[llvm] 9d7e1d9 - [AMDGPU][True16] added Pre-RA hint to improve copy elimination (#103366)

via llvm-commits llvm-commits at lists.llvm.org
Wed Mar 12 13:13:02 PDT 2025


Author: Brox Chen
Date: 2025-03-12T16:12:58-04:00
New Revision: 9d7e1d92dbef5aa4d11deed360685a86935953d2

URL: https://github.com/llvm/llvm-project/commit/9d7e1d92dbef5aa4d11deed360685a86935953d2
DIFF: https://github.com/llvm/llvm-project/commit/9d7e1d92dbef5aa4d11deed360685a86935953d2.diff

LOG: [AMDGPU][True16] added Pre-RA hint to improve copy elimination (#103366)

The allocation order of 16 bit registers is vgpr0lo16, vgpr0hi16,
vgpr1lo16, vgpr1hi16, vgpr2lo16.... We prefer (essentially require) that
allocation order, because it uses the minimum number of registers. But
when you have 16 bit data passing between 16 and 32 bit instructions you
get lots of COPY.

This patch teach the compiler that a COPY of a 16-bit value from a 32
bit register to a lo-half 16 bit register is free, to a hi-half 16 bit
register is not.

This might get improved to coalescing with additional cases, and perhaps
as an alternative to the RA hints. For now upstreaming this solution
first.

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp
    llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
    llvm/lib/Target/AMDGPU/SIRegisterInfo.h
    llvm/test/CodeGen/AMDGPU/bf16.ll
    llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll
    llvm/test/CodeGen/AMDGPU/fadd.f16.ll
    llvm/test/CodeGen/AMDGPU/fcmp.f16.ll
    llvm/test/CodeGen/AMDGPU/fma.f16.ll
    llvm/test/CodeGen/AMDGPU/fmul.f16.ll
    llvm/test/CodeGen/AMDGPU/fpext.f16.ll
    llvm/test/CodeGen/AMDGPU/fptosi.f16.ll
    llvm/test/CodeGen/AMDGPU/fptoui.f16.ll
    llvm/test/CodeGen/AMDGPU/fshr.ll
    llvm/test/CodeGen/AMDGPU/i1-to-bf16.ll
    llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll
    llvm/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll
    llvm/test/CodeGen/AMDGPU/mad-mix.ll
    llvm/test/CodeGen/AMDGPU/mad.u16.ll
    llvm/test/CodeGen/AMDGPU/minimummaximum.ll
    llvm/test/CodeGen/AMDGPU/minmax.ll
    llvm/test/CodeGen/AMDGPU/mul.i16.ll
    llvm/test/CodeGen/AMDGPU/preserve-hi16.ll
    llvm/test/CodeGen/AMDGPU/saddsat.ll
    llvm/test/CodeGen/AMDGPU/ssubsat.ll
    llvm/test/CodeGen/AMDGPU/uaddsat.ll
    llvm/test/CodeGen/AMDGPU/usubsat.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp b/llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp
index 0f008f70a6c3d..1a00b5a846b2f 100644
--- a/llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp
@@ -22,12 +22,22 @@
 /// although the same shall be possible with other register classes and
 /// instructions if necessary.
 ///
+/// This pass also adds register allocation hints to COPY.
+/// The hints will be post-processed by SIRegisterInfo::getRegAllocationHints.
+/// When using True16, we often see COPY moving a 16-bit value between a VGPR_32
+/// This pass also adds register allocation hints to COPY.
+/// The hints will be post-processed by SIRegisterInfo::getRegAllocationHints.
+/// When using True16, we often see COPY moving a 16-bit value between a VGPR_32
+/// and a VGPR_16. If we use the VGPR_16 that corresponds to the lo16 bits of
+/// the VGPR_32, the COPY can be completely eliminated.
+///
 //===----------------------------------------------------------------------===//
 
 #include "GCNPreRAOptimizations.h"
 #include "AMDGPU.h"
 #include "GCNSubtarget.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "SIRegisterInfo.h"
 #include "llvm/CodeGen/LiveIntervals.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/InitializePasses.h"
@@ -253,5 +263,38 @@ bool GCNPreRAOptimizationsImpl::run(MachineFunction &MF) {
     Changed |= processReg(Reg);
   }
 
+  if (!ST.useRealTrue16Insts())
+    return Changed;
+
+  // Add RA hints to improve True16 COPY elimination.
+  for (const MachineBasicBlock &MBB : MF) {
+    for (const MachineInstr &MI : MBB) {
+      if (MI.getOpcode() != AMDGPU::COPY)
+        continue;
+      Register Dst = MI.getOperand(0).getReg();
+      Register Src = MI.getOperand(1).getReg();
+      if (Dst.isVirtual() &&
+          MRI->getRegClass(Dst) == &AMDGPU::VGPR_16RegClass &&
+          Src.isPhysical() &&
+          TRI->getRegClassForReg(*MRI, Src) == &AMDGPU::VGPR_32RegClass)
+        MRI->setRegAllocationHint(Dst, 0, TRI->getSubReg(Src, AMDGPU::lo16));
+      if (Src.isVirtual() &&
+          MRI->getRegClass(Src) == &AMDGPU::VGPR_16RegClass &&
+          Dst.isPhysical() &&
+          TRI->getRegClassForReg(*MRI, Dst) == &AMDGPU::VGPR_32RegClass)
+        MRI->setRegAllocationHint(Src, 0, TRI->getSubReg(Dst, AMDGPU::lo16));
+      if (!Dst.isVirtual() || !Src.isVirtual())
+        continue;
+      if (MRI->getRegClass(Dst) == &AMDGPU::VGPR_32RegClass &&
+          MRI->getRegClass(Src) == &AMDGPU::VGPR_16RegClass) {
+        MRI->setRegAllocationHint(Dst, AMDGPURI::Size32, Src);
+        MRI->setRegAllocationHint(Src, AMDGPURI::Size16, Dst);
+      }
+      if (MRI->getRegClass(Dst) == &AMDGPU::VGPR_16RegClass &&
+          MRI->getRegClass(Src) == &AMDGPU::VGPR_32RegClass)
+        MRI->setRegAllocationHint(Dst, AMDGPURI::Size16, Src);
+    }
+  }
+
   return Changed;
 }

diff  --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
index adadf8e4e4e65..12b451ece3b96 100644
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -3713,6 +3713,73 @@ const int *SIRegisterInfo::getRegUnitPressureSets(unsigned RegUnit) const {
   return AMDGPUGenRegisterInfo::getRegUnitPressureSets(RegUnit);
 }
 
