AMDGPU/GFX9: Do not use v_pack_b32_f16 when packing

Differential Revision: https://reviews.llvm.org/D31819

llvm-svn: 300275

6 files changed, 52 insertions, 86 deletions

diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index b1e4c9a7aaa..b29cfe116c8 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -3560,36 +3560,22 @@ void SIInstrInfo::movePackToVALU(SmallVectorImpl<MachineInstr *> &Worklist,
 
   switch (Inst.getOpcode()) {
   case AMDGPU::S_PACK_LL_B32_B16: {
-    // v_pack_b32_f16 flushes denormals if not enabled. Use it if the default
-    // is to leave them untouched.
-    // XXX: Does this do anything to NaNs?
-    if (ST.hasFP16Denormals()) {
-      BuildMI(*MBB, Inst, DL, get(AMDGPU::V_PACK_B32_F16), ResultReg)
-        .addImm(0)  // src0_modifiers
-        .add(Src0)  // src0
-        .addImm(0)  // src1_modifiers
-        .add(Src1)  // src2
-        .addImm(0)  // clamp
-        .addImm(0); // omod
-    } else {
-      unsigned ImmReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
-      unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
-
-      // FIXME: Can do a lot better if we know the high bits of src0 or src1 are
-      // 0.
-      BuildMI(*MBB, Inst, DL, get(AMDGPU::V_MOV_B32_e32), ImmReg)
-        .addImm(0xffff);
-
-      BuildMI(*MBB, Inst, DL, get(AMDGPU::V_AND_B32_e64), TmpReg)
-        .addReg(ImmReg, RegState::Kill)
-        .add(Src0);
-
-      BuildMI(*MBB, Inst, DL, get(AMDGPU::V_LSHL_OR_B32), ResultReg)
-        .add(Src1)
-        .addImm(16)
-        .addReg(TmpReg, RegState::Kill);
-    }
+    unsigned ImmReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+    unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+
+    // FIXME: Can do a lot better if we know the high bits of src0 or src1 are
+    // 0.
+    BuildMI(*MBB, Inst, DL, get(AMDGPU::V_MOV_B32_e32), ImmReg)
+      .addImm(0xffff);
 
+    BuildMI(*MBB, Inst, DL, get(AMDGPU::V_AND_B32_e64), TmpReg)
+      .addReg(ImmReg, RegState::Kill)
+      .add(Src0);
+
+    BuildMI(*MBB, Inst, DL, get(AMDGPU::V_LSHL_OR_B32), ResultReg)
+      .add(Src1)
+      .addImm(16)
+      .addReg(TmpReg, RegState::Kill);
     break;
   }
   case AMDGPU::S_PACK_LH_B32_B16: {
diff --git a/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll b/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll
index cdcc7be8f2f..bc72f4424c9 100644
--- a/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll
@@ -1,7 +1,6 @@
 ; RUN: llc -march=amdgcn -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=SIVI %s
 ; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=SIVI %s
-; RUN: llc -march=amdgcn -mcpu=gfx901 -mattr=-flat-for-global,+fp64-fp16-denormals -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=GFX9-DENORM %s
-; RUN: llc -march=amdgcn -mcpu=gfx901 -mattr=-flat-for-global,-fp64-fp16-denormals -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=GFX9-FLUSH %s
+; RUN: llc -march=amdgcn -mcpu=gfx901 -mattr=-flat-for-global,-fp64-fp16-denormals -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 %s
 
 ; GCN-LABEL: {{^}}fptrunc_f32_to_f16:
 ; GCN: buffer_load_dword v[[A_F32:[0-9]+]]
@@ -45,10 +44,8 @@ entry:
 ; VI:     v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], v[[R_F16_0]]
 
 ; GFX9-DAG:   v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[A_F32_1]]
-; GFX9-FLUSH: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
-; GFX9-FLUSH: v_lshl_or_b32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], 16, v[[R_F16_LO]]
-
-; GFX9-DENORM: v_pack_b32_f16 v[[R_V2_F16:[0-9]+]], v[[R_F16_0]], v[[R_F16_1]]
+; GFX9: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; GFX9: v_lshl_or_b32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], 16, v[[R_F16_LO]]
 
