https://github.com/lialan created 
https://github.com/llvm/llvm-project/pull/237626

Allow legal `v4i16` and `v4bf16` shuffles in `isShuffleMaskLegal` so DAG 
combines can preserve packed 16-bit pairs and lowering can use `op_sel` and 
`v_pk_mov_b32` instead of extracting each half into a separate register. Keep 
scalarization for other vector shapes to avoid inhibiting load narrowing or 
adding register copies. The four existing AMDGPU tests improve in 52 
function/configuration cases, removing 54 instructions with no 
instruction-count increases.


>From cf23347e538fe74b1cbd03833634561c418e86bf Mon Sep 17 00:00:00 2001
From: Alan Li <[email protected]>
Date: Sat, 31 Oct 2026 12:34:54 -0700
Subject: [PATCH] [AMDGPU] Allow legal shuffles of four 16-bit lanes

Preserve packed 16-bit pairs so lowering can use op_sel and
v_pk_mov_b32 instead of extracting each half into a separate register.
Keep scalarization for other vector shapes to avoid inhibiting load
narrowing or introducing extra register copies.
---
 llvm/lib/Target/AMDGPU/SIISelLowering.cpp     |  12 +-
 .../AMDGPU/shufflevector.v3bf16.v2bf16.ll     |  16 +--
 .../AMDGPU/shufflevector.v3bf16.v3bf16.ll     | 127 +++++++-----------
 .../AMDGPU/shufflevector.v3f16.v2f16.ll       |  16 +--
 .../AMDGPU/shufflevector.v3f16.v3f16.ll       | 127 +++++++-----------
 5 files changed, 124 insertions(+), 174 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp 
b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index df1d56339342f3..e656cba2b68264 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1267,10 +1267,14 @@ bool SITargetLowering::isFPExtFoldable(const 
MachineInstr &MI, unsigned Opcode,
          denormalModeIsFlushAllF32(*MI.getMF());
 }
 
-bool SITargetLowering::isShuffleMaskLegal(ArrayRef<int>, EVT) const {
-  // SI has some legal vector types, but no legal vector operations. Say no
-  // shuffles are legal in order to prefer scalarizing some vector operations.
-  return false;
+bool SITargetLowering::isShuffleMaskLegal(ArrayRef<int>, EVT VT) const {
+  // Preserve packed 16-bit pairs so lowering can use op_sel and v_pk_mov_b32
+  // instead of extracting each half into a separate register. Prefer
+  // scalarization for other shapes. Larger vectors can inhibit load narrowing,
+  // while 32-bit shuffles can add register copies.
+  return VT.isFixedLengthVector() && VT.getScalarSizeInBits() == 16 &&
+         VT.getVectorNumElements() == 4 && isTypeLegal(VT) &&
+         isOperationLegalOrCustom(ISD::VECTOR_SHUFFLE, VT);
 }
 
 MVT SITargetLowering::getRegisterTypeForCallingConv(LLVMContext &Context,
diff --git a/llvm/test/CodeGen/AMDGPU/shufflevector.v3bf16.v2bf16.ll 
b/llvm/test/CodeGen/AMDGPU/shufflevector.v3bf16.v2bf16.ll
index 86dfc6f37680eb..832a20a456a1a1 100644
--- a/llvm/test/CodeGen/AMDGPU/shufflevector.v3bf16.v2bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/shufflevector.v3bf16.v2bf16.ll
@@ -1284,16 +1284,15 @@ define void @v_shuffle_v3bf16_v2bf16__3_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3bf16_v2bf16__3_0_1:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v1
 ; GFX900-NEXT:    ;;#ASMEND
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v2
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX900-NEXT:    v_alignbit_b32 v2, v1, v2, 16
-; GFX900-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX900-NEXT:    global_store_short v0, v1, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v0, v1, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v0, v2, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -1301,16 +1300,15 @@ define void @v_shuffle_v3bf16_v2bf16__3_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3bf16_v2bf16__3_0_1:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v1
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v2
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX90A-NEXT:    v_alignbit_b32 v2, v1, v2, 16
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX90A-NEXT:    global_store_short v0, v1, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v0, v1, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v0, v2, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -1318,16 +1316,16 @@ define void @v_shuffle_v3bf16_v2bf16__3_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3bf16_v2bf16__3_0_1:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v1
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v2
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v0, 0
+; GFX942-NEXT:    s_nop 0
 ; GFX942-NEXT:    v_alignbit_b32 v2, v1, v2, 16
-; GFX942-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX942-NEXT:    global_store_short v0, v1, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v0, v1, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v0, v2, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/shufflevector.v3bf16.v3bf16.ll 
b/llvm/test/CodeGen/AMDGPU/shufflevector.v3bf16.v3bf16.ll
index 08850a6625bc0f..ac9403b788c0e4 100644
--- a/llvm/test/CodeGen/AMDGPU/shufflevector.v3bf16.v3bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/shufflevector.v3bf16.v3bf16.ll
@@ -797,14 +797,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_5_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3bf16_v3bf16__5_5_4:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX900-NEXT:    v_perm_b32 v1, v1, v1, s4
-; GFX900-NEXT:    global_store_short v2, v0, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v2, v1, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -812,14 +811,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_5_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3bf16_v3bf16__5_5_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX90A-NEXT:    v_perm_b32 v1, v1, v1, s4
-; GFX90A-NEXT:    global_store_short v2, v0, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v2, v1, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -827,14 +825,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_5_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3bf16_v3bf16__5_5_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX942-NEXT:    v_perm_b32 v1, v1, v1, s2