+bool SIRegisterInfo::getRegAllocationHints(Register VirtReg,
+                                           ArrayRef<MCPhysReg> Order,
+                                           SmallVectorImpl<MCPhysReg> &Hints,
+                                           const MachineFunction &MF,
+                                           const VirtRegMap *VRM,
+                                           const LiveRegMatrix *Matrix) const {
+
+  const MachineRegisterInfo &MRI = MF.getRegInfo();
+  const SIRegisterInfo *TRI = ST.getRegisterInfo();
+
+  std::pair<unsigned, Register> Hint = MRI.getRegAllocationHint(VirtReg);
+
+  switch (Hint.first) {
+  case AMDGPURI::Size32: {
+    Register Paired = Hint.second;
+    assert(Paired);
+    Register PairedPhys;
+    if (Paired.isPhysical()) {
+      PairedPhys =
+          getMatchingSuperReg(Paired, AMDGPU::lo16, &AMDGPU::VGPR_32RegClass);
+    } else if (VRM && VRM->hasPhys(Paired)) {
+      PairedPhys = getMatchingSuperReg(VRM->getPhys(Paired), AMDGPU::lo16,
+                                       &AMDGPU::VGPR_32RegClass);
+    }
+
+    // Prefer the paired physreg.
+    if (PairedPhys)
+      // isLo(Paired) is implicitly true here from the API of
+      // getMatchingSuperReg.
+      Hints.push_back(PairedPhys);
+    return false;
+  }
+  case AMDGPURI::Size16: {
+    Register Paired = Hint.second;
+    assert(Paired);
+    Register PairedPhys;
+    if (Paired.isPhysical()) {
+      PairedPhys = TRI->getSubReg(Paired, AMDGPU::lo16);
+    } else if (VRM && VRM->hasPhys(Paired)) {
+      PairedPhys = TRI->getSubReg(VRM->getPhys(Paired), AMDGPU::lo16);
+    }
+
+    // First prefer the paired physreg.
+    if (PairedPhys)
+      Hints.push_back(PairedPhys);
+    else {
+      // Add all the lo16 physregs.
+      // When the Paired operand has not yet been assigned a physreg it is
+      // better to try putting VirtReg in a lo16 register, because possibly
+      // later Paired can be assigned to the overlapping register and the COPY
+      // can be eliminated.
+      for (MCPhysReg PhysReg : Order) {
+        if (PhysReg == PairedPhys || AMDGPU::isHi16Reg(PhysReg, *this))
+          continue;
+        if (AMDGPU::VGPR_16RegClass.contains(PhysReg) &&
+            !MRI.isReserved(PhysReg))
+          Hints.push_back(PhysReg);
+      }
+    }
+    return false;
+  }
+  default:
+    return TargetRegisterInfo::getRegAllocationHints(VirtReg, Order, Hints, MF,
+                                                     VRM);
+  }
+}
+
 MCRegister SIRegisterInfo::getReturnAddressReg(const MachineFunction &MF) const {
   // Not a callee saved register.
   return AMDGPU::SGPR30_SGPR31;

diff  --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.h b/llvm/lib/Target/AMDGPU/SIRegisterInfo.h
index a64180daea2ad..f3068963fd10f 100644
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.h
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.h
@@ -29,6 +29,13 @@ class LiveRegUnits;
 class RegisterBank;
 struct SGPRSpillBuilder;
 
+/// Register allocation hint types. Helps eliminate unneeded COPY with True16
+namespace AMDGPURI {
+
+enum { Size16 = 1, Size32 = 2 };
+
+} // end namespace AMDGPURI
+
 class SIRegisterInfo final : public AMDGPUGenRegisterInfo {
 private:
   const GCNSubtarget &ST;
@@ -329,6 +336,11 @@ class SIRegisterInfo final : public AMDGPUGenRegisterInfo {
   unsigned getRegPressureSetLimit(const MachineFunction &MF,
                                   unsigned Idx) const override;
 
+  bool getRegAllocationHints(Register VirtReg, ArrayRef<MCPhysReg> Order,
+                             SmallVectorImpl<MCPhysReg> &Hints,
+                             const MachineFunction &MF, const VirtRegMap *VRM,
+                             const LiveRegMatrix *Matrix) const override;
+
   const int *getRegUnitPressureSets(unsigned RegUnit) const override;
 
   MCRegister getReturnAddressReg(const MachineFunction &MF) const;

diff  --git a/llvm/test/CodeGen/AMDGPU/bf16.ll b/llvm/test/CodeGen/AMDGPU/bf16.ll
index 91598496eb984..669444e36079d 100644
--- a/llvm/test/CodeGen/AMDGPU/bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/bf16.ll
@@ -37712,12 +37712,10 @@ define bfloat @v_select_bf16(i1 %cond, bfloat %a, bfloat %b) {
 ; GFX11TRUE16-LABEL: v_select_bf16:
 ; GFX11TRUE16:       ; %bb.0:
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11TRUE16-NEXT:    v_and_b32_e32 v3, 1, v0
-; GFX11TRUE16-NEXT:    v_mov_b16_e32 v0.l, v2.l
-; GFX11TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v3
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v0.l, v0.l, v0.h, vcc_lo
+; GFX11TRUE16-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX11TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v0.l, v2.l, v1.l, vcc_lo
 ; GFX11TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11FAKE16-LABEL: v_select_bf16:
@@ -37785,14 +37783,11 @@ define bfloat @v_select_fneg_lhs_bf16(i1 %cond, bfloat %a, bfloat %b) {
 ; GFX11TRUE16-LABEL: v_select_fneg_lhs_bf16:
 ; GFX11TRUE16:       ; %bb.0:
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11TRUE16-NEXT:    v_and_b32_e32 v3, 1, v0
-; GFX11TRUE16-NEXT:    v_mov_b16_e32 v0.l, v1.l
-; GFX11TRUE16-NEXT:    v_mov_b16_e32 v0.h, v2.l
-; GFX11TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v3
-; GFX11TRUE16-NEXT:    v_xor_b16 v0.l, 0x8000, v0.l
-; GFX11TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v0.l, v0.h, v0.l, vcc_lo
+; GFX11TRUE16-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX11TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX11TRUE16-NEXT:    v_xor_b16 v0.l, 0x8000, v1.l
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo
 ; GFX11TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11FAKE16-LABEL: v_select_fneg_lhs_bf16:
@@ -37862,14 +37857,11 @@ define bfloat @v_select_fneg_rhs_bf16(i1 %cond, bfloat %a, bfloat %b) {
 ; GFX11TRUE16-LABEL: v_select_fneg_rhs_bf16:
 ; GFX11TRUE16:       ; %bb.0:
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11TRUE16-NEXT:    v_and_b32_e32 v3, 1, v0
-; GFX11TRUE16-NEXT:    v_mov_b16_e32 v0.l, v2.l
-; GFX11TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v3
-; GFX11TRUE16-NEXT:    v_xor_b16 v0.l, 0x8000, v0.l
-; GFX11TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v0.l, v0.l, v0.h, vcc_lo
+; GFX11TRUE16-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX11TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX11TRUE16-NEXT:    v_xor_b16 v0.l, 0x8000, v2.l
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v0.l, v0.l, v1.l, vcc_lo
 ; GFX11TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11FAKE16-LABEL: v_select_fneg_rhs_bf16:
@@ -42659,9 +42651,9 @@ define <32 x bfloat> @v_vselect_v32bf16(<32 x i1> %cond, <32 x bfloat> %a, <32 x
 ; GFX11TRUE16-NEXT:    scratch_load_b32 v85, off, s32 offset:72
 ; GFX11TRUE16-NEXT:    scratch_load_b32 v86, off, s32 offset:4
 ; GFX11TRUE16-NEXT:    scratch_load_b32 v87, off, s32 offset:68
-; GFX11TRUE16-NEXT:    v_and_b32_e32 v16, 1, v16
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v0, 1, v0
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v14, 1, v14
+; GFX11TRUE16-NEXT:    v_and_b32_e32 v16, 1, v16
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v18, 1, v18
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v20, 1, v20
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v22, 1, v22
@@ -42669,7 +42661,6 @@ define <32 x bfloat> @v_vselect_v32bf16(<32 x i1> %cond, <32 x bfloat> %a, <32 x
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v26, 1, v26
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v28, 1, v28
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v30, 1, v30
-; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e64 s15, 1, v16
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v1, 1, v1
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v2, 1, v2
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v3, 1, v3
@@ -42693,6 +42684,7 @@ define <32 x bfloat> @v_vselect_v32bf16(<32 x i1> %cond, <32 x bfloat> %a, <32 x
 ; GFX11TRUE16-NEXT:    v_and_b32_e32 v29, 1, v29
 ; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v0
 ; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e64 s13, 1, v14
+; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e64 s15, 1, v16
 ; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e64 s17, 1, v18
 ; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e64 s19, 1, v20
 ; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e64 s21, 1, v22
@@ -42722,45 +42714,44 @@ define <32 x bfloat> @v_vselect_v32bf16(<32 x i1> %cond, <32 x bfloat> %a, <32 x
 ; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e64 s26, 1, v27
 ; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e64 s29, 1, v29
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(32)
-; GFX11TRUE16-NEXT:    v_mov_b16_e32 v16.l, v31.l
+; GFX11TRUE16-NEXT:    v_and_b32_e32 v31, 1, v31
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(31)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v17, 16, v32
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v16, 16, v32
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(30)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v18, 16, v33
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v17, 16, v33
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v15.l, v33.l, v32.l, s28
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(29)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v19, 16, v34
-; GFX11TRUE16-NEXT:    v_and_b32_e32 v16, 1, v16
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v18, 16, v34
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(28)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v20, 16, v35
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v19, 16, v35
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v14.l, v35.l, v34.l, s27
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(27)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v21, 16, v36
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v20, 16, v36
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(26)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v22, 16, v37
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v21, 16, v37
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v13.l, v37.l, v36.l, s25
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(25)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v23, 16, v38
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v22, 16, v38
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(24)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v24, 16, v39
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v23, 16, v39
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v12.l, v39.l, v38.l, s23
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(23)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v25, 16, v48
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v24, 16, v48
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(22)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v26, 16, v49
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v25, 16, v49
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v11.l, v49.l, v48.l, s21
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(21)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v27, 16, v50
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v26, 16, v50
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(20)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v28, 16, v51
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v27, 16, v51
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v10.l, v51.l, v50.l, s19
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(19)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v29, 16, v52
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v28, 16, v52
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(18)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v30, 16, v53
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v29, 16, v53
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v9.l, v53.l, v52.l, s17
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(17)
-; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v31, 16, v54
+; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v30, 16, v54
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(16)
 ; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v32, 16, v55
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v8.l, v55.l, v54.l, s15
@@ -42798,20 +42789,20 @@ define <32 x bfloat> @v_vselect_v32bf16(<32 x i1> %cond, <32 x bfloat> %a, <32 x
 ; GFX11TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11TRUE16-NEXT:    v_lshrrev_b32_e32 v64, 16, v87
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v0.l, v87.l, v86.l, vcc_lo
-; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v16
+; GFX11TRUE16-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v31
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v6.l, v67.l, v66.l, s11
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v5.l, v69.l, v68.l, s9
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v4.l, v71.l, v70.l, s7
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v3.l, v81.l, v80.l, s5
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v2.l, v83.l, v82.l, s3
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v1.l, v85.l, v84.l, s1
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v14.h, v20.l, v19.l, s29
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v13.h, v22.l, v21.l, s26
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v12.h, v24.l, v23.l, s24
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v11.h, v26.l, v25.l, s22
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v10.h, v28.l, v27.l, s20
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v9.h, v30.l, v29.l, s18
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v8.h, v32.l, v31.l, s16
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v14.h, v19.l, v18.l, s29
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v13.h, v21.l, v20.l, s26
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v12.h, v23.l, v22.l, s24
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v11.h, v25.l, v24.l, s22
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v10.h, v27.l, v26.l, s20
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v9.h, v29.l, v28.l, s18
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v8.h, v32.l, v30.l, s16
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v7.h, v34.l, v33.l, s14
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v6.h, v36.l, v35.l, s12
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v5.h, v38.l, v37.l, s10
@@ -42820,7 +42811,7 @@ define <32 x bfloat> @v_vselect_v32bf16(<32 x i1> %cond, <32 x bfloat> %a, <32 x
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v1.h, v54.l, v53.l, s2
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v2.h, v52.l, v51.l, s4
 ; GFX11TRUE16-NEXT:    v_cndmask_b16 v3.h, v50.l, v49.l, s6
-; GFX11TRUE16-NEXT:    v_cndmask_b16 v15.h, v18.l, v17.l, vcc_lo
+; GFX11TRUE16-NEXT:    v_cndmask_b16 v15.h, v17.l, v16.l, vcc_lo
 ; GFX11TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11FAKE16-LABEL: v_vselect_v32bf16:

diff  --git a/llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll b/llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll
index 55f62058ec7af..806fe899a9149 100644
--- a/llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll
+++ b/llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll
@@ -908,10 +908,9 @@ define <2 x i16> @chain_hi_to_lo_global_other_dep(ptr addrspace(1) %ptr) {
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    global_load_d16_hi_b16 v0, v[0:1], off glc dlc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
 ; GFX11-TRUE16-NEXT:    v_pk_add_u16 v0, v0, 12 op_sel_hi:[1,0]
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v0, 0xffff, v1, v0
+; GFX11-TRUE16-NEXT:    v_bfi_b32 v0, 0xffff, v2, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: chain_hi_to_lo_global_other_dep:
@@ -981,12 +980,11 @@ define <2 x i16> @chain_hi_to_lo_flat_other_dep(ptr addrspace(0) %ptr) {
 ; GFX11-TRUE16-NEXT:    flat_load_d16_b16 v2, v[0:1] offset:2 glc dlc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    flat_load_d16_hi_b16 v0, v[0:1] glc dlc
-; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(1)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-TRUE16-NEXT:    v_pk_add_u16 v0, v0, 12 op_sel_hi:[1,0]
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v0, 0xffff, v1, v0
+; GFX11-TRUE16-NEXT:    v_bfi_b32 v0, 0xffff, v2, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: chain_hi_to_lo_flat_other_dep:

diff  --git a/llvm/test/CodeGen/AMDGPU/fadd.f16.ll b/llvm/test/CodeGen/AMDGPU/fadd.f16.ll
index e1ecd34ad24e0..fc3624cdfe118 100644
--- a/llvm/test/CodeGen/AMDGPU/fadd.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fadd.f16.ll
@@ -76,9 +76,7 @@ define amdgpu_kernel void @fadd_f16(
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-SDAG-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-SDAG-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-SDAG-NEXT:    v_add_f16_e32 v0.l, v0.l, v0.h
+; GFX11-SDAG-NEXT:    v_add_f16_e32 v0.l, v0.l, v1.l
 ; GFX11-SDAG-NEXT:    buffer_store_b16 v0, off, s[8:11], 0
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
@@ -98,9 +96,7 @@ define amdgpu_kernel void @fadd_f16(
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-GISEL-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-GISEL-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-GISEL-NEXT:    v_add_f16_e32 v0.l, v0.l, v0.h
+; GFX11-GISEL-NEXT:    v_add_f16_e32 v0.l, v0.l, v1.l
 ; GFX11-GISEL-NEXT:    buffer_store_b16 v0, off, s[0:3], 0
 ; GFX11-GISEL-NEXT:    s_endpgm
 ;

diff  --git a/llvm/test/CodeGen/AMDGPU/fcmp.f16.ll b/llvm/test/CodeGen/AMDGPU/fcmp.f16.ll
index 9b98bd6c9ab03..3156a1280afea 100644
--- a/llvm/test/CodeGen/AMDGPU/fcmp.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fcmp.f16.ll
@@ -76,9 +76,7 @@ define amdgpu_kernel void @fcmp_f16_lt(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -215,9 +213,8 @@ define amdgpu_kernel void @fcmp_f16_lt_abs(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_lt_f16_e64 s2, |v0.l|, |v0.h|
+; GFX11-TRUE16-NEXT:    v_cmp_lt_f16_e64 s2, |v0.l|, |v1.l|
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, s2
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -360,9 +357,7 @@ define amdgpu_kernel void @fcmp_f16_eq(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -499,9 +494,7 @@ define amdgpu_kernel void @fcmp_f16_le(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_le_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_le_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -638,9 +631,7 @@ define amdgpu_kernel void @fcmp_f16_gt(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -777,9 +768,7 @@ define amdgpu_kernel void @fcmp_f16_lg(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -916,9 +905,7 @@ define amdgpu_kernel void @fcmp_f16_ge(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -1055,9 +1042,7 @@ define amdgpu_kernel void @fcmp_f16_o(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_o_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_o_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -1194,9 +1179,7 @@ define amdgpu_kernel void @fcmp_f16_u(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_u_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_u_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -1333,9 +1316,7 @@ define amdgpu_kernel void @fcmp_f16_nge(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -1472,9 +1453,7 @@ define amdgpu_kernel void @fcmp_f16_nlg(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -1611,9 +1590,7 @@ define amdgpu_kernel void @fcmp_f16_ngt(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -1750,9 +1727,7 @@ define amdgpu_kernel void @fcmp_f16_nle(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -1889,9 +1864,7 @@ define amdgpu_kernel void @fcmp_f16_neq(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
@@ -2028,9 +2001,7 @@ define amdgpu_kernel void @fcmp_f16_nlt(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm

diff  --git a/llvm/test/CodeGen/AMDGPU/fma.f16.ll b/llvm/test/CodeGen/AMDGPU/fma.f16.ll
index eefcf56586033..8eda0072a7f4d 100644
--- a/llvm/test/CodeGen/AMDGPU/fma.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fma.f16.ll
@@ -29,10 +29,7 @@ define half @test_fma(half %x, half %y, half %z) {
 ; GFX11-SDAG-TRUE16-LABEL: test_fma:
 ; GFX11-SDAG-TRUE16:       ; %bb.0:
 ; GFX11-SDAG-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_fma_f16 v0.l, v0.l, v0.h, v1.l
+; GFX11-SDAG-TRUE16-NEXT:    v_fma_f16 v0.l, v0.l, v1.l, v2.l
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_fma:
@@ -85,9 +82,7 @@ define half @test_fmac(half %x, half %y, half %z) {
 ; GFX11-SDAG-TRUE16-LABEL: test_fmac:
 ; GFX11-SDAG-TRUE16:       ; %bb.0:
 ; GFX11-SDAG-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v1.h, v2.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_fmac_f16_e32 v0.l, v1.l, v1.h
+; GFX11-SDAG-TRUE16-NEXT:    v_fmac_f16_e32 v0.l, v1.l, v2.l
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_fmac:
@@ -146,9 +141,7 @@ define half @test_fmaak(half %x, half %y, half %z) {
 ; GFX11-SDAG-TRUE16-LABEL: test_fmaak:
 ; GFX11-SDAG-TRUE16:       ; %bb.0:
 ; GFX11-SDAG-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_fmaak_f16 v0.l, v0.l, v0.h, 0x4200
+; GFX11-SDAG-TRUE16-NEXT:    v_fmaak_f16 v0.l, v0.l, v1.l, 0x4200
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_fmaak:
@@ -207,9 +200,7 @@ define half @test_fmamk(half %x, half %y, half %z) {
 ; GFX11-SDAG-TRUE16-LABEL: test_fmamk:
 ; GFX11-SDAG-TRUE16:       ; %bb.0:
 ; GFX11-SDAG-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v2.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_fmamk_f16 v0.l, v0.l, 0x4200, v0.h
+; GFX11-SDAG-TRUE16-NEXT:    v_fmamk_f16 v0.l, v0.l, 0x4200, v2.l
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_fmamk:

diff  --git a/llvm/test/CodeGen/AMDGPU/fmul.f16.ll b/llvm/test/CodeGen/AMDGPU/fmul.f16.ll
index e9877ae5144f5..c16fa2d40097d 100644
--- a/llvm/test/CodeGen/AMDGPU/fmul.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmul.f16.ll
@@ -75,9 +75,7 @@ define amdgpu_kernel void @fmul_f16(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_mul_f16_e32 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_mul_f16_e32 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    buffer_store_b16 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;

diff  --git a/llvm/test/CodeGen/AMDGPU/fpext.f16.ll b/llvm/test/CodeGen/AMDGPU/fpext.f16.ll
index 79b5eca070302..f9ec35034c18b 100644
--- a/llvm/test/CodeGen/AMDGPU/fpext.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fpext.f16.ll
@@ -830,13 +830,11 @@ define amdgpu_kernel void @fneg_multi_foldable_use_fpext_f16_to_f32(
 ; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s5, s1
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mul_f16_e64 v0.h, -v0.l, v0.l
-; GFX11-TRUE16-NEXT:    v_cvt_f32_f16_e64 v1, -v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    buffer_store_b32 v1, off, s[4:7], 0 dlc
+; GFX11-TRUE16-NEXT:    v_mul_f16_e64 v1.l, -v0.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cvt_f32_f16_e64 v0, -v0.l
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0 dlc
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-TRUE16-NEXT:    buffer_store_b16 v0, off, s[4:7], 0 dlc
+; GFX11-TRUE16-NEXT:    buffer_store_b16 v1, off, s[4:7], 0 dlc
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;
@@ -1035,13 +1033,11 @@ define amdgpu_kernel void @fabs_multi_foldable_use_fpext_f16_to_f32(
 ; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s5, s1
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mul_f16_e64 v0.h, |v0.l|, v0.l
-; GFX11-TRUE16-NEXT:    v_cvt_f32_f16_e64 v1, |v0.l|
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    buffer_store_b32 v1, off, s[4:7], 0 dlc
+; GFX11-TRUE16-NEXT:    v_mul_f16_e64 v1.l, |v0.l|, v0.l
+; GFX11-TRUE16-NEXT:    v_cvt_f32_f16_e64 v0, |v0.l|
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0 dlc
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-TRUE16-NEXT:    buffer_store_b16 v0, off, s[4:7], 0 dlc
+; GFX11-TRUE16-NEXT:    buffer_store_b16 v1, off, s[4:7], 0 dlc
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;
@@ -1241,13 +1237,11 @@ define amdgpu_kernel void @fabs_fneg_multi_foldable_use_fpext_f16_to_f32(
 ; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s5, s1
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mul_f16_e64 v0.h, -|v0.l|, v0.l
-; GFX11-TRUE16-NEXT:    v_cvt_f32_f16_e64 v1, -|v0.l|
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    buffer_store_b32 v1, off, s[4:7], 0 dlc
+; GFX11-TRUE16-NEXT:    v_mul_f16_e64 v1.l, -|v0.l|, v0.l
+; GFX11-TRUE16-NEXT:    v_cvt_f32_f16_e64 v0, -|v0.l|
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0 dlc
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-TRUE16-NEXT:    buffer_store_b16 v0, off, s[4:7], 0 dlc
+; GFX11-TRUE16-NEXT:    buffer_store_b16 v1, off, s[4:7], 0 dlc
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;

diff  --git a/llvm/test/CodeGen/AMDGPU/fptosi.f16.ll b/llvm/test/CodeGen/AMDGPU/fptosi.f16.ll
index 407626a8e92d5..f84e14ea62273 100644
--- a/llvm/test/CodeGen/AMDGPU/fptosi.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fptosi.f16.ll
@@ -331,13 +331,10 @@ define amdgpu_kernel void @fptosi_v2f16_to_v2i16(
 ; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; GFX11-TRUE16-NEXT:    v_cvt_i16_f16_e32 v0.l, v0.l
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_cvt_i16_f16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX11-TRUE16-NEXT:    v_cvt_i16_f16_e32 v1.l, v1.l
+; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-TRUE16-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;
@@ -622,7 +619,7 @@ define amdgpu_kernel void @fptosi_f16_to_i1(ptr addrspace(1) %out, half %in) {
 ; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, s2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s2, -1
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, -1.0, v0
+; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, -1.0, v0.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b8 v0, off, s[0:3], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm

diff  --git a/llvm/test/CodeGen/AMDGPU/fptoui.f16.ll b/llvm/test/CodeGen/AMDGPU/fptoui.f16.ll
index ff00633cad492..bba3a23df11a5 100644
--- a/llvm/test/CodeGen/AMDGPU/fptoui.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fptoui.f16.ll
@@ -330,13 +330,10 @@ define amdgpu_kernel void @fptoui_v2f16_to_v2i16(
 ; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; GFX11-TRUE16-NEXT:    v_cvt_u16_f16_e32 v0.l, v0.l
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_cvt_u16_f16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX11-TRUE16-NEXT:    v_cvt_u16_f16_e32 v1.l, v1.l
+; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-TRUE16-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;
@@ -620,7 +617,7 @@ define amdgpu_kernel void @fptoui_f16_to_i1(ptr addrspace(1) %out, half %in) {
 ; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, s2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s2, -1
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, 1.0, v0
+; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, 1.0, v0.l
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc_lo
 ; GFX11-TRUE16-NEXT:    buffer_store_b8 v0, off, s[0:3], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm

diff  --git a/llvm/test/CodeGen/AMDGPU/fshr.ll b/llvm/test/CodeGen/AMDGPU/fshr.ll
index c3d82e38d1985..816c3fccbb237 100644
--- a/llvm/test/CodeGen/AMDGPU/fshr.ll
+++ b/llvm/test/CodeGen/AMDGPU/fshr.ll
@@ -1065,13 +1065,11 @@ define i16 @v_fshr_i16(i16 %src0, i16 %src1, i16 %src2) {
 ; GFX11-TRUE16-LABEL: v_fshr_i16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v2.l
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b16 v0.l, 1, v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_xor_b16 v2.l, v0.h, -1
-; GFX11-TRUE16-NEXT:    v_lshrrev_b16 v0.h, v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_lshlrev_b16 v0.l, v2.l, v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_xor_b16 v0.h, v2.l, -1
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_lshlrev_b16 v0.l, v0.h, v0.l
+; GFX11-TRUE16-NEXT:    v_lshrrev_b16 v0.h, v2.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_or_b16 v0.l, v0.l, v0.h
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -1093,13 +1091,11 @@ define i16 @v_fshr_i16(i16 %src0, i16 %src1, i16 %src2) {
 ; GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v2.l
 ; GFX12-TRUE16-NEXT:    v_lshlrev_b16 v0.l, 1, v0.l
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT:    v_xor_b16 v2.l, v0.h, -1
-; GFX12-TRUE16-NEXT:    v_lshrrev_b16 v0.h, v0.h, v1.l
-; GFX12-TRUE16-NEXT:    v_lshlrev_b16 v0.l, v2.l, v0.l
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12-TRUE16-NEXT:    v_xor_b16 v0.h, v2.l, -1
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX12-TRUE16-NEXT:    v_lshlrev_b16 v0.l, v0.h, v0.l
+; GFX12-TRUE16-NEXT:    v_lshrrev_b16 v0.h, v2.l, v1.l
 ; GFX12-TRUE16-NEXT:    v_or_b16 v0.l, v0.l, v0.h
 ; GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;

diff  --git a/llvm/test/CodeGen/AMDGPU/i1-to-bf16.ll b/llvm/test/CodeGen/AMDGPU/i1-to-bf16.ll
index 59b0ba2469a20..a46d629c02b85 100644
--- a/llvm/test/CodeGen/AMDGPU/i1-to-bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/i1-to-bf16.ll
@@ -108,14 +108,13 @@ define amdgpu_ps i32 @s_uitofp_i1_to_bf16(i1 inreg %num) {
 ; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, 1.0, s0
 ; GFX11-NEXT:    v_readfirstlane_b32 s0, v0
 ; GFX11-NEXT:    v_cmp_u_f32_e32 vcc_lo, v0, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_bfe_u32 s1, s0, 0x10010
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_add_i32 s1, s1, s0
 ; GFX11-NEXT:    s_bitset1_b32 s0, 22
 ; GFX11-NEXT:    s_addk_i32 s1, 0x7fff
 ; GFX11-NEXT:    s_and_b32 s2, vcc_lo, exec_lo
 ; GFX11-NEXT:    s_cselect_b32 s0, s0, s1
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_lshr_b32 s0, s0, 16
 ; GFX11-NEXT:    ; return to shader part epilog
 ;
@@ -126,7 +125,6 @@ define amdgpu_ps i32 @s_uitofp_i1_to_bf16(i1 inreg %num) {
 ; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX12-NEXT:    v_cndmask_b32_e64 v0, 0, 1.0, s0
 ; GFX12-NEXT:    v_readfirstlane_b32 s0, v0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX12-NEXT:    s_bfe_u32 s1, s0, 0x10010
 ; GFX12-NEXT:    s_or_b32 s2, s0, 0x400000
 ; GFX12-NEXT:    s_wait_alu 0xfffe
@@ -307,11 +305,10 @@ define amdgpu_ps <2 x i32> @s_uitofp_v2i1_to_v2bf16(<2 x i1> inreg %num) {
 ; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, 1.0, s0
 ; GFX11-NEXT:    v_readfirstlane_b32 s2, v0
 ; GFX11-NEXT:    v_cmp_u_f32_e64 s1, v0, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    v_readfirstlane_b32 s0, v1
 ; GFX11-NEXT:    v_cmp_u_f32_e32 vcc_lo, v1, v1
 ; GFX11-NEXT:    s_bfe_u32 s3, s0, 0x10010
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_add_i32 s3, s3, s0
 ; GFX11-NEXT:    s_bitset1_b32 s0, 22
 ; GFX11-NEXT:    s_addk_i32 s3, 0x7fff
@@ -341,7 +338,6 @@ define amdgpu_ps <2 x i32> @s_uitofp_v2i1_to_v2bf16(<2 x i1> inreg %num) {
 ; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX12-NEXT:    v_readfirstlane_b32 s2, v0
 ; GFX12-NEXT:    v_readfirstlane_b32 s0, v1
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX12-NEXT:    s_bfe_u32 s1, s0, 0x10010
 ; GFX12-NEXT:    s_or_b32 s3, s0, 0x400000
 ; GFX12-NEXT:    s_wait_alu 0xfffe
@@ -1165,14 +1161,13 @@ define amdgpu_ps i32 @s_sitofp_i1_to_bf16(i1 inreg %num) {
 ; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1.0, s0
 ; GFX11-NEXT:    v_readfirstlane_b32 s0, v0
 ; GFX11-NEXT:    v_cmp_u_f32_e32 vcc_lo, v0, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_bfe_u32 s1, s0, 0x10010
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_add_i32 s1, s1, s0
 ; GFX11-NEXT:    s_bitset1_b32 s0, 22
 ; GFX11-NEXT:    s_addk_i32 s1, 0x7fff
 ; GFX11-NEXT:    s_and_b32 s2, vcc_lo, exec_lo
 ; GFX11-NEXT:    s_cselect_b32 s0, s0, s1
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_ashr_i32 s0, s0, 16
 ; GFX11-NEXT:    ; return to shader part epilog
 ;
@@ -1183,7 +1178,6 @@ define amdgpu_ps i32 @s_sitofp_i1_to_bf16(i1 inreg %num) {
 ; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX12-NEXT:    v_cndmask_b32_e64 v0, 0, -1.0, s0
 ; GFX12-NEXT:    v_readfirstlane_b32 s0, v0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX12-NEXT:    s_bfe_u32 s1, s0, 0x10010
 ; GFX12-NEXT:    s_or_b32 s2, s0, 0x400000
 ; GFX12-NEXT:    s_wait_alu 0xfffe
@@ -1364,11 +1358,10 @@ define amdgpu_ps <2 x i32> @s_sitofp_v2i1_to_v2bf16(<2 x i1> inreg %num) {
 ; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1.0, s0
 ; GFX11-NEXT:    v_readfirstlane_b32 s2, v0
 ; GFX11-NEXT:    v_cmp_u_f32_e64 s0, v0, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    v_readfirstlane_b32 s1, v1
 ; GFX11-NEXT:    v_cmp_u_f32_e32 vcc_lo, v1, v1
 ; GFX11-NEXT:    s_bfe_u32 s3, s1, 0x10010
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_add_i32 s3, s3, s1
 ; GFX11-NEXT:    s_bitset1_b32 s1, 22
 ; GFX11-NEXT:    s_addk_i32 s3, 0x7fff
@@ -1398,7 +1391,6 @@ define amdgpu_ps <2 x i32> @s_sitofp_v2i1_to_v2bf16(<2 x i1> inreg %num) {
 ; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX12-NEXT:    v_readfirstlane_b32 s2, v0
 ; GFX12-NEXT:    v_readfirstlane_b32 s0, v1
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX12-NEXT:    s_bfe_u32 s1, s0, 0x10010
 ; GFX12-NEXT:    s_or_b32 s3, s0, 0x400000
 ; GFX12-NEXT:    s_wait_alu 0xfffe