 ; GCN:     buffer_store_dword v[[R_V2_F16]]
 ; GCN:     s_endpgm
@@ -74,10 +71,8 @@ entry:
 ; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
 ; GFX9-DAG: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[A_F32_1]]
-; GFX9-FLUSH: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
-; GFX9-FLUSH: v_lshl_or_b32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], 16, v[[R_F16_LO]]
-
-; GFX9-DENORM: v_pack_b32_f16 v[[R_V2_F16:[0-9]+]], v[[R_F16_0]], v[[R_F16_1]]
+; GFX9: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; GFX9: v_lshl_or_b32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], 16, v[[R_F16_LO]]
 
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.rint.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.rint.f16.ll
index f56655630be..59e81a7acc0 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.rint.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.rint.f16.ll
@@ -44,7 +44,8 @@ entry:
 ; GFX9: v_rndne_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
 ; GFX9: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; GFX9: v_rndne_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
-; GFX9: v_pack_b32_f16 v[[R_V2_F16:[0-9]+]], v[[R_F16_0]], v[[R_F16_1]]
+; GFX9: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; GFX9: v_lshl_or_b32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], 16, v[[R_F16_LO]]
 
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.round.ll b/llvm/test/CodeGen/AMDGPU/llvm.round.ll
index d211ad8ec9f..ffe87977870 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.round.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.round.ll
@@ -87,7 +87,8 @@ define amdgpu_kernel void @round_f16(half addrspace(1)* %out, i32 %x.arg) #0 {
 ; GFX89: v_bfi_b32 [[COPYSIGN0:v[0-9]+]], [[K]], [[BFI_K]],
 ; GFX89: v_bfi_b32 [[COPYSIGN1:v[0-9]+]], [[K]], [[BFI_K]],
 
-; GFX9: v_pack_b32_f16
+; GFX9: v_and_b32_e32
+; GFX9: v_lshl_or_b32
 define amdgpu_kernel void @round_v2f16(<2 x half> addrspace(1)* %out, i32 %in.arg) #0 {
   %in = bitcast i32 %in.arg to <2 x half>
   %result = call <2 x half> @llvm.round.v2f16(<2 x half> %in)
diff --git a/llvm/test/CodeGen/AMDGPU/pack.v2f16.ll b/llvm/test/CodeGen/AMDGPU/pack.v2f16.ll
index b8621562713..5a07f7ca6ae 100644
--- a/llvm/test/CodeGen/AMDGPU/pack.v2f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/pack.v2f16.ll
@@ -1,5 +1,4 @@
-; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx901 -mattr=-flat-for-global,+fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=GFX9-DENORM %s
-; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx901 -mattr=-flat-for-global,-fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=GFX9-FLUSH %s
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx901 -mattr=-flat-for-global,-fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 %s
 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI %s
 
@@ -59,10 +58,9 @@ define amdgpu_kernel void @s_pack_v2f16_imm_hi(i32 addrspace(2)* %in0) #0 {
 ; GCN-LABEL: {{^}}v_pack_v2f16:
 ; GFX9: flat_load_dword [[VAL0:v[0-9]+]]
 ; GFX9: flat_load_dword [[VAL1:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL0]], [[VAL1]]
 
-; GFX9-FLUSH: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VAL0]]
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[ELT0]]
+; GFX9: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VAL0]]
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[ELT0]]
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2f16(i32 addrspace(1)* %in0, i32 addrspace(1)* %in1) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -85,10 +83,9 @@ define amdgpu_kernel void @v_pack_v2f16(i32 addrspace(1)* %in0, i32 addrspace(1)
 ; GCN-LABEL: {{^}}v_pack_v2f16_user:
 ; GFX9: flat_load_dword [[VAL0:v[0-9]+]]
 ; GFX9: flat_load_dword [[VAL1:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL0]], [[VAL1]]
 
-; GFX9-FLUSH: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VAL0]]
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[ELT0]]
+; GFX9: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VAL0]]
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[ELT0]]
 
 ; GFX9: v_add_i32_e32 v{{[0-9]+}}, vcc, 9, [[PACKED]]
 define amdgpu_kernel void @v_pack_v2f16_user(i32 addrspace(1)* %in0, i32 addrspace(1)* %in1) #0 {
@@ -112,11 +109,9 @@ define amdgpu_kernel void @v_pack_v2f16_user(i32 addrspace(1)* %in0, i32 addrspa
 