-; GFX942-NEXT:    global_store_short v2, v0, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v2, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v2, v1, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -1985,14 +1982,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
+; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[1:2]
 ; GFX900-NEXT:    ;;#ASMEND
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    v_perm_b32 v1, v0, v2, s4
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX900-NEXT:    global_store_short v3, v0, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v3, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v3, v1, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -2000,17 +1996,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3bf16_v3bf16__5_0_1:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    v_perm_b32 v1, v0, v3, s4
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX90A-NEXT:    global_store_short v4, v0, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v1, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -2018,18 +2013,17 @@ define void @v_shuffle_v3bf16_v3bf16__5_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3bf16_v3bf16__5_0_1:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_nop 0
 ; GFX942-NEXT:    v_perm_b32 v1, v0, v3, s2
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX942-NEXT:    global_store_short v4, v0, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v1, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -2046,17 +2040,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_2_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3bf16_v3bf16__5_2_1:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[2:3]
 ; GFX900-NEXT:    ;;#ASMEND
 ; GFX900-NEXT:    v_perm_b32 v1, v1, v3, s4
-; GFX900-NEXT:    global_store_short v4, v0, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v4, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v4, v1, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -2064,17 +2057,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_2_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3bf16_v3bf16__5_2_1:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    v_perm_b32 v1, v1, v3, s4
-; GFX90A-NEXT:    global_store_short v4, v0, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v1, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -2082,18 +2074,17 @@ define void @v_shuffle_v3bf16_v3bf16__5_2_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3bf16_v3bf16__5_2_1:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_nop 0
 ; GFX942-NEXT:    v_perm_b32 v1, v1, v3, s2
-; GFX942-NEXT:    global_store_short v4, v0, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v1, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -3810,39 +3801,36 @@ define void @v_shuffle_v3bf16_v3bf16__5_u_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3bf16_v3bf16__5_u_4:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX900-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v2, v1, s[16:17]
-; GFX900-NEXT:    global_store_short v2, v0, s[16:17] offset:4
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: v_shuffle_v3bf16_v3bf16__5_u_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX90A-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v2, v1, s[16:17]
-; GFX90A-NEXT:    global_store_short v2, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX942-LABEL: v_shuffle_v3bf16_v3bf16__5_u_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v2, 0
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX942-NEXT:    global_store_short_d16_hi v2, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v2, v1, s[0:1]
-; GFX942-NEXT:    global_store_short v2, v0, s[0:1] offset:4
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
   %vec0 = call <4 x bfloat> asm "; def $0", "=v"()
@@ -3861,14 +3849,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_0_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
+; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[1:2]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
 ; GFX900-NEXT:    v_perm_b32 v0, v0, v2, s4
-; GFX900-NEXT:    global_store_short v3, v1, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v3, v1, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v3, v0, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -3876,17 +3863,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_0_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3bf16_v3bf16__5_0_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
 ; GFX90A-NEXT:    v_perm_b32 v0, v0, v3, s4
-; GFX90A-NEXT:    global_store_short v4, v1, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v2, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v0, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -3894,17 +3880,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_0_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3bf16_v3bf16__5_0_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
 ; GFX942-NEXT:    v_perm_b32 v0, v0, v3, s2
-; GFX942-NEXT:    global_store_short v4, v1, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v2, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v0, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -3924,14 +3909,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_1_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
+; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[1:2]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    s_mov_b32 s4, 0xffff
-; GFX900-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
 ; GFX900-NEXT:    v_bfi_b32 v0, s4, v2, v0
-; GFX900-NEXT:    global_store_short v3, v1, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v3, v1, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v3, v0, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -3939,17 +3923,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_1_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3bf16_v3bf16__5_1_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    s_mov_b32 s4, 0xffff
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