diff  --git a/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll b/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll
index 0071842baad27..6ba31118322bf 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll
@@ -217,9 +217,8 @@ define half @test_ldexp_f16_i8(half %a, i8 %b) {
 ; GFX11-SDAG-TRUE16:       ; %bb.0:
 ; GFX11-SDAG-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-TRUE16-NEXT:    v_bfe_i32 v1, v1, 0, 8
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v0.h
+; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v1.l
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_ldexp_f16_i8:
@@ -307,9 +306,7 @@ define half @test_ldexp_f16_i16(half %a, i16 %b) {
 ; GFX11-SDAG-TRUE16-LABEL: test_ldexp_f16_i16:
 ; GFX11-SDAG-TRUE16:       ; %bb.0:
 ; GFX11-SDAG-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v0.h
+; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v1.l
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_ldexp_f16_i16:

diff  --git a/llvm/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll
index 3faf84e5d58c8..863240cc591c3 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll
@@ -135,9 +135,8 @@ define amdgpu_kernel void @maxnum_f16(
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
 ; GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.l, v0.l, v0.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.h, v0.h, v0.h
+; GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.h, v1.l, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.l, v0.l, v0.h
 ; GFX11-TRUE16-NEXT:    buffer_store_b16 v0, off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm

diff  --git a/llvm/test/CodeGen/AMDGPU/mad-mix.ll b/llvm/test/CodeGen/AMDGPU/mad-mix.ll
index a4568b58661db..1bf6559f66feb 100644
--- a/llvm/test/CodeGen/AMDGPU/mad-mix.ll
+++ b/llvm/test/CodeGen/AMDGPU/mad-mix.ll
@@ -1970,15 +1970,12 @@ define float @v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd(half %src0,
 ; SDAG-GFX1100-TRUE16-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
 ; SDAG-GFX1100-TRUE16:       ; %bb.0:
 ; SDAG-GFX1100-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX1100-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-GFX1100-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v2, v0.l
-; SDAG-GFX1100-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v0, v0.h
+; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v0, v0.l
 ; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v1, v1.l
+; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v2, v2.l
 ; SDAG-GFX1100-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; SDAG-GFX1100-TRUE16-NEXT:    v_mul_f32_e32 v0, v2, v0
-; SDAG-GFX1100-TRUE16-NEXT:    v_add_f32_e32 v0, v0, v1
+; SDAG-GFX1100-TRUE16-NEXT:    v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX1100-TRUE16-NEXT:    v_add_f32_e32 v0, v0, v2
 ; SDAG-GFX1100-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; SDAG-GFX1100-FAKE16-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
@@ -2071,12 +2068,10 @@ define float @v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd(half %src0, half
 ; SDAG-GFX1100-TRUE16-LABEL: v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd:
 ; SDAG-GFX1100-TRUE16:       ; %bb.0:
 ; SDAG-GFX1100-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX1100-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v1, v0.l
-; SDAG-GFX1100-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v0, v0.h
-; SDAG-GFX1100-TRUE16-NEXT:    v_mul_f32_e32 v0, v1, v0
-; SDAG-GFX1100-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v0, v0.l
+; SDAG-GFX1100-TRUE16-NEXT:    v_cvt_f32_f16_e32 v1, v1.l
+; SDAG-GFX1100-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; SDAG-GFX1100-TRUE16-NEXT:    v_mul_f32_e32 v0, v0, v1
 ; SDAG-GFX1100-TRUE16-NEXT:    v_add_f32_e32 v0, v0, v2
 ; SDAG-GFX1100-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;

diff  --git a/llvm/test/CodeGen/AMDGPU/mad.u16.ll b/llvm/test/CodeGen/AMDGPU/mad.u16.ll
index 64668f006aab4..9ad5626cc7a2c 100644
--- a/llvm/test/CodeGen/AMDGPU/mad.u16.ll
+++ b/llvm/test/CodeGen/AMDGPU/mad.u16.ll
@@ -142,10 +142,7 @@ define i16 @v_mad_u16(i16 %arg0, i16 %arg1, i16 %arg2) {
 ; GFX11-TRUE16-LABEL: v_mad_u16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_mad_u16 v0.l, v0.l, v0.h, v1.l
+; GFX11-TRUE16-NEXT:    v_mad_u16 v0.l, v0.l, v1.l, v2.l
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: v_mad_u16:
@@ -181,10 +178,8 @@ define i32 @v_mad_u16_zext(i16 %arg0, i16 %arg1, i16 %arg2) {
 ; GFX11-TRUE16-LABEL: v_mad_u16_zext:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_mad_u16 v0.l, v0.l, v0.h, v1.l
+; GFX11-TRUE16-NEXT:    v_mad_u16 v0.l, v0.l, v1.l, v2.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -227,10 +222,8 @@ define i64 @v_mad_u16_zext64(i16 %arg0, i16 %arg1, i16 %arg2) {
 ; GFX11-TRUE16-LABEL: v_mad_u16_zext64:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_mad_u16 v0.l, v0.l, v0.h, v1.l
+; GFX11-TRUE16-NEXT:    v_mad_u16 v0.l, v0.l, v1.l, v2.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-TRUE16-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;

diff  --git a/llvm/test/CodeGen/AMDGPU/minimummaximum.ll b/llvm/test/CodeGen/AMDGPU/minimummaximum.ll
index deedb36a69cbd..36148313edbb5 100644
--- a/llvm/test/CodeGen/AMDGPU/minimummaximum.ll
+++ b/llvm/test/CodeGen/AMDGPU/minimummaximum.ll
@@ -76,10 +76,7 @@ define amdgpu_ps float @test_maxmin_commuted_f32(float %a, float %b, float %c) {
 define amdgpu_ps half @test_minmax_f16(half %a, half %b, half %c) {
 ; SDAG-TRUE16-LABEL: test_minmax_f16:
 ; SDAG-TRUE16:       ; %bb.0:
-; SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; SDAG-TRUE16-NEXT:    v_maximumminimum_f16 v0.l, v0.l, v0.h, v1.l
+; SDAG-TRUE16-NEXT:    v_maximumminimum_f16 v0.l, v0.l, v1.l, v2.l
 ; SDAG-TRUE16-NEXT:    ; return to shader part epilog
 ;
 ; SDAG-FAKE16-LABEL: test_minmax_f16:
@@ -104,10 +101,7 @@ define amdgpu_ps half @test_minmax_f16(half %a, half %b, half %c) {
 define amdgpu_ps half @test_minmax_commuted_f16(half %a, half %b, half %c) {
 ; SDAG-TRUE16-LABEL: test_minmax_commuted_f16:
 ; SDAG-TRUE16:       ; %bb.0:
-; SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; SDAG-TRUE16-NEXT:    v_maximumminimum_f16 v0.l, v0.l, v0.h, v1.l
+; SDAG-TRUE16-NEXT:    v_maximumminimum_f16 v0.l, v0.l, v1.l, v2.l
 ; SDAG-TRUE16-NEXT:    ; return to shader part epilog
 ;
 ; SDAG-FAKE16-LABEL: test_minmax_commuted_f16:
@@ -132,10 +126,7 @@ define amdgpu_ps half @test_minmax_commuted_f16(half %a, half %b, half %c) {
 define amdgpu_ps half @test_maxmin_commuted_f16(half %a, half %b, half %c) {
 ; SDAG-TRUE16-LABEL: test_maxmin_commuted_f16:
 ; SDAG-TRUE16:       ; %bb.0:
-; SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; SDAG-TRUE16-NEXT:    v_minimummaximum_f16 v0.l, v0.l, v0.h, v1.l
+; SDAG-TRUE16-NEXT:    v_minimummaximum_f16 v0.l, v0.l, v1.l, v2.l
 ; SDAG-TRUE16-NEXT:    ; return to shader part epilog
 ;
 ; SDAG-FAKE16-LABEL: test_maxmin_commuted_f16:

diff  --git a/llvm/test/CodeGen/AMDGPU/minmax.ll b/llvm/test/CodeGen/AMDGPU/minmax.ll
index b16d0b7e1213b..e8e1837c58bbb 100644
--- a/llvm/test/CodeGen/AMDGPU/minmax.ll
+++ b/llvm/test/CodeGen/AMDGPU/minmax.ll
@@ -473,9 +473,7 @@ define void @test_med3_f32(ptr addrspace(1) %arg, float %x, float %y, float %z)
 define amdgpu_ps half @test_minmax_f16_ieee_false(half %a, half %b, half %c) {
 ; SDAG-GFX11-TRUE16-LABEL: test_minmax_f16_ieee_false:
 ; SDAG-GFX11-TRUE16:       ; %bb.0:
-; SDAG-GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; SDAG-GFX11-TRUE16-NEXT:    v_maxmin_f16 v0.l, v0.l, v0.h, v1.l
+; SDAG-GFX11-TRUE16-NEXT:    v_maxmin_f16 v0.l, v0.l, v1.l, v2.l
 ; SDAG-GFX11-TRUE16-NEXT:    ; return to shader part epilog
 ;
 ; SDAG-GFX11-FAKE16-LABEL: test_minmax_f16_ieee_false:
@@ -495,9 +493,7 @@ define amdgpu_ps half @test_minmax_f16_ieee_false(half %a, half %b, half %c) {
 ;
 ; SDAG-GFX12-TRUE16-LABEL: test_minmax_f16_ieee_false:
 ; SDAG-GFX12-TRUE16:       ; %bb.0:
-; SDAG-GFX12-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-GFX12-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; SDAG-GFX12-TRUE16-NEXT:    v_maxmin_num_f16 v0.l, v0.l, v0.h, v1.l
+; SDAG-GFX12-TRUE16-NEXT:    v_maxmin_num_f16 v0.l, v0.l, v1.l, v2.l
 ; SDAG-GFX12-TRUE16-NEXT:    ; return to shader part epilog
 ;
 ; SDAG-GFX12-FAKE16-LABEL: test_minmax_f16_ieee_false:
@@ -608,11 +604,10 @@ define half @test_minmax_commuted_f16_ieee_true(half %a, half %b, half %c) {
 ; SDAG-GFX11-TRUE16-LABEL: test_minmax_commuted_f16_ieee_true:
 ; SDAG-GFX11-TRUE16:       ; %bb.0:
 ; SDAG-GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v2.l
-; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v1.l, v1.l, v1.l
+; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.h, v1.l, v1.l
 ; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.l, v0.l, v0.l
-; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.h, v0.h, v0.h
-; SDAG-GFX11-TRUE16-NEXT:    v_maxmin_f16 v0.l, v0.l, v1.l, v0.h
+; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v1.l, v2.l, v2.l
+; SDAG-GFX11-TRUE16-NEXT:    v_maxmin_f16 v0.l, v0.l, v0.h, v1.l
 ; SDAG-GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; SDAG-GFX11-FAKE16-LABEL: test_minmax_commuted_f16_ieee_true:
@@ -649,11 +644,10 @@ define half @test_minmax_commuted_f16_ieee_true(half %a, half %b, half %c) {
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
-; SDAG-GFX12-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v2.l
-; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v1.l, v1.l, v1.l
+; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v0.h, v1.l, v1.l
 ; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v0.l, v0.l, v0.l
-; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v0.h, v0.h, v0.h
-; SDAG-GFX12-TRUE16-NEXT:    v_maxmin_num_f16 v0.l, v0.l, v1.l, v0.h
+; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v1.l, v2.l, v2.l
+; SDAG-GFX12-TRUE16-NEXT:    v_maxmin_num_f16 v0.l, v0.l, v0.h, v1.l
 ; SDAG-GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; SDAG-GFX12-FAKE16-LABEL: test_minmax_commuted_f16_ieee_true:
@@ -702,9 +696,7 @@ define half @test_minmax_commuted_f16_ieee_true(half %a, half %b, half %c) {
 define amdgpu_ps half @test_maxmin_f16_ieee_false(half %a, half %b, half %c) {
 ; SDAG-GFX11-TRUE16-LABEL: test_maxmin_f16_ieee_false:
 ; SDAG-GFX11-TRUE16:       ; %bb.0:
-; SDAG-GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; SDAG-GFX11-TRUE16-NEXT:    v_minmax_f16 v0.l, v0.l, v0.h, v1.l
+; SDAG-GFX11-TRUE16-NEXT:    v_minmax_f16 v0.l, v0.l, v1.l, v2.l
 ; SDAG-GFX11-TRUE16-NEXT:    ; return to shader part epilog
 ;
 ; SDAG-GFX11-FAKE16-LABEL: test_maxmin_f16_ieee_false:
@@ -724,9 +716,7 @@ define amdgpu_ps half @test_maxmin_f16_ieee_false(half %a, half %b, half %c) {
 ;
 ; SDAG-GFX12-TRUE16-LABEL: test_maxmin_f16_ieee_false:
 ; SDAG-GFX12-TRUE16:       ; %bb.0:
-; SDAG-GFX12-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; SDAG-GFX12-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; SDAG-GFX12-TRUE16-NEXT:    v_minmax_num_f16 v0.l, v0.l, v0.h, v1.l
+; SDAG-GFX12-TRUE16-NEXT:    v_minmax_num_f16 v0.l, v0.l, v1.l, v2.l
 ; SDAG-GFX12-TRUE16-NEXT:    ; return to shader part epilog
 ;
 ; SDAG-GFX12-FAKE16-LABEL: test_maxmin_f16_ieee_false:
@@ -752,11 +742,10 @@ define half @test_maxmin_commuted_f16_ieee_true(half %a, half %b, half %c) {
 ; SDAG-GFX11-TRUE16-LABEL: test_maxmin_commuted_f16_ieee_true:
 ; SDAG-GFX11-TRUE16:       ; %bb.0:
 ; SDAG-GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v2.l
-; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v1.l, v1.l, v1.l
+; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.h, v1.l, v1.l
 ; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.l, v0.l, v0.l
-; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v0.h, v0.h, v0.h
-; SDAG-GFX11-TRUE16-NEXT:    v_minmax_f16 v0.l, v0.l, v1.l, v0.h
+; SDAG-GFX11-TRUE16-NEXT:    v_max_f16_e32 v1.l, v2.l, v2.l
+; SDAG-GFX11-TRUE16-NEXT:    v_minmax_f16 v0.l, v0.l, v0.h, v1.l
 ; SDAG-GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; SDAG-GFX11-FAKE16-LABEL: test_maxmin_commuted_f16_ieee_true:
@@ -793,11 +782,10 @@ define half @test_maxmin_commuted_f16_ieee_true(half %a, half %b, half %c) {
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
-; SDAG-GFX12-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v2.l
-; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v1.l, v1.l, v1.l
+; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v0.h, v1.l, v1.l
 ; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v0.l, v0.l, v0.l
-; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v0.h, v0.h, v0.h
-; SDAG-GFX12-TRUE16-NEXT:    v_minmax_num_f16 v0.l, v0.l, v1.l, v0.h
+; SDAG-GFX12-TRUE16-NEXT:    v_max_num_f16_e32 v1.l, v2.l, v2.l
+; SDAG-GFX12-TRUE16-NEXT:    v_minmax_num_f16 v0.l, v0.l, v0.h, v1.l
 ; SDAG-GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; SDAG-GFX12-FAKE16-LABEL: test_maxmin_commuted_f16_ieee_true:
@@ -847,9 +835,7 @@ define void @test_med3_f16(ptr addrspace(1) %arg, half %x, half %y, half %z) #0
 ; SDAG-GFX11-TRUE16-LABEL: test_med3_f16:
 ; SDAG-GFX11-TRUE16:       ; %bb.0:
 ; SDAG-GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.h, v3.l
-; SDAG-GFX11-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v4.l
-; SDAG-GFX11-TRUE16-NEXT:    v_med3_f16 v2.l, v2.l, v2.h, v3.l
+; SDAG-GFX11-TRUE16-NEXT:    v_med3_f16 v2.l, v2.l, v3.l, v4.l
 ; SDAG-GFX11-TRUE16-NEXT:    global_store_b16 v[0:1], v2, off
 ; SDAG-GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -881,9 +867,7 @@ define void @test_med3_f16(ptr addrspace(1) %arg, half %x, half %y, half %z) #0
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
 ; SDAG-GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
-; SDAG-GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.h, v3.l
-; SDAG-GFX12-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v4.l
-; SDAG-GFX12-TRUE16-NEXT:    v_med3_num_f16 v2.l, v2.l, v2.h, v3.l
+; SDAG-GFX12-TRUE16-NEXT:    v_med3_num_f16 v2.l, v2.l, v3.l, v4.l
 ; SDAG-GFX12-TRUE16-NEXT:    global_store_b16 v[0:1], v2, off
 ; SDAG-GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;

diff  --git a/llvm/test/CodeGen/AMDGPU/mul.i16.ll b/llvm/test/CodeGen/AMDGPU/mul.i16.ll
index 2ba719c4138c6..ba4c29e78514c 100644
--- a/llvm/test/CodeGen/AMDGPU/mul.i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/mul.i16.ll
@@ -11,7 +11,7 @@
 
