_co_ci_u32_e64 v1, null, 0, v1, s2
-; GFX1250-NEXT:    v_mad_co_u64_u32 v[12:13], s4, v4, v9, v[22:23]
-; GFX1250-NEXT:    v_mul_lo_u32 v22, v16, v15
-; GFX1250-NEXT:    v_mul_lo_u32 v9, v17, v14
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[10:11], s2, v3, v8, v[18:19]
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v3, 0, 1, s5
-; GFX1250-NEXT:    v_add_co_ci_u32_e64 v1, null, 0, v1, s2
+; GFX1250-NEXT:    v_mul_lo_u32 v22, v16, v15
+; GFX1250-NEXT:    v_mul_lo_u32 v9, v17, v14
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[14:15], s5, v5, v8, v[12:13]
+; GFX1250-NEXT:    v_add_co_ci_u32_e64 v1, null, 0, v1, s2
 ; GFX1250-NEXT:    v_mad_co_u64_u32 v[12:13], s2, v17, v8, v[20:21]
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    v_add_co_ci_u32_e64 v3, s2, v3, v10, s2
diff --git a/llvm/test/CodeGen/AMDGPU/addrspacecast-gas.ll b/llvm/test/CodeGen/AMDGPU/addrspacecast-gas.ll
index 208c8530781d5f..e478fa00bcf2f1 100644
--- a/llvm/test/CodeGen/AMDGPU/addrspacecast-gas.ll
+++ b/llvm/test/CodeGen/AMDGPU/addrspacecast-gas.ll
@@ -34,12 +34,12 @@ define amdgpu_kernel void @use_private_to_flat_addrspacecast(ptr addrspace(5) %p
 ; GFX1250-GISEL-NEXT:    global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
 ; GFX1250-GISEL-NEXT:    s_load_b32 s0, s[4:5], 0x24 nv
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[0:1], src_flat_scratch_base_lo
-; GFX1250-GISEL-NEXT:    v_mbcnt_lo_u32_b32 v2, -1, 0
 ; GFX1250-GISEL-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-GISEL-NEXT:    s_cmp_lg_u32 s0, -1
+; GFX1250-GISEL-NEXT:    v_mbcnt_lo_u32_b32 v2, -1, 0
 ; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, s0, v0
 ; GFX1250-GISEL-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
-; GFX1250-GISEL-NEXT:    s_cmp_lg_u32 s0, -1
 ; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
 ; GFX1250-GISEL-NEXT:    v_add_co_ci_u32_e64 v1, null, v2, v1, vcc_lo
 ; GFX1250-GISEL-NEXT:    s_cselect_b32 vcc_lo, exec_lo, 0
diff --git a/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll b/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
index 0450b6cf7401ad..bd829a29457295 100644
--- a/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
+++ b/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
@@ -235,15 +235,15 @@ define amdgpu_ps float @v_test_cvt_v2f64_v2bf16_v(<2 x double> %src) {
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v8
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[6:7], v9
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[2:3]|, |v[4:5]|
 ; GFX1250-NEXT:    v_cmp_nlg_f64_e32 vcc_lo, v[2:3], v[4:5]
-; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, v[0:1], v[6:7]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v2, -1, 1, s1
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[0:1]|, |v[6:7]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, v[0:1], v[6:7]
 ; GFX1250-NEXT:    v_dual_add_nc_u32 v1, v8, v2 :: v_dual_bitop2_b32 v10, 1, v8 bitop3:0x40
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s1
 ; GFX1250-NEXT:    v_and_b32_e32 v11, 1, v9
 ; GFX1250-NEXT:    v_cmp_ne_u32_e64 s1, 0, v10
diff --git a/llvm/test/CodeGen/AMDGPU/bf16.ll b/llvm/test/CodeGen/AMDGPU/bf16.ll
index 5da163b24d948d..45e6859c61cba2 100644
--- a/llvm/test/CodeGen/AMDGPU/bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/bf16.ll
@@ -9105,54 +9105,45 @@ define <32 x double> @global_extload_v32bf16_to_v32f64(ptr addrspace(1) %ptr) #0
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[2:3], v3
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[8:9], v7
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[6:7], v11
-; GFX1250-NEXT:    s_wait_loadcnt 0x11
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v69, 16, v17 :: v_dual_lshlrev_b32 v37, 16, v16
-; GFX1250-NEXT:    s_wait_loadcnt 0xe
+; GFX1250-NEXT:    s_clause 0x1
+; GFX1250-NEXT:    scratch_store_b128 v0, v[2:5], off offset:240
+; GFX1250-NEXT:    scratch_store_b128 v0, v[6:9], off offset:224
+; GFX1250-NEXT:    s_wait_loadcnt 0x0
+; GFX1250-NEXT:    v_dual_lshlrev_b32 v34, 16, v34 :: v_dual_lshlrev_b32 v66, 16, v32
+; GFX1250-NEXT:    v_dual_lshlrev_b32 v49, 16, v29 :: v_dual_lshlrev_b32 v52, 16, v30
+; GFX1250-NEXT:    v_dual_lshlrev_b32 v64, 16, v25 :: v_dual_lshlrev_b32 v53, 16, v24
+; GFX1250-NEXT:    v_dual_lshlrev_b32 v54, 16, v31 :: v_dual_lshlrev_b32 v65, 16, v33
+; GFX1250-NEXT:    v_dual_lshlrev_b32 v23, 16, v23 :: v_dual_lshlrev_b32 v48, 16, v28
+; GFX1250-NEXT:    v_dual_lshlrev_b32 v27, 16, v27 :: v_dual_lshlrev_b32 v26, 16, v26
 ; GFX1250-NEXT:    v_dual_lshlrev_b32 v36, 16, v20 :: v_dual_lshlrev_b32 v82, 16, v18
-; GFX1250-NEXT:    s_wait_loadcnt 0xc
+; GFX1250-NEXT:    v_dual_lshlrev_b32 v69, 16, v17 :: v_dual_lshlrev_b32 v37, 16, v16
+; GFX1250-NEXT:    s_wait_xcnt 0x0
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[8:9], v34
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[6:7], v66
 ; GFX1250-NEXT:    v_dual_lshlrev_b32 v68, 16, v21 :: v_dual_lshlrev_b32 v22, 16, v22
-; GFX1250-NEXT:    s_wait_loadcnt 0x9
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v53, 16, v24 :: v_dual_lshlrev_b32 v64, 16, v25
-; GFX1250-NEXT:    v_lshlrev_b32_e32 v23, 16, v23
-; GFX1250-NEXT:    s_wait_loadcnt 0x7
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v27, 16, v27 :: v_dual_lshlrev_b32 v26, 16, v26
-; GFX1250-NEXT:    s_wait_loadcnt 0x4
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v49, 16, v29 :: v_dual_lshlrev_b32 v52, 16, v30
-; GFX1250-NEXT:    s_wait_loadcnt 0x3
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v54, 16, v31 :: v_dual_lshlrev_b32 v48, 16, v28
-; GFX1250-NEXT:    s_wait_loadcnt 0x0
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v66, 16, v32 :: v_dual_lshlrev_b32 v34, 16, v34
-; GFX1250-NEXT:    v_lshlrev_b32_e32 v65, 16, v33
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v52
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[2:3], v53
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[54:55], v54
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[52:53], v65
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[66:67], v49
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[64:65], v64
 ; GFX1250-NEXT:    v_dual_lshlrev_b32 v11, 16, v14 :: v_dual_lshlrev_b32 v14, 16, v15
 ; GFX1250-NEXT:    v_lshlrev_b32_e32 v15, 16, v19
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[50:51], v48
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[48:49], v26
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[16:17], v35
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[28:29], v36
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[30:31], v37
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[36:37], v23
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[34:35], v27
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[32:33], v22
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[70:71], v68
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[68:69], v69
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[26:27], v82
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[24:25], v14
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[22:23], v15
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[20:21], v11
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[18:19], v10
-; GFX1250-NEXT:    scratch_store_b128 v0, v[2:5], off offset:240
-; GFX1250-NEXT:    s_wait_xcnt 0x0
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v52
-; GFX1250-NEXT:    scratch_store_b128 v0, v[6:9], off offset:224
-; GFX1250-NEXT:    s_wait_xcnt 0x0
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[8:9], v34
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[6:7], v66
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[2:3], v53
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[52:53], v65
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[66:67], v49
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[64:65], v64
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[48:49], v26
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[34:35], v27
-; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[26:27], v82
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[82:83], v80
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[80:81], v81
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[14:15], v39