 ; GFX90A-NEXT:    v_bfi_b32 v0, s4, v3, v0
-; GFX90A-NEXT:    global_store_short v4, v1, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v2, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v0, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -3957,17 +3940,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_1_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3bf16_v3bf16__5_1_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_mov_b32 s2, 0xffff
-; GFX942-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
 ; GFX942-NEXT:    v_bfi_b32 v0, s2, v3, v0
-; GFX942-NEXT:    global_store_short v4, v1, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v2, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v0, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -3984,17 +3966,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_2_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3bf16_v3bf16__5_2_4:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[2:3]
 ; GFX900-NEXT:    ;;#ASMEND
+; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX900-NEXT:    v_perm_b32 v0, v1, v3, s4
-; GFX900-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX900-NEXT:    global_store_short v4, v1, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v4, v2, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v4, v0, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -4002,17 +3983,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_2_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3bf16_v3bf16__5_2_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
+; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX90A-NEXT:    v_perm_b32 v0, v1, v3, s4
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX90A-NEXT:    global_store_short v4, v1, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v2, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v0, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -4020,18 +4000,16 @@ define void @v_shuffle_v3bf16_v3bf16__5_2_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3bf16_v3bf16__5_2_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    s_nop 0
+; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
 ; GFX942-NEXT:    v_perm_b32 v0, v1, v3, s2
-; GFX942-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX942-NEXT:    global_store_short v4, v1, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v2, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v0, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -4048,14 +4026,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_3_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3bf16_v3bf16__5_3_4:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    v_perm_b32 v1, v0, v1, s4
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX900-NEXT:    global_store_short v2, v0, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v2, v1, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -4063,14 +4040,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_3_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3bf16_v3bf16__5_3_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    v_perm_b32 v1, v0, v1, s4
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX90A-NEXT:    global_store_short v2, v0, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v2, v1, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -4078,14 +4054,13 @@ define void @v_shuffle_v3bf16_v3bf16__5_3_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3bf16_v3bf16__5_3_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    v_perm_b32 v1, v0, v1, s2
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX942-NEXT:    global_store_short v2, v0, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v2, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v2, v1, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/shufflevector.v3f16.v2f16.ll 
b/llvm/test/CodeGen/AMDGPU/shufflevector.v3f16.v2f16.ll
index 6aa9657fd06f3b..bcbf8622cf550b 100644
--- a/llvm/test/CodeGen/AMDGPU/shufflevector.v3f16.v2f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/shufflevector.v3f16.v2f16.ll
@@ -1284,16 +1284,15 @@ define void @v_shuffle_v3f16_v2f16__3_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3f16_v2f16__3_0_1:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v1
 ; GFX900-NEXT:    ;;#ASMEND
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v2
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX900-NEXT:    v_alignbit_b32 v2, v1, v2, 16
-; GFX900-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX900-NEXT:    global_store_short v0, v1, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v0, v1, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v0, v2, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -1301,16 +1300,15 @@ define void @v_shuffle_v3f16_v2f16__3_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3f16_v2f16__3_0_1:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v1
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v2
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX90A-NEXT:    v_alignbit_b32 v2, v1, v2, 16
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX90A-NEXT:    global_store_short v0, v1, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v0, v1, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v0, v2, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -1318,16 +1316,16 @@ define void @v_shuffle_v3f16_v2f16__3_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3f16_v2f16__3_0_1:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v1
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v2
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v0, 0
+; GFX942-NEXT:    s_nop 0
 ; GFX942-NEXT:    v_alignbit_b32 v2, v1, v2, 16
-; GFX942-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX942-NEXT:    global_store_short v0, v1, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v0, v1, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v0, v2, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/shufflevector.v3f16.v3f16.ll 
b/llvm/test/CodeGen/AMDGPU/shufflevector.v3f16.v3f16.ll
index bf3f0a319df95a..4a5e9d3eb3512f 100644