 ; GFX89: v_mul_lo_u16_e32 v0, v0, v1
 
-; GFX11-TRUE16: v_mul_lo_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16: v_mul_lo_u16 v{{[0-9]+}}.{{(l|h)}}, v{{[0-9]+}}.{{(l|h)}}, v{{[0-9]+}}.{{(l|h)}}
 ; GFX11-FAKE16: v_mul_lo_u16 v0, v0, v1
 define i16 @v_mul_i16(i16 %a, i16 %b) {
   %r.val = mul i16 %a, %b

diff  --git a/llvm/test/CodeGen/AMDGPU/preserve-hi16.ll b/llvm/test/CodeGen/AMDGPU/preserve-hi16.ll
index 1f36101c7b53a..20ca575e1aa47 100644
--- a/llvm/test/CodeGen/AMDGPU/preserve-hi16.ll
+++ b/llvm/test/CodeGen/AMDGPU/preserve-hi16.ll
@@ -130,8 +130,7 @@ define i16 @add_u16(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: add_u16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_add_nc_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_add_nc_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: add_u16:
@@ -165,8 +164,7 @@ define i16 @sub_u16(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: sub_u16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_sub_nc_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_sub_nc_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: sub_u16:
@@ -200,8 +198,7 @@ define i16 @mul_lo_u16(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: mul_lo_u16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mul_lo_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_mul_lo_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: mul_lo_u16:
@@ -235,8 +232,7 @@ define i16 @min_u16(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: min_u16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_min_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_min_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: min_u16:
@@ -271,8 +267,7 @@ define i16 @min_i16(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: min_i16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_min_i16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_min_i16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: min_i16:
@@ -307,8 +302,7 @@ define i16 @max_u16(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: max_u16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_max_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_max_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: max_u16:
@@ -343,8 +337,7 @@ define i16 @max_i16(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: max_i16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_max_i16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_max_i16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: max_i16:
@@ -494,8 +487,7 @@ define i32 @add_u16_zext_i32(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: add_u16_zext_i32:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_add_nc_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_add_nc_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -533,8 +525,7 @@ define i32 @sub_u16_zext_i32(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: sub_u16_zext_i32:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_sub_nc_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_sub_nc_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -572,8 +563,7 @@ define i32 @mul_lo_u16_zext_i32(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: mul_lo_u16_zext_i32:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mul_lo_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_mul_lo_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -611,8 +601,7 @@ define i32 @min_u16_zext_i32(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: min_u16_zext_i32:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_min_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_min_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -651,8 +640,7 @@ define i32 @min_i16_zext_i32(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: min_i16_zext_i32:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_min_i16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_min_i16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -691,8 +679,7 @@ define i32 @max_u16_zext_i32(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: max_u16_zext_i32:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_max_u16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_max_u16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -731,8 +718,7 @@ define i32 @max_i16_zext_i32(i16 %x, i16 %y) {
 ; GFX11-TRUE16-LABEL: max_i16_zext_i32:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_max_i16 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_max_i16 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -771,8 +757,7 @@ define i32 @zext_fadd_f16(half %x, half %y) {
 ; GFX11-TRUE16-LABEL: zext_fadd_f16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_add_f16_e32 v0.l, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_add_f16_e32 v0.l, v0.l, v1.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -812,11 +797,8 @@ define i32 @zext_fma_f16(half %x, half %y, half %z) {
 ; GFX11-TRUE16-LABEL: zext_fma_f16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; GFX11-TRUE16-NEXT:    v_fmac_f16_e32 v1.l, v0.l, v0.h
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v1.l
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX11-TRUE16-NEXT:    v_fmac_f16_e32 v2.l, v0.l, v1.l
+; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v2
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: zext_fma_f16:
@@ -855,9 +837,7 @@ define i32 @zext_div_fixup_f16(half %x, half %y, half %z) {
 ; GFX11-TRUE16-LABEL: zext_div_fixup_f16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v2.l
-; GFX11-TRUE16-NEXT:    v_div_fixup_f16 v0.l, v0.l, v0.h, v1.l
+; GFX11-TRUE16-NEXT:    v_div_fixup_f16 v0.l, v0.l, v1.l, v2.l
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;

diff  --git a/llvm/test/CodeGen/AMDGPU/saddsat.ll b/llvm/test/CodeGen/AMDGPU/saddsat.ll
index a034cf6941c26..1a17efa562b79 100644
--- a/llvm/test/CodeGen/AMDGPU/saddsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/saddsat.ll
@@ -104,8 +104,7 @@ define i16 @v_saddsat_i16(i16 %lhs, i16 %rhs) {
 ; GFX11-TRUE16-LABEL: v_saddsat_i16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_add_nc_i16 v0.l, v0.l, v0.h clamp
+; GFX11-TRUE16-NEXT:    v_add_nc_i16 v0.l, v0.l, v1.l clamp
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: v_saddsat_i16:

diff  --git a/llvm/test/CodeGen/AMDGPU/ssubsat.ll b/llvm/test/CodeGen/AMDGPU/ssubsat.ll
index 30a0a26ca173e..877088be27086 100644
--- a/llvm/test/CodeGen/AMDGPU/ssubsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/ssubsat.ll
@@ -104,8 +104,7 @@ define i16 @v_ssubsat_i16(i16 %lhs, i16 %rhs) {
 ; GFX11-TRUE16-LABEL: v_ssubsat_i16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_sub_nc_i16 v0.l, v0.l, v0.h clamp
+; GFX11-TRUE16-NEXT:    v_sub_nc_i16 v0.l, v0.l, v1.l clamp
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: v_ssubsat_i16:

diff  --git a/llvm/test/CodeGen/AMDGPU/uaddsat.ll b/llvm/test/CodeGen/AMDGPU/uaddsat.ll
index 572793e1c5d71..056e1d038571b 100644
--- a/llvm/test/CodeGen/AMDGPU/uaddsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/uaddsat.ll
@@ -93,9 +93,7 @@ define i16 @v_uaddsat_i16(i16 %lhs, i16 %rhs) {
 ; GFX11-TRUE16-LABEL: v_uaddsat_i16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_add_nc_u16 v0.l, v0.l, v0.h clamp
+; GFX11-TRUE16-NEXT:    v_add_nc_u16 v0.l, v0.l, v1.l clamp
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: v_uaddsat_i16:

diff  --git a/llvm/test/CodeGen/AMDGPU/usubsat.ll b/llvm/test/CodeGen/AMDGPU/usubsat.ll
index 75866e33da23a..cfcb2438e1768 100644
--- a/llvm/test/CodeGen/AMDGPU/usubsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/usubsat.ll
@@ -86,8 +86,7 @@ define i16 @v_usubsat_i16(i16 %lhs, i16 %rhs) {
 ; GFX11-TRUE16-LABEL: v_usubsat_i16:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_sub_nc_u16 v0.l, v0.l, v0.h clamp
+; GFX11-TRUE16-NEXT:    v_sub_nc_u16 v0.l, v0.l, v1.l clamp
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-FAKE16-LABEL: v_usubsat_i16:


        


More information about the llvm-commits mailing list