@@ -37372,22 +37363,23 @@ define <4 x bfloat> @v_sitofp_v4i64_to_v4bf16(<4 x i64> %x) #0 {
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_min_u32_e32 v9, v12, v9
 ; GFX1250-NEXT:    v_min_u32_e32 v11, v13, v14
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[4:5], v10, v[4:5]
-; GFX1250-NEXT:    v_lshlrev_b64_e32 v[2:3], v9, v[2:3]
 ; GFX1250-NEXT:    v_min_u32_e32 v6, 1, v6
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[2:3], v9, v[2:3]
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[0:1], v11, v[0:1]
-; GFX1250-NEXT:    v_min_u32_e32 v4, 1, v4
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1250-NEXT:    v_min_u32_e32 v2, 1, v2
+; GFX1250-NEXT:    v_min_u32_e32 v4, 1, v4
 ; GFX1250-NEXT:    v_or_b32_e32 v6, v7, v6
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_3) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_min_u32_e32 v2, 1, v2
+; GFX1250-NEXT:    v_sub_nc_u32_e32 v7, 32, v10
 ; GFX1250-NEXT:    v_min_u32_e32 v0, 1, v0
 ; GFX1250-NEXT:    v_dual_sub_nc_u32 v5, 32, v9 :: v_dual_bitop2_b32 v4, v5, v4 bitop3:0x54
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_or_b32_e32 v2, v3, v2
-; GFX1250-NEXT:    v_dual_sub_nc_u32 v7, 32, v10 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX1250-NEXT:    v_or_b32_e32 v0, v1, v0
 ; GFX1250-NEXT:    v_sub_nc_u32_e32 v1, 32, v8
 ; GFX1250-NEXT:    v_cvt_f32_i32_e32 v3, v6
 ; GFX1250-NEXT:    v_cvt_f32_i32_e32 v4, v4
@@ -38886,8 +38878,8 @@ define <2 x bfloat> @v_uitofp_v2i64_to_v2bf16(<2 x i64> %x) #0 {
 ; GFX1250-NEXT:    v_min_u32_e32 v2, 1, v2
 ; GFX1250-NEXT:    v_min_u32_e32 v0, 1, v0
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1250-NEXT:    v_or_b32_e32 v2, v3, v2
-; GFX1250-NEXT:    v_dual_sub_nc_u32 v3, 32, v5 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54
+; GFX1250-NEXT:    v_dual_sub_nc_u32 v3, 32, v5 :: v_dual_bitop2_b32 v2, v3, v2 bitop3:0x54
+; GFX1250-NEXT:    v_or_b32_e32 v0, v1, v0
 ; GFX1250-NEXT:    v_sub_nc_u32_e32 v1, 32, v4
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_cvt_f32_u32_e32 v2, v2
@@ -39231,12 +39223,13 @@ define <3 x bfloat> @v_uitofp_v3i64_to_v3bf16(<3 x i64> %x) #0 {
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[0:1], v7, v[0:1]
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[4:5], v8, v[4:5]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_min_u32_e32 v2, 1, v2
+; GFX1250-NEXT:    v_sub_nc_u32_e32 v8, 32, v8
 ; GFX1250-NEXT:    v_min_u32_e32 v0, 1, v0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_min_u32_e32 v4, 1, v4
-; GFX1250-NEXT:    v_dual_sub_nc_u32 v8, 32, v8 :: v_dual_bitop2_b32 v2, v3, v2 bitop3:0x54
+; GFX1250-NEXT:    v_or_b32_e32 v2, v3, v2
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54
 ; GFX1250-NEXT:    v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54
@@ -39670,17 +39663,17 @@ define <4 x bfloat> @v_uitofp_v4i64_to_v4bf16(<4 x i64> %x) #0 {
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_min_u32_e32 v6, 1, v6
 ; GFX1250-NEXT:    v_min_u32_e32 v2, 1, v2
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX1250-NEXT:    v_min_u32_e32 v0, 1, v0
+; GFX1250-NEXT:    v_sub_nc_u32_e32 v8, 32, v8
 ; GFX1250-NEXT:    v_min_u32_e32 v4, 1, v4
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1250-NEXT:    v_dual_sub_nc_u32 v8, 32, v8 :: v_dual_bitop2_b32 v6, v7, v6 bitop3:0x54
+; GFX1250-NEXT:    v_or_b32_e32 v6, v7, v6
 ; GFX1250-NEXT:    v_dual_sub_nc_u32 v3, 32, v11 :: v_dual_bitop2_b32 v2, v3, v2 bitop3:0x54
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_or_b32_e32 v0, v1, v0
-; GFX1250-NEXT:    v_or_b32_e32 v1, v5, v4
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_or_b32_e32 v1, v5, v4
 ; GFX1250-NEXT:    v_cvt_f32_u32_e32 v4, v6
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX1250-NEXT:    v_cvt_f32_u32_e32 v2, v2
 ; GFX1250-NEXT:    v_sub_nc_u32_e32 v5, 32, v9
 ; GFX1250-NEXT:    v_cvt_f32_u32_e32 v0, v0
diff --git a/llvm/test/CodeGen/AMDGPU/calling-conventions.ll b/llvm/test/CodeGen/AMDGPU/calling-conventions.ll
index 17aa94a597c566..4c7d91053e3bfd 100644
--- a/llvm/test/CodeGen/AMDGPU/calling-conventions.ll
+++ b/llvm/test/CodeGen/AMDGPU/calling-conventions.ll
@@ -1802,16 +1802,15 @@ define amdgpu_kernel void @amd_kernel_v3i8(<3 x i8> %arg0) {
 ; GFX1250-NEXT:    global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
 ; GFX1250-NEXT:    s_load_b32 s0, s[4:5], 0x24 nv
 ; GFX1250-NEXT:    v_mov_b64_e32 v[0:1], 2
-; GFX1250-NEXT:    v_mov_b64_e32 v[2:3], 0
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    s_lshr_b32 s1, s0, 16
 ; GFX1250-NEXT:    s_bfe_u32 s2, s0, 0x80008
 ; GFX1250-NEXT:    s_add_co_i32 s1, s1, s1
-; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_dual_mov_b32 v4, 0x6050400 :: v_dual_mov_b32 v5, s1
 ; GFX1250-NEXT:    s_add_co_i32 s0, s0, s0
 ; GFX1250-NEXT:    s_add_co_i32 s2, s2, s2
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX1250-NEXT:    v_mov_b64_e32 v[2:3], 0
 ; GFX1250-NEXT:    v_perm_b32 v4, s2, s0, v4
 ; GFX1250-NEXT:    s_clause 0x1
 ; GFX1250-NEXT:    global_store_b8 v[0:1], v5, off
diff --git a/llvm/test/CodeGen/AMDGPU/clmul.ll b/llvm/test/CodeGen/AMDGPU/clmul.ll
index 08bd7ccda07745..0e0ced47809309 100644
--- a/llvm/test/CodeGen/AMDGPU/clmul.ll
+++ b/llvm/test/CodeGen/AMDGPU/clmul.ll
@@ -368,8 +368,9 @@ define amdgpu_kernel void @test_clmul_i32(ptr addrspace(1) %out, ptr addrspace(1
 ; GFX1250-NEXT:    v_and_b32_e32 v8, 0x88888888, v0
 ; GFX1250-NEXT:    v_mul_lo_u32 v9, v3, v2
 ; GFX1250-NEXT:    v_mul_lo_u32 v13, v3, v6
-; GFX1250-NEXT:    v_mul_lo_u32 v15, v7, v2
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX1250-NEXT:    v_mul_lo_u32 v14, v5, v1
+; GFX1250-NEXT:    v_mul_lo_u32 v15, v7, v2
 ; GFX1250-NEXT:    v_mul_lo_u32 v16, v5, v6
 ; GFX1250-NEXT:    v_mul_lo_u32 v17, v7, v1
 ; GFX1250-NEXT:    v_mul_lo_u32 v10, v5, v4
diff --git a/llvm/test/CodeGen/AMDGPU/coexec-scheduler.ll b/llvm/test/CodeGen/AMDGPU/coexec-scheduler.ll
index 50f3858e83f9a9..cbf9c42a39591a 100644
--- a/llvm/test/CodeGen/AMDGPU/coexec-scheduler.ll
+++ b/llvm/test/CodeGen/AMDGPU/coexec-scheduler.ll
@@ -746,9 +746,9 @@ define amdgpu_kernel void @ds_wmma_block_carried(ptr addrspace(3) %base, ptr add
 ; GCN-NEXT:    v_mov_b64_e32 v[70:71], s[14:15]
 ; GCN-NEXT:    s_add_co_i32 s0, s0, s1
 ; GCN-NEXT:    v_mov_b64_e32 v[68:69], s[12:13]
-; GCN-NEXT:    v_mov_b64_e32 v[66:67], s[10:11]
 ; GCN-NEXT:    s_wait_dscnt 0x0
 ; GCN-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[40:47], v[56:63], v[16:23]
+; GCN-NEXT:    v_mov_b64_e32 v[66:67], s[10:11]
 ; GCN-NEXT:    v_mov_b64_e32 v[64:65], s[8:9]
 ; GCN-NEXT:    v_wmma_f32_16x16x32_f16 v[48:55], v[24:31], v[32:39], v[48:55]
 ; GCN-NEXT:    ds_load_tr16_b128 v[24:27], v72
@@ -1006,25 +1006,25 @@ define amdgpu_kernel void @ds_wmma_loop_carried(ptr addrspace(3) %base, ptr addr
 ; GCN-NEXT:    s_mov_b32 s14, s8
 ; GCN-NEXT:    s_mov_b32 s15, s8
 ; GCN-NEXT:    v_mov_b64_e32 v[68:69], s[12:13]
-; GCN-NEXT:    v_mov_b64_e32 v[70:71], s[14:15]
-; GCN-NEXT:    v_mov_b64_e32 v[66:67], s[10:11]
-; GCN-NEXT:    v_mov_b64_e32 v[64:65], s[8:9]
 ; GCN-NEXT:    s_wait_kmcnt 0x0
-; GCN-NEXT:    v_mov_b32_e32 v80, s2
 ; GCN-NEXT:    s_bitcmp1_b32 s0, 0
+; GCN-NEXT:    v_mov_b32_e32 v80, s2
+; GCN-NEXT:    v_mov_b64_e32 v[70:71], s[14:15]
+; GCN-NEXT:    s_cselect_b32 s3, -1, 0
 ; GCN-NEXT:    ds_load_tr16_b128 v[32:35], v80 offset:512
 ; GCN-NEXT:    ds_load_tr16_b128 v[36:39], v80 offset:576
 ; GCN-NEXT:    ds_load_tr16_b128 v[48:51], v80 offset:768
 ; GCN-NEXT:    ds_load_tr16_b128 v[52:55], v80 offset:832
 ; GCN-NEXT:    ds_load_tr16_b128 v[72:75], v80 offset:640
 ; GCN-NEXT:    ds_load_tr16_b128 v[76:79], v80 offset:704
+; GCN-NEXT:    v_mov_b64_e32 v[66:67], s[10:11]
 ; GCN-NEXT:    ds_load_tr16_b128 v[8:11], v80
 ; GCN-NEXT:    ds_load_tr16_b128 v[12:15], v80 offset:64
+; GCN-NEXT:    v_mov_b64_e32 v[64:65], s[8:9]
 ; GCN-NEXT:    ds_load_tr16_b128 v[40:43], v80 offset:256
 ; GCN-NEXT:    ds_load_tr16_b128 v[44:47], v80 offset:320
 ; GCN-NEXT:    ds_load_tr16_b128 v[56:59], v80 offset:384
 ; GCN-NEXT:    ds_load_tr16_b128 v[60:63], v80 offset:448
-; GCN-NEXT:    s_cselect_b32 s3, -1, 0
 ; GCN-NEXT:    s_add_co_i32 s0, s2, s1
 ; GCN-NEXT:    s_and_b32 vcc_lo, exec_lo, s3
 ; GCN-NEXT:    s_wait_dscnt 0xa
@@ -1059,9 +1059,9 @@ define amdgpu_kernel void @ds_wmma_loop_carried(ptr addrspace(3) %base, ptr addr