--- a/llvm/test/CodeGen/AMDGPU/shufflevector.v3f16.v3f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/shufflevector.v3f16.v3f16.ll
@@ -797,14 +797,13 @@ define void @v_shuffle_v3f16_v3f16__5_5_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3f16_v3f16__5_5_4:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX900-NEXT:    v_perm_b32 v1, v1, v1, s4
-; GFX900-NEXT:    global_store_short v2, v0, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v2, v1, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -812,14 +811,13 @@ define void @v_shuffle_v3f16_v3f16__5_5_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3f16_v3f16__5_5_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX90A-NEXT:    v_perm_b32 v1, v1, v1, s4
-; GFX90A-NEXT:    global_store_short v2, v0, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v2, v1, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -827,14 +825,13 @@ define void @v_shuffle_v3f16_v3f16__5_5_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3f16_v3f16__5_5_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX942-NEXT:    v_perm_b32 v1, v1, v1, s2
-; GFX942-NEXT:    global_store_short v2, v0, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v2, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v2, v1, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -1985,14 +1982,13 @@ define void @v_shuffle_v3f16_v3f16__5_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
+; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[1:2]
 ; GFX900-NEXT:    ;;#ASMEND
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    v_perm_b32 v1, v0, v2, s4
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX900-NEXT:    global_store_short v3, v0, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v3, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v3, v1, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -2000,17 +1996,16 @@ define void @v_shuffle_v3f16_v3f16__5_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3f16_v3f16__5_0_1:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    v_perm_b32 v1, v0, v3, s4
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX90A-NEXT:    global_store_short v4, v0, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v1, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -2018,18 +2013,17 @@ define void @v_shuffle_v3f16_v3f16__5_0_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3f16_v3f16__5_0_1:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_nop 0
 ; GFX942-NEXT:    v_perm_b32 v1, v0, v3, s2
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX942-NEXT:    global_store_short v4, v0, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v1, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -2046,17 +2040,16 @@ define void @v_shuffle_v3f16_v3f16__5_2_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3f16_v3f16__5_2_1:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[2:3]
 ; GFX900-NEXT:    ;;#ASMEND
 ; GFX900-NEXT:    v_perm_b32 v1, v1, v3, s4
-; GFX900-NEXT:    global_store_short v4, v0, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v4, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v4, v1, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -2064,17 +2057,16 @@ define void @v_shuffle_v3f16_v3f16__5_2_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3f16_v3f16__5_2_1:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    v_perm_b32 v1, v1, v3, s4
-; GFX90A-NEXT:    global_store_short v4, v0, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v1, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -2082,18 +2074,17 @@ define void @v_shuffle_v3f16_v3f16__5_2_1(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3f16_v3f16__5_2_1:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_nop 0
 ; GFX942-NEXT:    v_perm_b32 v1, v1, v3, s2
-; GFX942-NEXT:    global_store_short v4, v0, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v1, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -3810,39 +3801,36 @@ define void @v_shuffle_v3f16_v3f16__5_u_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3f16_v3f16__5_u_4:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX900-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v2, v1, s[16:17]
-; GFX900-NEXT:    global_store_short v2, v0, s[16:17] offset:4
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: v_shuffle_v3f16_v3f16__5_u_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX90A-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v2, v1, s[16:17]
-; GFX90A-NEXT:    global_store_short v2, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX942-LABEL: v_shuffle_v3f16_v3f16__5_u_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v2, 0
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX942-NEXT:    global_store_short_d16_hi v2, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v2, v1, s[0:1]
-; GFX942-NEXT:    global_store_short v2, v0, s[0:1] offset:4
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
   %vec0 = call <4 x half> asm "; def $0", "=v"()
@@ -3861,14 +3849,13 @@ define void @v_shuffle_v3f16_v3f16__5_0_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
+; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[1:2]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
 ; GFX900-NEXT:    v_perm_b32 v0, v0, v2, s4
-; GFX900-NEXT:    global_store_short v3, v1, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v3, v1, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v3, v0, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -3876,17 +3863,16 @@ define void @v_shuffle_v3f16_v3f16__5_0_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3f16_v3f16__5_0_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
 ; GFX90A-NEXT:    v_perm_b32 v0, v0, v3, s4
-; GFX90A-NEXT:    global_store_short v4, v1, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v2, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v0, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -3894,17 +3880,16 @@ define void @v_shuffle_v3f16_v3f16__5_0_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3f16_v3f16__5_0_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