 ; GCN-LABEL: {{^}}v_pack_v2f16_imm_lo:
 ; GFX9-DAG: flat_load_dword [[VAL1:v[0-9]+]]
-; GFX9-DENORM-DAG: s_movk_i32 [[K:s[0-9]+]], 0x1234{{$}}
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[K]], [[VAL1]]
 
-; GFX9-FLUSH-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x1234{{$}}
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[K]]
+; GFX9-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x1234{{$}}
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[K]]
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2f16_imm_lo(i32 addrspace(1)* %in1) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -134,10 +129,9 @@ define amdgpu_kernel void @v_pack_v2f16_imm_lo(i32 addrspace(1)* %in1) #0 {
 
 ; GCN-LABEL: {{^}}v_pack_v2f16_inline_imm_lo:
 ; GFX9-DAG: flat_load_dword [[VAL1:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], 4.0, [[VAL1]]
 
-; GFX9-FLUSH-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x4400{{$}}
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[K]]
+; GFX9-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x4400{{$}}
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[K]]
 
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2f16_inline_imm_lo(i32 addrspace(1)* %in1) #0 {
@@ -156,12 +150,10 @@ define amdgpu_kernel void @v_pack_v2f16_inline_imm_lo(i32 addrspace(1)* %in1) #0
 
 ; GCN-LABEL: {{^}}v_pack_v2f16_imm_hi:
 ; GFX9-DAG: flat_load_dword [[VAL0:v[0-9]+]]
-; GFX9-DENORM-DAG: s_movk_i32 [[K:s[0-9]+]], 0x1234
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL0]], [[K]]
 
-; GFX9-FLUSH-DAG: s_movk_i32 [[K:s[0-9]+]], 0x1234
-; GFX9-FLUSH: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL0]]
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[K]], 16, [[MASKED]]
+; GFX9-DAG: s_movk_i32 [[K:s[0-9]+]], 0x1234
+; GFX9: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL0]]
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[K]], 16, [[MASKED]]
 
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2f16_imm_hi(i32 addrspace(1)* %in0) #0 {
@@ -180,11 +172,10 @@ define amdgpu_kernel void @v_pack_v2f16_imm_hi(i32 addrspace(1)* %in0) #0 {
 
 ; GCN-LABEL: {{^}}v_pack_v2f16_inline_f16imm_hi:
 ; GFX9-DAG: flat_load_dword [[VAL:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL]], 1.0
 
-; GFX9-FLUSH-DAG: s_movk_i32 [[K:s[0-9]+]], 0x3c00
-; GFX9-FLUSH: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL]]
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[K]], 16, [[MASKED]]
+; GFX9-DAG: s_movk_i32 [[K:s[0-9]+]], 0x3c00
+; GFX9: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL]]
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[K]], 16, [[MASKED]]
 
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2f16_inline_f16imm_hi(i32 addrspace(1)* %in0) #0 {
@@ -203,10 +194,9 @@ define amdgpu_kernel void @v_pack_v2f16_inline_f16imm_hi(i32 addrspace(1)* %in0)
 
 ; GCN-LABEL: {{^}}v_pack_v2f16_inline_imm_hi:
 ; GFX9: flat_load_dword [[VAL:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL]], 64
 
-; GFX9-FLUSH: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL]]
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], 64, 16, [[MASKED]]
+; GFX9: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL]]
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], 64, 16, [[MASKED]]
 
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2f16_inline_imm_hi(i32 addrspace(1)* %in0) #0 {
diff --git a/llvm/test/CodeGen/AMDGPU/pack.v2i16.ll b/llvm/test/CodeGen/AMDGPU/pack.v2i16.ll
index 9ffd16754a1..8515fbc6dba 100644
--- a/llvm/test/CodeGen/AMDGPU/pack.v2i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/pack.v2i16.ll
@@ -1,5 +1,4 @@
-; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx901 -mattr=-flat-for-global,+fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=GFX9-DENORM %s
-; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx901 -mattr=-flat-for-global,-fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=GFX9-FLUSH %s
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx901 -mattr=-flat-for-global,-fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 %s
 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI %s
 