 ; GCN-NEXT:    v_mov_b64_e32 v[70:71], s[14:15]
 ; GCN-NEXT:    s_add_co_i32 s0, s0, s1
 ; GCN-NEXT:    v_mov_b64_e32 v[68:69], s[12:13]
-; GCN-NEXT:    v_mov_b64_e32 v[66:67], s[10:11]
 ; GCN-NEXT:    s_wait_dscnt 0x0
 ; GCN-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[40:47], v[56:63], v[16:23]
+; GCN-NEXT:    v_mov_b64_e32 v[66:67], s[10:11]
 ; GCN-NEXT:    v_mov_b64_e32 v[64:65], s[8:9]
 ; GCN-NEXT:    v_wmma_f32_16x16x32_f16 v[0:7], v[8:15], v[48:55], v[0:7]
 ; GCN-NEXT:    ds_load_tr16_b128 v[8:11], v72
@@ -1308,12 +1308,11 @@ define amdgpu_kernel void @ds_wmma_no_block_carried(ptr addrspace(3) %base, ptr
 ; GCN-NEXT:    s_mov_b32 s14, s8
 ; GCN-NEXT:    s_mov_b32 s15, s8
 ; GCN-NEXT:    v_mov_b64_e32 v[100:101], s[12:13]
-; GCN-NEXT:    v_mov_b64_e32 v[102:103], s[14:15]
-; GCN-NEXT:    v_mov_b64_e32 v[98:99], s[10:11]
-; GCN-NEXT:    v_mov_b64_e32 v[96:97], s[8:9]
 ; GCN-NEXT:    s_wait_kmcnt 0x0
-; GCN-NEXT:    v_mov_b32_e32 v92, s0
 ; GCN-NEXT:    s_bitcmp1_b32 s1, 0
+; GCN-NEXT:    v_mov_b32_e32 v92, s0
+; GCN-NEXT:    v_mov_b64_e32 v[102:103], s[14:15]
+; GCN-NEXT:    s_cselect_b32 s0, -1, 0
 ; GCN-NEXT:    ds_load_tr16_b128 v[0:3], v92
 ; GCN-NEXT:    ds_load_tr16_b128 v[4:7], v92 offset:64
 ; GCN-NEXT:    ds_load_tr16_b128 v[40:43], v92 offset:256
@@ -1330,10 +1329,11 @@ define amdgpu_kernel void @ds_wmma_no_block_carried(ptr addrspace(3) %base, ptr
 ; GCN-NEXT:    ds_load_tr16_b128 v[84:87], v92 offset:704
 ; GCN-NEXT:    ds_load_tr16_b128 v[88:91], v92 offset:896
 ; GCN-NEXT:    ds_load_tr16_b128 v[92:95], v92 offset:960
-; GCN-NEXT:    s_cselect_b32 s0, -1, 0
+; GCN-NEXT:    v_mov_b64_e32 v[98:99], s[10:11]
+; GCN-NEXT:    s_and_b32 vcc_lo, exec_lo, s0
+; GCN-NEXT:    v_mov_b64_e32 v[96:97], s[8:9]
 ; GCN-NEXT:    s_wait_dscnt 0xe
 ; GCN-NEXT:    v_wmma_f32_16x16x32_f16 v[8:15], v[0:7], v[96:103], 0
-; GCN-NEXT:    s_and_b32 vcc_lo, exec_lo, s0
 ; GCN-NEXT:    s_wait_dscnt 0xc
 ; GCN-NEXT:    v_wmma_f32_16x16x32_f16 v[16:23], v[40:47], v[96:103], 0
 ; GCN-NEXT:    s_wait_dscnt 0xa
diff --git a/llvm/test/CodeGen/AMDGPU/flat-saddr-atomics.ll b/llvm/test/CodeGen/AMDGPU/flat-saddr-atomics.ll
index e215c80b472ed5..4cfbee75f05487 100644
--- a/llvm/test/CodeGen/AMDGPU/flat-saddr-atomics.ll
+++ b/llvm/test/CodeGen/AMDGPU/flat-saddr-atomics.ll
@@ -12511,7 +12511,7 @@ define amdgpu_ps <2 x float> @flat_dec_saddr_i64_rtn(ptr inreg %sbase, i32 %voff
 ; GFX1250-SDAG-NEXT:    v_sub_co_u32 v5, s0, v0, 1
 ; GFX1250-SDAG-NEXT:    v_subrev_co_ci_u32_e64 v6, s0, 0, v1, s0
 ; GFX1250-SDAG-NEXT:    s_or_b32 vcc_lo, s0, vcc_lo
-; GFX1250-SDAG-NEXT:    v_dual_cndmask_b32 v3, v6, v3 :: v_dual_cndmask_b32 v2, v5, v2
+; GFX1250-SDAG-NEXT:    v_dual_cndmask_b32 v2, v5, v2 :: v_dual_cndmask_b32 v3, v6, v3
 ; GFX1250-SDAG-NEXT:    scratch_store_b64 v4, v[2:3], off
 ; GFX1250-SDAG-NEXT:    s_wait_xcnt 0x0
 ; GFX1250-SDAG-NEXT:    s_or_b32 exec_lo, exec_lo, s1
@@ -12718,7 +12718,7 @@ define amdgpu_ps <2 x float> @flat_dec_saddr_i64_rtn_neg128(ptr inreg %sbase, i3
 ; GFX1250-SDAG-NEXT:    v_sub_co_u32 v5, s0, v0, 1
 ; GFX1250-SDAG-NEXT:    v_subrev_co_ci_u32_e64 v6, s0, 0, v1, s0
 ; GFX1250-SDAG-NEXT:    s_or_b32 vcc_lo, s0, vcc_lo
-; GFX1250-SDAG-NEXT:    v_dual_cndmask_b32 v3, v6, v3 :: v_dual_cndmask_b32 v2, v5, v2
+; GFX1250-SDAG-NEXT:    v_dual_cndmask_b32 v2, v5, v2 :: v_dual_cndmask_b32 v3, v6, v3
 ; GFX1250-SDAG-NEXT:    scratch_store_b64 v4, v[2:3], off
 ; GFX1250-SDAG-NEXT:    s_wait_xcnt 0x0
 ; GFX1250-SDAG-NEXT:    s_or_b32 exec_lo, exec_lo, s1
@@ -12927,7 +12927,7 @@ define amdgpu_ps void @flat_dec_saddr_i64_nortn(ptr inreg %sbase, i32 %voffset,
 ; GFX1250-SDAG-NEXT:    v_sub_co_u32 v0, s0, v0, 1
 ; GFX1250-SDAG-NEXT:    v_subrev_co_ci_u32_e64 v1, s0, 0, v1, s0
 ; GFX1250-SDAG-NEXT:    s_or_b32 vcc_lo, s0, vcc_lo
-; GFX1250-SDAG-NEXT:    v_dual_cndmask_b32 v1, v1, v3 :: v_dual_cndmask_b32 v0, v0, v2
+; GFX1250-SDAG-NEXT:    v_dual_cndmask_b32 v0, v0, v2 :: v_dual_cndmask_b32 v1, v1, v3
 ; GFX1250-SDAG-NEXT:    scratch_store_b64 v4, v[0:1], off
 ; GFX1250-SDAG-NEXT:    s_endpgm
 ;
@@ -13102,7 +13102,7 @@ define amdgpu_ps void @flat_dec_saddr_i64_nortn_neg128(ptr inreg %sbase, i32 %vo
 ; GFX1250-SDAG-NEXT:    v_sub_co_u32 v0, s0, v0, 1
 ; GFX1250-SDAG-NEXT:    v_subrev_co_ci_u32_e64 v1, s0, 0, v1, s0
 ; GFX1250-SDAG-NEXT:    s_or_b32 vcc_lo, s0, vcc_lo
-; GFX1250-SDAG-NEXT:    v_dual_cndmask_b32 v1, v1, v3 :: v_dual_cndmask_b32 v0, v0, v2
+; GFX1250-SDAG-NEXT:    v_dual_cndmask_b32 v0, v0, v2 :: v_dual_cndmask_b32 v1, v1, v3
 ; GFX1250-SDAG-NEXT:    scratch_store_b64 v4, v[0:1], off
 ; GFX1250-SDAG-NEXT:    s_endpgm
 ;
@@ -13276,8 +13276,8 @@ define double @flat_atomic_fadd_f64_saddr_rtn(ptr inreg %ptr, double %data) {
 ; GFX1250-SDAG-NEXT:    scratch_load_b64 v[2:3], off, s0
 ; GFX1250-SDAG-NEXT:    s_wait_loadcnt 0x0
 ; GFX1250-SDAG-NEXT:    v_add_f64_e32 v[4:5], v[2:3], v[0:1]
-; GFX1250-SDAG-NEXT:    v_mov_b64_e32 v[0:1], v[2:3]
 ; GFX1250-SDAG-NEXT:    scratch_store_b64 off, v[4:5], s0
+; GFX1250-SDAG-NEXT:    v_mov_b64_e32 v[0:1], v[2:3]
 ; GFX1250-SDAG-NEXT:    s_set_pc_i64 s[30:31]
 ; GFX1250-SDAG-NEXT:  .LBB110_4: ; %atomicrmw.global
 ; GFX1250-SDAG-NEXT:    v_mov_b32_e32 v2, 0
@@ -13317,8 +13317,8 @@ define double @flat_atomic_fadd_f64_saddr_rtn(ptr inreg %ptr, double %data) {
 ; GFX1250-GISEL-NEXT:    scratch_load_b64 v[2:3], off, s0
 ; GFX1250-GISEL-NEXT:    s_wait_loadcnt 0x0
 ; GFX1250-GISEL-NEXT:    v_add_f64_e32 v[4:5], v[2:3], v[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[0:1], v[2:3]
 ; GFX1250-GISEL-NEXT:    scratch_store_b64 off, v[4:5], s0
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[0:1], v[2:3]
 ; GFX1250-GISEL-NEXT:    s_set_pc_i64 s[30:31]
 ; GFX1250-GISEL-NEXT:  .LBB110_4: ; %atomicrmw.global
 ; GFX1250-GISEL-NEXT:    v_mov_b32_e32 v2, 0
diff --git a/llvm/test/CodeGen/AMDGPU/float-to-arbitrary-fp-fp8-hw.ll b/llvm/test/CodeGen/AMDGPU/float-to-arbitrary-fp-fp8-hw.ll
index 954d6a3f54cec8..97f14d353b7fa5 100644
--- a/llvm/test/CodeGen/AMDGPU/float-to-arbitrary-fp-fp8-hw.ll
+++ b/llvm/test/CodeGen/AMDGPU/float-to-arbitrary-fp-fp8-hw.ll
@@ -1337,16 +1337,16 @@ define i8 @to_fp8_f64(double %x) {
 ; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    v_frexp_exp_i32_f64_e32 v2, v[0:1]
-; GFX1250-NEXT:    v_frexp_mant_f64_e32 v[4:5], v[0:1]
 ; GFX1250-NEXT:    s_mov_b32 s0, 0xfffff
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_frexp_mant_f64_e32 v[4:5], v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    v_ashrrev_i32_e32 v3, 31, v2
+; GFX1250-NEXT:    v_sub_nc_u64_e32 v[6:7], 37, v[2:3]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_mov_b32_e32 v10, v4
 ; GFX1250-NEXT:    v_and_or_b32 v11, v5, s0, 0x100000
 ; GFX1250-NEXT:    s_mov_b32 s0, 0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1250-NEXT:    v_mov_b32_e32 v10, v4
-; GFX1250-NEXT:    v_sub_nc_u64_e32 v[6:7], 37, v[2:3]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    v_cmp_gt_u64_e32 vcc_lo, 63, v[6:7]
 ; GFX1250-NEXT:    v_dual_cndmask_b32 v7, 0, v7 :: v_dual_cndmask_b32 v6, 63, v6
 ; GFX1250-NEXT:    v_cmp_lt_u64_e32 vcc_lo, 1, v[6:7]
@@ -1374,11 +1374,10 @@ define i8 @to_fp8_f64(double %x) {
 ; GFX1250-NEXT:    v_and_b32_e32 v9, 0x1ffff, v5
 ; GFX1250-NEXT:    v_cmp_ne_u64_e32 vcc_lo, 0, v[8:9]
 ; GFX1250-NEXT:    v_mov_b32_e32 v9, 0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3)
-; GFX1250-NEXT:    v_mov_b32_e32 v5, v9
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v11, 0, 1, vcc_lo
 ; GFX1250-NEXT:    v_cmp_ne_u64_e32 vcc_lo, 0, v[6:7]
-; GFX1250-NEXT:    v_cndmask_b32_e32 v8, 0, v10, vcc_lo
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_dual_mov_b32 v5, v9 :: v_dual_cndmask_b32 v8, 0, v10
 ; GFX1250-NEXT:    v_and_or_b32 v10, v4, 1, v11
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_add_nc_u64_e32 v[6:7], v[12:13], v[8:9]
@@ -1401,26 +1400,26 @@ define i8 @to_fp8_f64(double %x) {
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[8:9], 2, v[2:3]
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v9, 0, 4, s0
 ; GFX1250-NEXT:    v_cmp_eq_u64_e64 s0, 30, v[2:3]
-; GFX1250-NEXT:    v_cmp_lt_i64_e64 s1, 30, v[2:3]
-; GFX1250-NEXT:    v_cmp_gt_i64_e64 s2, 1, v[2:3]
 ; GFX1250-NEXT:    s_and_b32 s0, s0, vcc_lo
-; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cmp_lt_i64_e64 s1, 30, v[2:3]