 ; GFX942-NEXT:    v_perm_b32 v0, v0, v3, s2
-; GFX942-NEXT:    global_store_short v4, v1, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v2, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v0, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -3924,14 +3909,13 @@ define void @v_shuffle_v3f16_v3f16__5_1_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
+; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[1:2]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX900-NEXT:    s_mov_b32 s4, 0xffff
-; GFX900-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
 ; GFX900-NEXT:    v_bfi_b32 v0, s4, v2, v0
-; GFX900-NEXT:    global_store_short v3, v1, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v3, v1, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v3, v0, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -3939,17 +3923,16 @@ define void @v_shuffle_v3f16_v3f16__5_1_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3f16_v3f16__5_1_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    s_mov_b32 s4, 0xffff
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
 ; GFX90A-NEXT:    v_bfi_b32 v0, s4, v3, v0
-; GFX90A-NEXT:    global_store_short v4, v1, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v2, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v0, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -3957,17 +3940,16 @@ define void @v_shuffle_v3f16_v3f16__5_1_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3f16_v3f16__5_1_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_mov_b32 s2, 0xffff
-; GFX942-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
 ; GFX942-NEXT:    v_bfi_b32 v0, s2, v3, v0
-; GFX942-NEXT:    global_store_short v4, v1, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v2, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v0, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -3984,17 +3966,16 @@ define void @v_shuffle_v3f16_v3f16__5_2_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3f16_v3f16__5_2_4:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
-; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[2:3]
 ; GFX900-NEXT:    ;;#ASMEND
+; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX900-NEXT:    v_perm_b32 v0, v1, v3, s4
-; GFX900-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX900-NEXT:    global_store_short v4, v1, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v4, v2, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v4, v0, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -4002,17 +3983,16 @@ define void @v_shuffle_v3f16_v3f16__5_2_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3f16_v3f16__5_2_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
-; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[2:3]
 ; GFX90A-NEXT:    ;;#ASMEND
+; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX90A-NEXT:    v_perm_b32 v0, v1, v3, s4
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX90A-NEXT:    global_store_short v4, v1, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v4, v2, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v4, v0, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -4020,18 +4000,16 @@ define void @v_shuffle_v3f16_v3f16__5_2_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3f16_v3f16__5_2_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[2:3]
 ; GFX942-NEXT:    ;;#ASMEND
-; GFX942-NEXT:    s_nop 0
+; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
 ; GFX942-NEXT:    v_perm_b32 v0, v1, v3, s2
-; GFX942-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX942-NEXT:    global_store_short v4, v1, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v4, v2, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v4, v0, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]
@@ -4048,14 +4026,13 @@ define void @v_shuffle_v3f16_v3f16__5_3_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX900-LABEL: v_shuffle_v3f16_v3f16__5_3_4:
 ; GFX900:       ; %bb.0:
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    ;;#ASMSTART
 ; GFX900-NEXT:    ; def v[0:1]
 ; GFX900-NEXT:    ;;#ASMEND
 ; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    v_perm_b32 v1, v0, v1, s4
-; GFX900-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX900-NEXT:    global_store_short v2, v0, s[16:17] offset:4
+; GFX900-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX900-NEXT:    global_store_dword v2, v1, s[16:17]
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
@@ -4063,14 +4040,13 @@ define void @v_shuffle_v3f16_v3f16__5_3_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX90A-LABEL: v_shuffle_v3f16_v3f16__5_3_4:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    ;;#ASMSTART
 ; GFX90A-NEXT:    ; def v[0:1]
 ; GFX90A-NEXT:    ;;#ASMEND
 ; GFX90A-NEXT:    s_mov_b32 s4, 0x5040100
-; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX90A-NEXT:    v_perm_b32 v1, v0, v1, s4
-; GFX90A-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX90A-NEXT:    global_store_short v2, v0, s[16:17] offset:4
+; GFX90A-NEXT:    global_store_short_d16_hi v2, v0, s[16:17] offset:4
 ; GFX90A-NEXT:    global_store_dword v2, v1, s[16:17]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
@@ -4078,14 +4054,13 @@ define void @v_shuffle_v3f16_v3f16__5_3_4(ptr 
addrspace(1) inreg %ptr) {
 ; GFX942-LABEL: v_shuffle_v3f16_v3f16__5_3_4:
 ; GFX942:       ; %bb.0:
 ; GFX942-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    ;;#ASMSTART
 ; GFX942-NEXT:    ; def v[0:1]
 ; GFX942-NEXT:    ;;#ASMEND
 ; GFX942-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX942-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX942-NEXT:    v_perm_b32 v1, v0, v1, s2
-; GFX942-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX942-NEXT:    global_store_short v2, v0, s[0:1] offset:4
+; GFX942-NEXT:    global_store_short_d16_hi v2, v0, s[0:1] offset:4
 ; GFX942-NEXT:    global_store_dword v2, v1, s[0:1]
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GFX942-NEXT:    s_setpc_b64 s[30:31]

_______________________________________________
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

Reply via email to