@@ -55,10 +54,9 @@ define amdgpu_kernel void @s_pack_v2i16_imm_hi(i32 addrspace(2)* %in0) #0 {
 ; GCN-LABEL: {{^}}v_pack_v2i16:
 ; GFX9: flat_load_dword [[VAL0:v[0-9]+]]
 ; GFX9: flat_load_dword [[VAL1:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL0]], [[VAL1]]
 
-; GFX9-FLUSH: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL0]]
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[MASKED]]
+; GFX9: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL0]]
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[MASKED]]
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2i16(i32 addrspace(1)* %in0, i32 addrspace(1)* %in1) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -79,10 +77,9 @@ define amdgpu_kernel void @v_pack_v2i16(i32 addrspace(1)* %in0, i32 addrspace(1)
 ; GCN-LABEL: {{^}}v_pack_v2i16_user:
 ; GFX9: flat_load_dword [[VAL0:v[0-9]+]]
 ; GFX9: flat_load_dword [[VAL1:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL0]], [[VAL1]]
 
-; GFX9-FLUSH: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL0]]
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[MASKED]]
+; GFX9: v_and_b32_e32 [[MASKED:v[0-9]+]], 0xffff, [[VAL0]]
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[MASKED]]
 
 ; GFX9: v_add_i32_e32 v{{[0-9]+}}, vcc, 9, [[PACKED]]
 define amdgpu_kernel void @v_pack_v2i16_user(i32 addrspace(1)* %in0, i32 addrspace(1)* %in1) #0 {
@@ -105,10 +102,9 @@ define amdgpu_kernel void @v_pack_v2i16_user(i32 addrspace(1)* %in0, i32 addrspa
 ; GCN-LABEL: {{^}}v_pack_v2i16_imm_lo:
 ; GFX9-DAG: flat_load_dword [[VAL1:v[0-9]+]]
 ; GFX9-DENORM-DAG: s_movk_i32 [[K:s[0-9]+]], 0x7b{{$}}
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[K]], [[VAL1]]
 
-; GFX9-FLUSH-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x7b{{$}}
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[K]]
+; GFX9-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x7b{{$}}
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, [[K]]
 
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2i16_imm_lo(i32 addrspace(1)* %in1) #0 {
@@ -126,9 +122,8 @@ define amdgpu_kernel void @v_pack_v2i16_imm_lo(i32 addrspace(1)* %in1) #0 {
 
 ; GCN-LABEL: {{^}}v_pack_v2i16_inline_imm_lo:
 ; GFX9: flat_load_dword [[VAL1:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], 64, [[VAL1]]
 
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, 64
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[VAL1]], 16, 64
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2i16_inline_imm_lo(i32 addrspace(1)* %in1) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -145,10 +140,9 @@ define amdgpu_kernel void @v_pack_v2i16_inline_imm_lo(i32 addrspace(1)* %in1) #0
 
 ; GCN-LABEL: {{^}}v_pack_v2i16_imm_hi:
 ; GFX9-DAG: flat_load_dword [[VAL0:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL0]], [[K]]
 
-; GFX9-FLUSH-DAG: s_movk_i32 [[K:s[0-9]+]], 0x7b{{$}}
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[K]], 16, [[VAL0]]
+; GFX9-DAG: s_movk_i32 [[K:s[0-9]+]], 0x7b{{$}}
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], [[K]], 16, [[VAL0]]
 
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2i16_imm_hi(i32 addrspace(1)* %in0) #0 {
@@ -166,8 +160,7 @@ define amdgpu_kernel void @v_pack_v2i16_imm_hi(i32 addrspace(1)* %in0) #0 {
 
 ; GCN-LABEL: {{^}}v_pack_v2i16_inline_imm_hi:
 ; GFX9: flat_load_dword [[VAL:v[0-9]+]]
-; GFX9-DENORM: v_pack_b32_f16 [[PACKED:v[0-9]+]], [[VAL]], 7
-; GFX9-FLUSH: v_lshl_or_b32 [[PACKED:v[0-9]+]], 7, 16, [[VAL0]]
+; GFX9: v_lshl_or_b32 [[PACKED:v[0-9]+]], 7, 16, [[VAL0]]
 ; GFX9: ; use [[PACKED]]
 define amdgpu_kernel void @v_pack_v2i16_inline_imm_hi(i32 addrspace(1)* %in0) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()