 ; GFX1250-NEXT:    s_or_b32 vcc_lo, s1, s0
 ; GFX1250-NEXT:    v_and_b32_e32 v7, 0x80, v10
+; GFX1250-NEXT:    v_cmp_gt_i64_e64 s2, 1, v[2:3]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_or_b32_e32 v8, v8, v7
 ; GFX1250-NEXT:    v_or3_b32 v6, v7, v9, v6
 ; GFX1250-NEXT:    v_or_b32_e32 v3, 0x7c, v7
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    v_or_b32_e32 v8, v8, v4
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v2, v8, v6, s2
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc_lo
 ; GFX1250-NEXT:    v_cmp_eq_f64_e32 vcc_lo, 0, v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cndmask_b32_e32 v2, v2, v7, vcc_lo
 ; GFX1250-NEXT:    v_cmp_class_f64_e64 vcc_lo, v[0:1], 0x204
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc_lo
 ; GFX1250-NEXT:    v_cmp_o_f64_e32 vcc_lo, v[0:1], v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cndmask_b32_e32 v0, 0x7e, v2, vcc_lo
 ; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
 ;
diff --git a/llvm/test/CodeGen/AMDGPU/float-to-arbitrary-fp-widen.ll b/llvm/test/CodeGen/AMDGPU/float-to-arbitrary-fp-widen.ll
index 93bb7db3755126..0f428ad8aefbb6 100644
--- a/llvm/test/CodeGen/AMDGPU/float-to-arbitrary-fp-widen.ll
+++ b/llvm/test/CodeGen/AMDGPU/float-to-arbitrary-fp-widen.ll
@@ -2823,7 +2823,7 @@ define <7 x i8> @to_fp8_v7f16(<7 x half> %x) {
 ; GFX1250-FAKE16-NEXT:    v_lshrrev_b64 v[8:9], 24, v[8:9]
 ; GFX1250-FAKE16-NEXT:    v_lshl_or_b32 v1, v2, 16, v3
 ; GFX1250-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT:    v_dual_lshrrev_b32 v1, 8, v1 :: v_dual_mov_b32 v3, v8
+; GFX1250-FAKE16-NEXT:    v_dual_mov_b32 v3, v8 :: v_dual_lshrrev_b32 v1, 8, v1
 ; GFX1250-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
 ;
 ; GFX1310-LABEL: to_fp8_v7f16:
diff --git a/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll b/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
index 4a16912a14589f..3b8005a51a3a35 100644
--- a/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
+++ b/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
@@ -15728,8 +15728,8 @@ define void @global_agent_atomic_fadd_noret_bf16__offset12b_pos__amdgpu_no_fine_
 ; GFX1250-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
 ; GFX1250-TRUE16-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-TRUE16-NEXT:    v_add_nc_u64_e32 v[4:5], 0x7fe, v[0:1]
-; GFX1250-TRUE16-NEXT:    v_mov_b16_e32 v6.l, v2.l
 ; GFX1250-TRUE16-NEXT:    s_mov_b32 s0, 0
+; GFX1250-TRUE16-NEXT:    v_mov_b16_e32 v6.l, v2.l
 ; GFX1250-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
 ; GFX1250-TRUE16-NEXT:    v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
 ; GFX1250-TRUE16-NEXT:    v_and_b32_e32 v4, 3, v4
@@ -18087,8 +18087,8 @@ define void @global_system_atomic_fadd_noret_bf16__offset12b_pos__amdgpu_no_fine
 ; GFX1250-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
 ; GFX1250-TRUE16-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-TRUE16-NEXT:    v_add_nc_u64_e32 v[4:5], 0x7fe, v[0:1]
-; GFX1250-TRUE16-NEXT:    v_mov_b16_e32 v6.l, v2.l
 ; GFX1250-TRUE16-NEXT:    s_mov_b32 s0, 0
+; GFX1250-TRUE16-NEXT:    v_mov_b16_e32 v6.l, v2.l
 ; GFX1250-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1)
 ; GFX1250-TRUE16-NEXT:    v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
 ; GFX1250-TRUE16-NEXT:    v_and_b32_e32 v4, 3, v4
diff --git a/llvm/test/CodeGen/AMDGPU/global-load-xcnt.ll b/llvm/test/CodeGen/AMDGPU/global-load-xcnt.ll
index 20e8e1e7371bda..a61d1a3ff167ab 100644
--- a/llvm/test/CodeGen/AMDGPU/global-load-xcnt.ll
+++ b/llvm/test/CodeGen/AMDGPU/global-load-xcnt.ll
@@ -84,9 +84,9 @@ define i16 @test_v7i16_load_store(ptr addrspace(1) %ptr1, ptr addrspace(1) %ptr2
 ; GCN-SDAG-FAKE16-NEXT:    v_pk_add_u16 v14, v7, v11
 ; GCN-SDAG-FAKE16-NEXT:    v_mov_b64_e32 v[6:7], 12
 ; GCN-SDAG-FAKE16-NEXT:    v_mov_b64_e32 v[10:11], 8
-; GCN-SDAG-FAKE16-NEXT:    v_pk_add_u16 v4, v4, v8
 ; GCN-SDAG-FAKE16-NEXT:    s_wait_xcnt 0x1
 ; GCN-SDAG-FAKE16-NEXT:    v_lshrrev_b32_e32 v0, 16, v9
+; GCN-SDAG-FAKE16-NEXT:    v_pk_add_u16 v4, v4, v8
 ; GCN-SDAG-FAKE16-NEXT:    s_clause 0x2
 ; GCN-SDAG-FAKE16-NEXT:    global_store_b16 v[6:7], v14, off
 ; GCN-SDAG-FAKE16-NEXT:    global_store_b32 v[10:11], v9, off
@@ -324,30 +324,30 @@ define i64 @test_v16i64_load_store(ptr addrspace(1) %ptr_a, ptr addrspace(1) %pt
 ; GCN-SDAG-NEXT:    v_mov_b64_e32 v[0:1], 0x70
 ; GCN-SDAG-NEXT:    v_mov_b64_e32 v[54:55], 48
 ; GCN-SDAG-NEXT:    v_mov_b64_e32 v[2:3], 0x60
-; GCN-SDAG-NEXT:    v_mov_b64_e32 v[64:65], 32
-; GCN-SDAG-NEXT:    v_mov_b64_e32 v[66:67], 16
-; GCN-SDAG-NEXT:    v_mov_b64_e32 v[68:69], 0
-; GCN-SDAG-NEXT:    v_mov_b64_e32 v[38:39], 0x50
-; GCN-SDAG-NEXT:    v_dual_mov_b32 v49, 0 :: v_dual_mov_b32 v48, 0xc8
-; GCN-SDAG-NEXT:    v_mov_b64_e32 v[52:53], 64
 ; GCN-SDAG-NEXT:    s_wait_loadcnt 0x7
 ; GCN-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off
 ; GCN-SDAG-NEXT:    s_wait_loadcnt 0x6
 ; GCN-SDAG-NEXT:    global_store_b128 v[2:3], v[10:13], off
-; GCN-SDAG-NEXT:    s_wait_loadcnt 0x5
-; GCN-SDAG-NEXT:    v_dual_mov_b32 v50, v16 :: v_dual_mov_b32 v51, v17
-; GCN-SDAG-NEXT:    s_wait_xcnt 0x0
-; GCN-SDAG-NEXT:    v_add_nc_u64_e32 v[12:13], v[12:13], v[12:13]
-; GCN-SDAG-NEXT:    s_wait_loadcnt 0x2
-; GCN-SDAG-NEXT:    v_add_nc_u64_e32 v[2:3], v[28:29], v[28:29]
-; GCN-SDAG-NEXT:    s_clause 0x2
+; GCN-SDAG-NEXT:    v_mov_b64_e32 v[64:65], 32
+; GCN-SDAG-NEXT:    v_mov_b64_e32 v[66:67], 16
+; GCN-SDAG-NEXT:    v_mov_b64_e32 v[68:69], 0
+; GCN-SDAG-NEXT:    s_wait_loadcnt 0x4
 ; GCN-SDAG-NEXT:    global_store_b128 v[54:55], v[18:21], off
+; GCN-SDAG-NEXT:    s_wait_loadcnt 0x3
 ; GCN-SDAG-NEXT:    global_store_b128 v[64:65], v[22:25], off
+; GCN-SDAG-NEXT:    s_wait_loadcnt 0x2
 ; GCN-SDAG-NEXT:    global_store_b128 v[66:67], v[26:29], off
 ; GCN-SDAG-NEXT:    s_wait_loadcnt 0x1
 ; GCN-SDAG-NEXT:    global_store_b128 v[68:69], v[30:33], off
+; GCN-SDAG-NEXT:    v_mov_b64_e32 v[38:39], 0x50
+; GCN-SDAG-NEXT:    s_wait_xcnt 0x4
+; GCN-SDAG-NEXT:    v_add_nc_u64_e32 v[2:3], v[28:29], v[28:29]
 ; GCN-SDAG-NEXT:    s_wait_xcnt 0x1
 ; GCN-SDAG-NEXT:    v_add_nc_u64_e32 v[28:29], v[32:33], v[32:33]
+; GCN-SDAG-NEXT:    v_dual_mov_b32 v49, 0 :: v_dual_mov_b32 v48, 0xc8
+; GCN-SDAG-NEXT:    v_mov_b64_e32 v[52:53], 64
+; GCN-SDAG-NEXT:    v_dual_mov_b32 v50, v16 :: v_dual_mov_b32 v51, v17
+; GCN-SDAG-NEXT:    v_add_nc_u64_e32 v[12:13], v[12:13], v[12:13]
 ; GCN-SDAG-NEXT:    v_add_nc_u64_e32 v[10:11], v[10:11], v[10:11]
 ; GCN-SDAG-NEXT:    v_add_nc_u64_e32 v[8:9], v[8:9], v[8:9]
 ; GCN-SDAG-NEXT:    v_add_nc_u64_e32 v[6:7], v[6:7], v[6:7]
@@ -399,9 +399,6 @@ define i64 @test_v16i64_load_store(ptr addrspace(1) %ptr_a, ptr addrspace(1) %pt
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[52:53], 48
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[66:67], 0x60
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[68:69], 0x70
-; GCN-GISEL-NEXT:    v_mov_b64_e32 v[54:55], 64
-; GCN-GISEL-NEXT:    v_mov_b64_e32 v[48:49], 0xc8
-; GCN-GISEL-NEXT:    v_mov_b64_e32 v[64:65], 0x50
 ; GCN-GISEL-NEXT:    s_wait_loadcnt 0x6
 ; GCN-GISEL-NEXT:    global_store_b128 v[0:1], v[10:13], off
 ; GCN-GISEL-NEXT:    s_wait_loadcnt 0x5
@@ -414,8 +411,11 @@ define i64 @test_v16i64_load_store(ptr addrspace(1) %ptr_a, ptr addrspace(1) %pt
 ; GCN-GISEL-NEXT:    global_store_b128 v[66:67], v[26:29], off
 ; GCN-GISEL-NEXT:    s_wait_loadcnt 0x1
 ; GCN-GISEL-NEXT:    global_store_b128 v[68:69], v[30:33], off
+; GCN-GISEL-NEXT:    v_mov_b64_e32 v[54:55], 64
 ; GCN-GISEL-NEXT:    s_wait_xcnt 0x4
 ; GCN-GISEL-NEXT:    v_add_nc_u64_e32 v[2:3], v[12:13], v[12:13]
+; GCN-GISEL-NEXT:    v_mov_b64_e32 v[48:49], 0xc8
+; GCN-GISEL-NEXT:    v_mov_b64_e32 v[64:65], 0x50
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[50:51], v[8:9]
 ; GCN-GISEL-NEXT:    v_add_nc_u64_e32 v[0:1], v[10:11], v[10:11]
 ; GCN-GISEL-NEXT:    v_add_nc_u64_e32 v[10:11], v[14:15], v[14:15]
@@ -476,12 +476,12 @@ define amdgpu_kernel void @test_v7i16_load_store_kernel(ptr addrspace(1) %ptr1,
 ; GCN-SDAG-NEXT:    s_wait_xcnt 0x0
 ; GCN-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x10 nv
 ; GCN-SDAG-NEXT:    v_mov_b64_e32 v[8:9], 12
-; GCN-SDAG-NEXT:    v_mov_b64_e32 v[10:11], 8
-; GCN-SDAG-NEXT:    v_mov_b64_e32 v[12:13], 0
 ; GCN-SDAG-NEXT:    s_wait_kmcnt 0x0
 ; GCN-SDAG-NEXT:    s_clause 0x1
 ; GCN-SDAG-NEXT:    global_load_b128 v[0:3], v14, s[0:1] scale_offset
 ; GCN-SDAG-NEXT:    global_load_b128 v[4:7], v14, s[2:3] scale_offset
+; GCN-SDAG-NEXT:    v_mov_b64_e32 v[10:11], 8
+; GCN-SDAG-NEXT:    v_mov_b64_e32 v[12:13], 0
 ; GCN-SDAG-NEXT:    s_wait_loadcnt 0x0
 ; GCN-SDAG-NEXT:    v_pk_add_u16 v3, v3, v7
 ; GCN-SDAG-NEXT:    v_pk_add_u16 v2, v2, v6
@@ -506,16 +506,16 @@ define amdgpu_kernel void @test_v7i16_load_store_kernel(ptr addrspace(1) %ptr1,
 ; GCN-GISEL-NEXT:    s_wait_xcnt 0x0
 ; GCN-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x10 nv
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[8:9], 0
+; GCN-GISEL-NEXT:    s_wait_kmcnt 0x0
+; GCN-GISEL-NEXT:    s_clause 0x1
+; GCN-GISEL-NEXT:    global_load_b128 v[0:3], v22, s[0:1] scale_offset
+; GCN-GISEL-NEXT:    global_load_b128 v[4:7], v22, s[2:3] scale_offset
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[10:11], 2
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[12:13], 4
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[14:15], 6
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[16:17], 8
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[18:19], 10
 ; GCN-GISEL-NEXT:    v_mov_b64_e32 v[20:21], 12
-; GCN-GISEL-NEXT:    s_wait_kmcnt 0x0
-; GCN-GISEL-NEXT:    s_clause 0x1
-; GCN-GISEL-NEXT:    global_load_b128 v[0:3], v22, s[0:1] scale_offset
-; GCN-GISEL-NEXT:    global_load_b128 v[4:7], v22, s[2:3] scale_offset
 ; GCN-GISEL-NEXT:    s_wait_loadcnt 0x0
 ; GCN-GISEL-NEXT:    v_pk_add_u16 v0, v0, v4
 ; GCN-GISEL-NEXT:    v_pk_add_u16 v1, v1, v5
diff --git a/llvm/test/CodeGen/AMDGPU/integer-mad-patterns.ll b/llvm/test/CodeGen/AMDGPU/integer-mad-patterns.ll
index 1a4f06cfcf694f..2a71b2118b588e 100644
--- a/llvm/test/CodeGen/AMDGPU/integer-mad-patterns.ll
+++ b/llvm/test/CodeGen/AMDGPU/integer-mad-patterns.ll
@@ -4174,16 +4174,16 @@ define <3 x i32> @clpeak_imad_pat_v3i32(<3 x i32> %x, <3 x i32> %y) {
 ; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-SDAG-NEXT:    v_add_nc_u32_e32 v2, v8, v2
 ; GFX1250-SDAG-NEXT:    v_mul_lo_u32 v1, v1, v4
-; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1250-SDAG-NEXT:    v_mad_u32 v3, v0, v6, v0
+; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-SDAG-NEXT:    v_mul_lo_u32 v2, v2, v5
+; GFX1250-SDAG-NEXT:    v_mad_u32 v3, v0, v6, v0
 ; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-SDAG-NEXT:    v_mad_u32 v4, v1, v7, v1
-; GFX1250-SDAG-NEXT:    v_mad_u32 v0, v3, v0, v3
-; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-SDAG-NEXT:    v_mad_u32 v5, v2, v8, v2
+; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-SDAG-NEXT:    v_mad_u32 v0, v3, v0, v3
 ; GFX1250-SDAG-NEXT:    v_mad_u32 v1, v4, v1, v4
-; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-SDAG-NEXT:    v_mad_u32 v2, v5, v2, v5
 ; GFX1250-SDAG-NEXT:    s_set_pc_i64 s[30:31]
 ;
@@ -4202,23 +4202,22 @@ define <3 x i32> @clpeak_imad_pat_v3i32(<3 x i32> %x, <3 x i32> %y) {
 ; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-GISEL-FAKE16-NEXT:    v_add_nc_u32_e32 v1, v7, v1
 ; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v0, v0, v3
-; GFX1250-GISEL-FAKE16-NEXT:    v_add_nc_u32_e32 v3, 1, v6
-; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-GISEL-FAKE16-NEXT:    v_add_nc_u32_e32 v2, v8, v2
 ; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v1, v1, v4
-; GFX1250-GISEL-FAKE16-NEXT:    v_add_nc_u32_e32 v4, 1, v7
-; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v3, v0, v3
+; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v2, v2, v5
+; GFX1250-GISEL-FAKE16-NEXT:    v_dual_add_nc_u32 v3, 1, v6 :: v_dual_add_nc_u32 v4, 1, v7
 ; GFX1250-GISEL-FAKE16-NEXT:    v_add_nc_u32_e32 v5, 1, v8
-; GFX1250-GISEL-FAKE16-NEXT:    v_add_nc_u32_e32 v0, 1, v0
+; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v3, v0, v3
+; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_4)
 ; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v4, v1, v4
 ; GFX1250-GISEL-FAKE16-NEXT:    v_add_nc_u32_e32 v1, 1, v1
-; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_2) | instid1(VALU_DEP_4)
 ; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v5, v2, v5
-; GFX1250-GISEL-FAKE16-NEXT:    v_add_nc_u32_e32 v2, 1, v2
-; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v0, v3, v0
+; GFX1250-GISEL-FAKE16-NEXT:    v_dual_add_nc_u32 v0, 1, v0 :: v_dual_add_nc_u32 v2, 1, v2
+; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v1, v4, v1
+; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v0, v3, v0
 ; GFX1250-GISEL-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-FAKE16-NEXT:    v_mul_lo_u32 v2, v5, v2
 ; GFX1250-GISEL-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
@@ -4238,23 +4237,22 @@ define <3 x i32> @clpeak_imad_pat_v3i32(<3 x i32> %x, <3 x i32> %y) {
 ; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-GISEL-REAL16-NEXT:    v_add_nc_u32_e32 v1, v7, v1
 ; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v0, v0, v3
-; GFX1250-GISEL-REAL16-NEXT:    v_add_nc_u32_e32 v3, 1, v6
-; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-GISEL-REAL16-NEXT:    v_add_nc_u32_e32 v2, v8, v2
 ; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v1, v1, v4
-; GFX1250-GISEL-REAL16-NEXT:    v_add_nc_u32_e32 v4, 1, v7
-; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v3, v0, v3
+; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v2, v2, v5
+; GFX1250-GISEL-REAL16-NEXT:    v_dual_add_nc_u32 v3, 1, v6 :: v_dual_add_nc_u32 v4, 1, v7
 ; GFX1250-GISEL-REAL16-NEXT:    v_add_nc_u32_e32 v5, 1, v8
-; GFX1250-GISEL-REAL16-NEXT:    v_add_nc_u32_e32 v0, 1, v0
+; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v3, v0, v3
+; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_4)
 ; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v4, v1, v4
 ; GFX1250-GISEL-REAL16-NEXT:    v_add_nc_u32_e32 v1, 1, v1
-; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_2) | instid1(VALU_DEP_4)
 ; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v5, v2, v5
-; GFX1250-GISEL-REAL16-NEXT:    v_add_nc_u32_e32 v2, 1, v2
-; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v0, v3, v0
+; GFX1250-GISEL-REAL16-NEXT:    v_dual_add_nc_u32 v0, 1, v0 :: v_dual_add_nc_u32 v2, 1, v2
+; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v1, v4, v1
+; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v0, v3, v0
 ; GFX1250-GISEL-REAL16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-REAL16-NEXT:    v_mul_lo_u32 v2, v5, v2
 ; GFX1250-GISEL-REAL16-NEXT:    s_set_pc_i64 s[30:31]
@@ -11984,9 +11982,8 @@ define i32 @other_use_mul_mad_i32_var(i32 %x, i32 %y, i32 %z, ptr addrspace(3) %
 ; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    v_mul_lo_u32 v1, v0, v1
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1250-NEXT:    v_add_nc_u32_e32 v0, v1, v2
 ; GFX1250-NEXT:    ds_store_b32 v3, v1
+; GFX1250-NEXT:    v_add_nc_u32_e32 v0, v1, v2
 ; GFX1250-NEXT:    s_wait_dscnt 0x0
 ; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
 entry:
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.scale.pk.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.scale.pk.ll
index 91b38c7f421354..8dd9c3f13e60d1 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.scale.pk.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.scale.pk.ll
@@ -309,9 +309,9 @@ define void @test_cvt_scale_pk8_f32_fp8_ss(<2 x i32> inreg %src, i32 inreg %scal
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
@@ -380,9 +380,9 @@ define void @test_cvt_scale_pk8_f32_bf8_ss(<2 x i32> inreg %src, i32 inreg %scal
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
@@ -449,9 +449,9 @@ define void @test_cvt_scale_pk8_f32_fp4_ss(i32 inreg %src, i32 inreg %scale, ptr
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
@@ -520,9 +520,9 @@ define void @test_cvt_scale_pk16_f16_fp6_ss(<3 x i32> inreg %src, i32 inreg %sca
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
@@ -566,9 +566,9 @@ define void @test_cvt_scale_pk16_f16_fp6_sl(<3 x i32> inreg %src, ptr addrspace(
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
@@ -637,9 +637,9 @@ define void @test_cvt_scale_pk16_bf16_fp6_sl(<3 x i32> inreg %src, ptr addrspace
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
@@ -708,9 +708,9 @@ define void @test_cvt_scale_pk16_f16_bf6_ss(<3 x i32> inreg %src, i32 inreg %sca
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
@@ -754,9 +754,9 @@ define void @test_cvt_scale_pk16_f16_bf6_sl(<3 x i32> inreg %src, ptr addrspace(
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
@@ -825,9 +825,9 @@ define void @test_cvt_scale_pk16_bf16_bf6_sl(<3 x i32> inreg %src, ptr addrspace
 ; GFX1250-GISEL-NEXT:    v_readfirstlane_b32 s7, v9
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[2:3]
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[0:1]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[8:9], s[6:7]
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[6:7], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_clause 0x1
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
 ; GFX1250-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
index 22abc40fb4404e..a3c8950224b2b7 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
@@ -342,9 +342,9 @@ define weak_odr amdgpu_kernel void @dpp_test1(ptr %arg) local_unnamed_addr {
 ; GFX1250-NEXT:    s_wait_dscnt 0x0
 ; GFX1250-NEXT:    v_add_nc_u32_e32 v1, v1, v1
 ; GFX1250-NEXT:    v_mov_b32_dpp v2, v1 quad_perm:[1,0,3,2] row_mask:0xf bank_mask:0xf
+; GFX1250-NEXT:    s_barrier_wait -1
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX1250-NEXT:    v_add_nc_u32_e32 v1, v2, v1
-; GFX1250-NEXT:    s_barrier_wait -1
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    flat_store_b32 v0, v1, s[0:1]
 ; GFX1250-NEXT:    s_endpgm
diff --git a/llvm/test/CodeGen/AMDGPU/lshl-add-u64.ll b/llvm/test/CodeGen/AMDGPU/lshl-add-u64.ll
index 3f3cde65dfdf7a..1a3d978e3c7e9b 100644
--- a/llvm/test/CodeGen/AMDGPU/lshl-add-u64.ll
+++ b/llvm/test/CodeGen/AMDGPU/lshl-add-u64.ll
@@ -754,11 +754,10 @@ define <4 x i64> @lshl_add_u64_vvv_and_2_v4(<4 x i64> %v, <4 x i64> %a, <4 x i64
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[6:7], v17, v[6:7]
 ; GFX1250-NEXT:    v_lshl_add_u64 v[0:1], v[0:1], v16, v[8:9]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1250-NEXT:    v_lshl_add_u64 v[4:5], v[4:5], v19, v[12:13]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_lshl_add_u64 v[2:3], v[2:3], v18, v[10:11]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX1250-NEXT:    v_add_nc_u64_e32 v[6:7], v[6:7], v[14:15]
+; GFX1250-NEXT:    v_lshl_add_u64 v[4:5], v[4:5], v19, v[12:13]
 ; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
 ;
 ; GFX942-GISEL-LABEL: lshl_add_u64_vvv_and_2_v4:
@@ -797,11 +796,10 @@ define <4 x i64> @lshl_add_u64_vvv_and_2_v4(<4 x i64> %v, <4 x i64> %a, <4 x i64
 ; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-GISEL-NEXT:    v_lshlrev_b64_e32 v[6:7], v17, v[6:7]
 ; GFX1250-GISEL-NEXT:    v_lshl_add_u64 v[0:1], v[0:1], v16, v[8:9]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-GISEL-NEXT:    v_lshl_add_u64 v[2:3], v[2:3], v18, v[10:11]
-; GFX1250-GISEL-NEXT:    v_lshl_add_u64 v[4:5], v[4:5], v19, v[12:13]
-; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX1250-GISEL-NEXT:    v_add_nc_u64_e32 v[6:7], v[6:7], v[14:15]
+; GFX1250-GISEL-NEXT:    v_lshl_add_u64 v[4:5], v[4:5], v19, v[12:13]
 ; GFX1250-GISEL-NEXT:    s_set_pc_i64 s[30:31]
   %and = and <4 x i64> %s, <i64 1, i64 2, i64 3, i64 5>
   %shl = shl <4 x i64> %v, %and
diff --git a/llvm/test/CodeGen/AMDGPU/mad_64_32.ll b/llvm/test/CodeGen/AMDGPU/mad_64_32.ll
index 68767c5183c3ab..416d0e5130b10f 100644
--- a/llvm/test/CodeGen/AMDGPU/mad_64_32.ll
+++ b/llvm/test/CodeGen/AMDGPU/mad_64_32.ll
@@ -428,10 +428,10 @@ define i128 @mad_i64_i32_sextops_i32_i128(i32 %arg0, i32 %arg1, i128 %arg2) #0 {
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[6:7], v0, v1, 0
 ; GFX1250-NEXT:    v_dual_mov_b32 v9, 0 :: v_dual_ashrrev_i32 v16, 31, v0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1250-NEXT:    v_mad_nc_i64_i32 v[14:15], v1, v16, 0
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_dual_mov_b32 v8, v7 :: v_dual_ashrrev_i32 v7, 31, v1
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_mad_nc_i64_i32 v[14:15], v1, v16, 0
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[10:11], v16, v1, v[8:9]
 ; GFX1250-NEXT:    v_dual_mov_b32 v8, v11 :: v_dual_mov_b32 v11, v9
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3)
@@ -2318,9 +2318,9 @@ define <2 x i64> @lshr_mad_i64_vec(<2 x i64> %arg0) #0 {
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[4:5], v1, 0xffff1c18, v[0:1]
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[6:7], v3, 0xffff1118, v[2:3]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1250-NEXT:    v_dual_sub_nc_u32 v1, v5, v1 :: v_dual_sub_nc_u32 v3, v7, v3
-; GFX1250-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v2, v6
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_sub_nc_u32 v1, v5, v1
+; GFX1250-NEXT:    v_dual_mov_b32 v2, v6 :: v_dual_sub_nc_u32 v3, v7, v3
 ; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
   %lsh = lshr <2 x i64> %arg0, <i64 32, i64 32>
   %mul = mul <2 x i64> %lsh, <i64 s0xffffffffffff1c18, i64 s0xffffffffffff1118>
diff --git a/llvm/test/CodeGen/AMDGPU/mul.ll b/llvm/test/CodeGen/AMDGPU/mul.ll
index a0b580a3a774cc..14e1aa694352ee 100644
--- a/llvm/test/CodeGen/AMDGPU/mul.ll
+++ b/llvm/test/CodeGen/AMDGPU/mul.ll
@@ -3937,11 +3937,11 @@ define amdgpu_kernel void @v_mul_i128(ptr addrspace(1) %out, ptr addrspace(1) %a
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[2:3], v0, v5, v[14:15]
 ; GFX1250-NEXT:    v_mad_u32 v0, v7, v0, v11
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_dual_mov_b32 v14, v3 :: v_dual_mov_b32 v9, v2
-; GFX1250-NEXT:    v_add_nc_u64_e32 v[12:13], v[12:13], v[14:15]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_mad_u32 v11, v6, v1, v0
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_add_nc_u64_e32 v[12:13], v[12:13], v[14:15]
 ; GFX1250-NEXT:    v_mad_nc_u64_u32 v[0:1], v1, v5, v[12:13]
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX1250-NEXT:    v_add_nc_u64_e32 v[10:11], v[0:1], v[10:11]
diff --git a/llvm/test/CodeGen/AMDGPU/packed-fp32.ll b/llvm/test/CodeGen/AMDGPU/packed-fp32.ll
index 79d6e37c7f21e6..8fecac089f08c2 100644
--- a/llvm/test/CodeGen/AMDGPU/packed-fp32.ll
+++ b/llvm/test/CodeGen/AMDGPU/packed-fp32.ll
@@ -3425,10 +3425,11 @@ define amdgpu_kernel void @fma_v2_v_unfoldable_lit(ptr addrspace(1) %a) {
 ; GFX1250-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24 nv
 ; GFX1250-SDAG-NEXT:    v_and_b32_e32 v6, 0x3ff, v0
 ; GFX1250-SDAG-NEXT:    v_mov_b64_e32 v[2:3], 0x4040000040800000
-; GFX1250-SDAG-NEXT:    v_mov_b64_e32 v[4:5], 0x400000003f800000
 ; GFX1250-SDAG-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-SDAG-NEXT:    global_load_b64 v[0:1], v6, s[0:1] scale_offset
+; GFX1250-SDAG-NEXT:    v_mov_b64_e32 v[4:5], 0x400000003f800000
 ; GFX1250-SDAG-NEXT:    s_wait_loadcnt 0x0
+; GFX1250-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX1250-SDAG-NEXT:    v_pk_fma_f32 v[0:1], v[0:1], v[2:3], v[4:5]
 ; GFX1250-SDAG-NEXT:    global_store_b64 v6, v[0:1], s[0:1] scale_offset
 ; GFX1250-SDAG-NEXT:    s_endpgm
@@ -3445,10 +3446,11 @@ define amdgpu_kernel void @fma_v2_v_unfoldable_lit(ptr addrspace(1) %a) {
 ; GFX1250-GISEL-NEXT:    s_wait_xcnt 0x0
 ; GFX1250-GISEL-NEXT:    s_mov_b64 s[4:5], 0x400000003f800000
 ; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[2:3], s[2:3]
-; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-GISEL-NEXT:    global_load_b64 v[0:1], v6, s[0:1] scale_offset
+; GFX1250-GISEL-NEXT:    v_mov_b64_e32 v[4:5], s[4:5]
 ; GFX1250-GISEL-NEXT:    s_wait_loadcnt 0x0
+; GFX1250-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX1250-GISEL-NEXT:    v_pk_fma_f32 v[0:1], v[0:1], v[2:3], v[4:5]
 ; GFX1250-GISEL-NEXT:    global_store_b64 v6, v[0:1], s[0:1] scale_offset
 ; GFX1250-GISEL-NEXT:    s_endpgm
diff --git a/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll b/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll
index 8bf811e335b13e..6e968c757e827d 100644
--- a/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll
+++ b/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll
@@ -5236,37 +5236,38 @@ define amdgpu_kernel void @compute_mad(ptr addrspace(4) %i18, ptr addrspace(4) %
 ; GFX1250-NEXT:    s_and_b32 s4, ttmp6, 15
 ; GFX1250-NEXT:    s_getreg_b32 s5, hwreg(HW_REG_IB_STS2, 6, 4)
 ; GFX1250-NEXT:    v_mul_lo_u32 v2, v2, v0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX1250-NEXT:    v_mul_lo_u32 v3, v2, v1
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    s_load_b32 s2, s[2:3], 0x4 nv
-; GFX1250-NEXT:    v_add_nc_u32_e32 v1, v3, v1
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(SALU_CYCLE_1)
+; GFX1250-NEXT:    v_mul_lo_u32 v3, v2, v1
 ; GFX1250-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0 nv
 ; GFX1250-NEXT:    s_wait_xcnt 0x0
 ; GFX1250-NEXT:    s_bfe_u32 s3, ttmp6, 0x4000c
-; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(SALU_CYCLE_1)
 ; GFX1250-NEXT:    s_add_co_i32 s3, s3, 1
-; GFX1250-NEXT:    v_mul_lo_u32 v1, v1, v2
-; GFX1250-NEXT:    v_add_nc_u32_e32 v2, 1, v3
+; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1250-NEXT:    s_mul_i32 s3, ttmp9, s3
+; GFX1250-NEXT:    v_add_nc_u32_e32 v1, v3, v1
 ; GFX1250-NEXT:    s_add_co_i32 s4, s4, s3
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
-; GFX1250-NEXT:    v_mul_lo_u32 v3, v1, v2
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_mul_lo_u32 v1, v1, v2
 ; GFX1250-NEXT:    s_wait_kmcnt 0x0
 ; GFX1250-NEXT:    s_and_b32 s2, s2, 0xffff
+; GFX1250-NEXT:    v_add_nc_u32_e32 v2, 1, v3
 ; GFX1250-NEXT:    s_cmp_eq_u32 s5, 0
-; GFX1250-NEXT:    v_add_nc_u32_e32 v2, v3, v2
 ; GFX1250-NEXT:    s_cselect_b32 s3, ttmp9, s4
-; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_mul_lo_u32 v3, v1, v2
 ; GFX1250-NEXT:    v_mad_u32 v0, s3, s2, v0
+; GFX1250-NEXT:    v_add_nc_u32_e32 v2, v3, v2
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_mul_lo_u32 v2, v2, v1
 ; GFX1250-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1250-NEXT:    v_add_nc_u64_e32 v[0:1], s[0:1], v[0:1]
 ; GFX1250-NEXT:    v_mad_u32 v3, v2, v3, v2
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1250-NEXT:    v_lshl_add_u64 v[0:1], v[0:1], 2, s[8:9]
+; GFX1250-NEXT:    v_add_nc_u64_e32 v[0:1], s[0:1], v[0:1]
 ; GFX1250-NEXT:    v_mad_u32 v2, v3, v2, v3
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1250-NEXT:    v_lshl_add_u64 v[0:1], v[0:1], 2, s[8:9]
 ; GFX1250-NEXT:    global_store_b32 v[0:1], v2, off
 ; GFX1250-NEXT:    s_endpgm
 ;
diff --git a/llvm/test/CodeGen/AMDGPU/shl.v2i64.ll b/llvm/test/CodeGen/AMDGPU/shl.v2i64.ll
index 067255d621c2c0..359ec532408fae 100644
--- a/llvm/test/CodeGen/AMDGPU/shl.v2i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/shl.v2i64.ll
@@ -334,44 +334,82 @@ define amdgpu_kernel void @shl_v_imm_v4i64(ptr addrspace(1) %out, ptr addrspace(
 }
 
 define amdgpu_kernel void @v_shl_v8i64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
-; GCN-LABEL: v_shl_v8i64:
-; GCN:       ; %bb.0:
-; GCN-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
-; GCN-NEXT:    s_mov_b64 s[64:65], 0
-; GCN-NEXT:    v_nop
-; GCN-NEXT:    global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
-; GCN-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24 nv
-; GCN-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
-; GCN-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GCN-NEXT:    v_lshlrev_b32_e32 v32, 6, v0
-; GCN-NEXT:    s_wait_kmcnt 0x0
-; GCN-NEXT:    s_clause 0x7
-; GCN-NEXT:    global_load_b128 v[0:3], v32, s[2:3]
-; GCN-NEXT:    global_load_b128 v[4:7], v32, s[2:3] offset:16
-; GCN-NEXT:    global_load_b128 v[8:11], v32, s[2:3] offset:64
-; GCN-NEXT:    global_load_b128 v[12:15], v32, s[2:3] offset:80
-; GCN-NEXT:    global_load_b128 v[16:19], v32, s[2:3] offset:32
-; GCN-NEXT:    global_load_b128 v[20:23], v32, s[2:3] offset:48
-; GCN-NEXT:    global_load_b128 v[24:27], v32, s[2:3] offset:112
-; GCN-NEXT:    global_load_b128 v[28:31], v32, s[2:3] offset:96
-; GCN-NEXT:    s_wait_loadcnt 0x5
-; GCN-NEXT:    v_lshlrev_b64_e32 v[2:3], v10, v[2:3]
-; GCN-NEXT:    v_lshlrev_b64_e32 v[0:1], v8, v[0:1]
-; GCN-NEXT:    s_wait_loadcnt 0x4
-; GCN-NEXT:    v_lshlrev_b64_e32 v[6:7], v14, v[6:7]
-; GCN-NEXT:    v_lshlrev_b64_e32 v[4:5], v12, v[4:5]
-; GCN-NEXT:    s_wait_loadcnt 0x1
-; GCN-NEXT:    v_lshlrev_b64_e32 v[22:23], v26, v[22:23]
-; GCN-NEXT:    s_wait_loadcnt 0x0
-; GCN-NEXT:    v_lshlrev_b64_e32 v[18:19], v30, v[18:19]
-; GCN-NEXT:    v_lshlrev_b64_e32 v[16:17], v28, v[16:17]
-; GCN-NEXT:    v_lshlrev_b64_e32 v[20:21], v24, v[20:21]
-; GCN-NEXT:    s_clause 0x3
-; GCN-NEXT:    global_store_b128 v32, v[16:19], s[0:1] offset:32
-; GCN-NEXT:    global_store_b128 v32, v[20:23], s[0:1] offset:48
-; GCN-NEXT:    global_store_b128 v32, v[0:3], s[0:1]
-; GCN-NEXT:    global_store_b128 v32, v[4:7], s[0:1] offset:16
-; GCN-NEXT:    s_endpgm
+; GFX1250-LABEL: v_shl_v8i64:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
+; GFX1250-NEXT:    s_mov_b64 s[64:65], 0
+; GFX1250-NEXT:    v_nop
+; GFX1250-NEXT:    global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
+; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24 nv
+; GFX1250-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_lshlrev_b32_e32 v32, 6, v0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    s_clause 0x7
+; GFX1250-NEXT:    global_load_b128 v[0:3], v32, s[2:3]
+; GFX1250-NEXT:    global_load_b128 v[4:7], v32, s[2:3] offset:16
+; GFX1250-NEXT:    global_load_b128 v[8:11], v32, s[2:3] offset:64
+; GFX1250-NEXT:    global_load_b128 v[12:15], v32, s[2:3] offset:80
+; GFX1250-NEXT:    global_load_b128 v[16:19], v32, s[2:3] offset:32
+; GFX1250-NEXT:    global_load_b128 v[20:23], v32, s[2:3] offset:48
+; GFX1250-NEXT:    global_load_b128 v[24:27], v32, s[2:3] offset:112
+; GFX1250-NEXT:    global_load_b128 v[28:31], v32, s[2:3] offset:96
+; GFX1250-NEXT:    s_wait_loadcnt 0x5
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[2:3], v10, v[2:3]
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[0:1], v8, v[0:1]
+; GFX1250-NEXT:    s_wait_loadcnt 0x1
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[22:23], v26, v[22:23]
+; GFX1250-NEXT:    s_wait_loadcnt 0x0
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[18:19], v30, v[18:19]
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[16:17], v28, v[16:17]
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[20:21], v24, v[20:21]
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[6:7], v14, v[6:7]
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[4:5], v12, v[4:5]
+; GFX1250-NEXT:    s_clause 0x3
+; GFX1250-NEXT:    global_store_b128 v32, v[16:19], s[0:1] offset:32
+; GFX1250-NEXT:    global_store_b128 v32, v[20:23], s[0:1] offset:48
+; GFX1250-NEXT:    global_store_b128 v32, v[0:3], s[0:1]
+; GFX1250-NEXT:    global_store_b128 v32, v[4:7], s[0:1] offset:16
+; GFX1250-NEXT:    s_endpgm
+;
+; GFX1251-LABEL: v_shl_v8i64:
+; GFX1251:       ; %bb.0:
+; GFX1251-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
+; GFX1251-NEXT:    s_mov_b64 s[64:65], 0
+; GFX1251-NEXT:    v_nop
+; GFX1251-NEXT:    global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
+; GFX1251-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24 nv
+; GFX1251-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GFX1251-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1251-NEXT:    v_lshlrev_b32_e32 v32, 6, v0
+; GFX1251-NEXT:    s_wait_kmcnt 0x0
+; GFX1251-NEXT:    s_clause 0x7
+; GFX1251-NEXT:    global_load_b128 v[0:3], v32, s[2:3]
+; GFX1251-NEXT:    global_load_b128 v[4:7], v32, s[2:3] offset:16
+; GFX1251-NEXT:    global_load_b128 v[8:11], v32, s[2:3] offset:64
+; GFX1251-NEXT:    global_load_b128 v[12:15], v32, s[2:3] offset:80
+; GFX1251-NEXT:    global_load_b128 v[16:19], v32, s[2:3] offset:32
+; GFX1251-NEXT:    global_load_b128 v[20:23], v32, s[2:3] offset:48
+; GFX1251-NEXT:    global_load_b128 v[24:27], v32, s[2:3] offset:112
+; GFX1251-NEXT:    global_load_b128 v[28:31], v32, s[2:3] offset:96
+; GFX1251-NEXT:    s_wait_loadcnt 0x5
+; GFX1251-NEXT:    v_lshlrev_b64_e32 v[2:3], v10, v[2:3]
+; GFX1251-NEXT:    v_lshlrev_b64_e32 v[0:1], v8, v[0:1]
+; GFX1251-NEXT:    s_wait_loadcnt 0x4
+; GFX1251-NEXT:    v_lshlrev_b64_e32 v[6:7], v14, v[6:7]
+; GFX1251-NEXT:    v_lshlrev_b64_e32 v[4:5], v12, v[4:5]
+; GFX1251-NEXT:    s_wait_loadcnt 0x1
+; GFX1251-NEXT:    v_lshlrev_b64_e32 v[22:23], v26, v[22:23]
+; GFX1251-NEXT:    s_wait_loadcnt 0x0
+; GFX1251-NEXT:    v_lshlrev_b64_e32 v[18:19], v30, v[18:19]
+; GFX1251-NEXT:    v_lshlrev_b64_e32 v[16:17], v28, v[16:17]
+; GFX1251-NEXT:    v_lshlrev_b64_e32 v[20:21], v24, v[20:21]
+; GFX1251-NEXT:    s_clause 0x3
+; GFX1251-NEXT:    global_store_b128 v32, v[16:19], s[0:1] offset:32
+; GFX1251-NEXT:    global_store_b128 v32, v[20:23], s[0:1] offset:48
+; GFX1251-NEXT:    global_store_b128 v32, v[0:3], s[0:1]
+; GFX1251-NEXT:    global_store_b128 v32, v[4:7], s[0:1] offset:16
+; GFX1251-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <8 x i64>, ptr addrspace(1) %in, i64 %tid.ext
diff --git a/llvm/test/CodeGen/AMDGPU/slp-int-to-fp.ll b/llvm/test/CodeGen/AMDGPU/slp-int-to-fp.ll
index f56b30d227fc7e..d0e2bb112a931a 100644
--- a/llvm/test/CodeGen/AMDGPU/slp-int-to-fp.ll
+++ b/llvm/test/CodeGen/AMDGPU/slp-int-to-fp.ll
@@ -1147,34 +1147,33 @@ define void @sitofp_i48_to_half(ptr addrspace(1) noalias %out, ptr addrspace(1)
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[4:5], v12, v[4:5]
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[8:9], v13, v[8:9]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1250-NEXT:    v_lshlrev_b64_e32 v[10:11], v14, v[10:11]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[6:7], v16, v[6:7]
-; GFX1250-NEXT:    v_dual_sub_nc_u32 v12, 32, v12 :: v_dual_sub_nc_u32 v14, 32, v14
+; GFX1250-NEXT:    v_lshlrev_b64_e32 v[10:11], v14, v[10:11]
 ; GFX1250-NEXT:    v_min_u32_e32 v4, 1, v4
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_min_u32_e32 v8, 1, v8
-; GFX1250-NEXT:    v_min_u32_e32 v10, 1, v10
 ; GFX1250-NEXT:    v_min_u32_e32 v6, 1, v6
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1250-NEXT:    v_dual_sub_nc_u32 v13, 32, v13 :: v_dual_bitop2_b32 v4, v5, v4 bitop3:0x54
+; GFX1250-NEXT:    v_sub_nc_u32_e32 v13, 32, v13
+; GFX1250-NEXT:    v_min_u32_e32 v10, 1, v10
+; GFX1250-NEXT:    v_or_b32_e32 v4, v5, v4
 ; GFX1250-NEXT:    v_or_b32_e32 v5, v9, v8
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_dual_sub_nc_u32 v8, 32, v16 :: v_dual_bitop2_b32 v6, v7, v6 bitop3:0x54
-; GFX1250-NEXT:    v_or_b32_e32 v7, v11, v10
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX1250-NEXT:    v_dual_sub_nc_u32 v12, 32, v12 :: v_dual_bitop2_b32 v7, v11, v10 bitop3:0x54
+; GFX1250-NEXT:    v_sub_nc_u32_e32 v14, 32, v14
 ; GFX1250-NEXT:    v_cvt_f32_i32_e32 v4, v4
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_cvt_f32_i32_e32 v5, v5
 ; GFX1250-NEXT:    v_cvt_f32_i32_e32 v6, v6
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_cvt_f32_i32_e32 v7, v7
-; GFX1250-NEXT:    v_ldexp_f32 v4, v4, v12
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_ldexp_f32 v4, v4, v12
 ; GFX1250-NEXT:    v_ldexp_f32 v5, v5, v13
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_ldexp_f32 v6, v6, v8
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_ldexp_f32 v7, v7, v14
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cvt_pk_f16_f32 v5, v5, v6
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cvt_pk_f16_f32 v4, v4, v7
 ; GFX1250-NEXT:    global_store_b64 v[0:1], v[4:5], off
 ; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
@@ -1458,34 +1457,34 @@ define void @uitofp_i56_to_half(ptr addrspace(1) noalias %out, ptr addrspace(1)
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[10:11], v14, v[10:11]
 ; GFX1250-NEXT:    v_min_u32_e32 v15, 32, v15
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[8:9], v12, v[8:9]
-; GFX1250-NEXT:    v_sub_nc_u32_e32 v12, 32, v12
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[6:7], v13, v[6:7]
-; GFX1250-NEXT:    v_min_u32_e32 v10, 1, v10
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1250-NEXT:    v_lshlrev_b64_e32 v[4:5], v15, v[4:5]
 ; GFX1250-NEXT:    v_min_u32_e32 v8, 1, v8
-; GFX1250-NEXT:    v_sub_nc_u32_e32 v13, 32, v13
+; GFX1250-NEXT:    v_min_u32_e32 v10, 1, v10
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX1250-NEXT:    v_min_u32_e32 v6, 1, v6
-; GFX1250-NEXT:    v_sub_nc_u32_e32 v15, 32, v15
+; GFX1250-NEXT:    v_sub_nc_u32_e32 v13, 32, v13
 ; GFX1250-NEXT:    v_min_u32_e32 v4, 1, v4
 ; GFX1250-NEXT:    v_or_b32_e32 v8, v9, v8
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_dual_sub_nc_u32 v12, 32, v12 :: v_dual_sub_nc_u32 v15, 32, v15
 ; GFX1250-NEXT:    v_or_b32_e32 v6, v7, v6
 ; GFX1250-NEXT:    v_or_b32_e32 v7, v11, v10
 ; GFX1250-NEXT:    v_dual_sub_nc_u32 v5, 32, v14 :: v_dual_bitop2_b32 v4, v5, v4 bitop3:0x54
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1250-NEXT:    v_cvt_f32_u32_e32 v8, v8
-; GFX1250-NEXT:    v_cvt_f32_u32_e32 v6, v6
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_cvt_f32_u32_e32 v6, v6
 ; GFX1250-NEXT:    v_cvt_f32_u32_e32 v7, v7
-; GFX1250-NEXT:    v_cvt_f32_u32_e32 v4, v4
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_cvt_f32_u32_e32 v4, v4
 ; GFX1250-NEXT:    v_ldexp_f32 v8, v8, v12
-; GFX1250-NEXT:    v_ldexp_f32 v6, v6, v13
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1250-NEXT:    v_ldexp_f32 v6, v6, v13
 ; GFX1250-NEXT:    v_ldexp_f32 v5, v7, v5
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_ldexp_f32 v4, v4, v15
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cvt_pk_f16_f32 v5, v6, v5
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cvt_pk_f16_f32 v4, v8, v4
 ; GFX1250-NEXT:    global_store_b64 v[0:1], v[4:5], off
 ; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fptrunc_bf16.ll b/llvm/test/CodeGen/AMDGPU/strict_fptrunc_bf16.ll
index 0eaa01d2685620..cba21eb769a8ac 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fptrunc_bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fptrunc_bf16.ll
@@ -128,25 +128,25 @@ define amdgpu_ps void @strict_fptrunc_v2f64_to_v2bf16(<2 x double> %a, ptr %out)
 ; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[6:7], v10
 ; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[8:9], v11
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[2:3]|, |v[6:7]|
 ; GFX1250-NEXT:    v_cmp_nlg_f64_e32 vcc_lo, v[2:3], v[6:7]
-; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, v[0:1], v[8:9]
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v2, -1, 1, s1
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[0:1]|, |v[8:9]|
-; GFX1250-NEXT:    v_dual_add_nc_u32 v1, v10, v2 :: v_dual_bitop2_b32 v13, 1, v11 bitop3:0x40
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1250-NEXT:    v_cmp_ne_u32_e64 s2, 0, v13
+; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, v[0:1], v[8:9]
 ; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s1
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-NEXT:    v_dual_add_nc_u32 v0, v11, v0 :: v_dual_bitop2_b32 v12, 1, v10 bitop3:0x40
+; GFX1250-NEXT:    v_and_b32_e32 v12, 1, v10
+; GFX1250-NEXT:    v_dual_add_nc_u32 v1, v10, v2 :: v_dual_bitop2_b32 v13, 1, v11 bitop3:0x40
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_add_nc_u32_e32 v0, v11, v0
 ; GFX1250-NEXT:    v_cmp_ne_u32_e64 s1, 0, v12
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cmp_ne_u32_e64 s2, 0, v13
 ; GFX1250-NEXT:    s_or_b32 vcc_lo, s1, vcc_lo
 ; GFX1250-NEXT:    v_cndmask_b32_e32 v1, v1, v10, vcc_lo
 ; GFX1250-NEXT:    s_or_b32 vcc_lo, s2, s0
 ; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v11, vcc_lo
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
 ; GFX1250-NEXT:    flat_store_b32 v[4:5], v0
 ; GFX1250-NEXT:    s_endpgm

>From 0f14737f4efa2910f7fae109fe9c9cdd37790500 Mon Sep 17 00:00:00 2001
From: Jeffrey Byrnes <Jeffrey.Byrnes at amd.com>
Date: Fri, 6 Nov 2026 16:54:42 -0800
Subject: [PATCH 2/2] Formatting

Change-Id: I18ed65c34f9afe8c590d321858efd87ea73a11df
---
 llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
index ccc34018fdc52b..8887a2f3fc668f 100644
--- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
@@ -279,7 +279,8 @@ void GCNHazardRecognizer::updateTRANSState(const MachineInstr &MI) {
 void GCNHazardRecognizer::updateMultiCycleVALUState(const MachineInstr &MI) {
   if (!hasCoExecWindowModel())
     return;
-  // Multi-cycle VALU (CVT, etc.) blocks subsequent VALU for BlockingCycles cycles.
+  // Multi-cycle VALU (CVT, etc.) blocks subsequent VALU for BlockingCycles
+  // cycles.
   if (!SIInstrInfo::isVALU(MI, /*AllowLDSDMA=*/true))
     return;
 
@@ -294,7 +295,8 @@ void GCNHazardRecognizer::updateMultiCycleVALUState(const MachineInstr &MI) {
     // hazard check (same convention as CyclesUntilTRANS), so to expose
     // BlockingCycles-1 cycles of shadow we must seed with BlockingCycles.
     CyclesUntilVALU = BlockingCycles;
-    LLVM_DEBUG(dbgs() << "    Multi-cycle VALU: BlockingCycles=" << BlockingCycles
+    LLVM_DEBUG(dbgs() << "    Multi-cycle VALU: BlockingCycles="
+                      << BlockingCycles
                       << ", CyclesUntilVALU=" << CyclesUntilVALU << "\n");
   }
 }



More information about the llvm-branch-commits mailing list