diff --git a/lib/Target/AMDGPU/SIInstructions.td b/lib/Target/AMDGPU/SIInstructions.td
index 7744f7ae6fd..1ed5e8e0937 100644
--- a/lib/Target/AMDGPU/SIInstructions.td
+++ b/lib/Target/AMDGPU/SIInstructions.td
@@ -1283,6 +1283,11 @@ def : Pat<
   (fcanonicalize (f16 (VOP3Mods f16:$src, i32:$src_mods))),
   (V_MUL_F16_e64 0, (i32 CONST.FP16_ONE), $src_mods, $src, 0, 0)
 >;
+
+def : Pat<
+  (fcanonicalize (v2f16 (VOP3PMods v2f16:$src, i32:$src_mods))),
+  (V_PK_MUL_F16 0, (i32 CONST.V2FP16_ONE), $src_mods, $src, DSTCLAMP.NONE)
+>;
 }
 
 let Predicates = [FP16Denormals] in {
@@ -1290,6 +1295,11 @@ def : Pat<
   (fcanonicalize (f16 (VOP3Mods f16:$src, i32:$src_mods))),
   (V_MAX_F16_e64 $src_mods, $src, $src_mods, $src, 0, 0)
 >;
+
+def : Pat<
+  (fcanonicalize (v2f16 (VOP3PMods v2f16:$src, i32:$src_mods))),
+  (V_PK_MAX_F16 $src_mods, $src, $src_mods, $src, DSTCLAMP.NONE)
+>;
 }
 
 let Predicates = [NoFP32Denormals] in {
@@ -1320,11 +1330,6 @@ def : Pat<
 >;
 }
 
-def : Pat<
-  (fcanonicalize (v2f16 (VOP3PMods v2f16:$src, i32:$src_mods))),
-  (V_PK_MUL_F16 0, (i32 CONST.V2FP16_ONE), $src_mods, $src, DSTCLAMP.NONE)
->;
-
 
 // Allow integer inputs
 class ExpPattern<SDPatternOperator node, ValueType vt, Instruction Inst> : Pat<
diff --git a/test/CodeGen/AMDGPU/fcanonicalize.f16.ll b/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
index 1dfd2fe1522..80f802bdce5 100644
--- a/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
+++ b/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
@@ -211,7 +211,7 @@ define amdgpu_kernel void @test_fold_canonicalize_snan3_value_f16(half addrspace
 ; VI-DAG: v_max_f16_e32 [[REG1:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}
 ; VI-NOT: v_and_b32
 
-; GFX9: v_pk_mul_f16 [[REG:v[0-9]+]], 1.0, {{v[0-9]+}} op_sel_hi:[0,1]{{$}}
+; GFX9: v_pk_max_f16 [[REG:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+$}}
 ; GFX9: buffer_store_dword [[REG]]
 define amdgpu_kernel void @v_test_canonicalize_var_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -232,7 +232,7 @@ define amdgpu_kernel void @v_test_canonicalize_var_v2f16(<2 x half> addrspace(1)
 ; VI: v_or_b32
 
 ; GFX9: v_and_b32_e32 [[ABS:v[0-9]+]], 0x7fff7fff, v{{[0-9]+}}
-; GFX9: v_pk_mul_f16 [[REG:v[0-9]+]], 1.0, [[ABS]] op_sel_hi:[0,1]{{$}}
+; GFX9: v_pk_max_f16 [[REG:v[0-9]+]], [[ABS]], [[ABS]]{{$}}
 ; GCN: buffer_store_dword
 define amdgpu_kernel void @v_test_canonicalize_fabs_var_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -251,7 +251,7 @@ define amdgpu_kernel void @v_test_canonicalize_fabs_var_v2f16(<2 x half> addrspa
 ; VI: v_or_b32
 
 ; GFX9: v_and_b32_e32 [[ABS:v[0-9]+]], 0x7fff7fff, v{{[0-9]+}}
-; GFX9: v_pk_mul_f16 [[REG:v[0-9]+]], 1.0, [[ABS]] op_sel_hi:[0,1] neg_lo:[0,1] neg_hi:[0,1]{{$}}
+; GFX9: v_pk_max_f16 [[REG:v[0-9]+]], [[ABS]], [[ABS]] neg_lo:[1,1] neg_hi:[1,1]{{$}}
 ; GCN: buffer_store_dword
 define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -271,7 +271,7 @@ define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_v2f16(<2 x half> ad
 ; VI-DAG: v_max_f16_e32 [[REG0:v[0-9]+]], [[FNEG]], [[FNEG]]
 ; VI-NOT: 0xffff
 
-; GFX9: v_pk_mul_f16 [[REG:v[0-9]+]], 1.0, {{v[0-9]+}} op_sel_hi:[0,1] neg_lo:[0,1] neg_hi:[0,1]{{$}}
+; GFX9: v_pk_max_f16 [[REG:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}} neg_lo:[1,1] neg_hi:[1,1]{{$}}
 ; GFX9: buffer_store_dword [[REG]]
 define amdgpu_kernel void @v_test_canonicalize_fneg_var_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -288,7 +288,7 @@ define amdgpu_kernel void @v_test_canonicalize_fneg_var_v2f16(<2 x half> addrspa
 ; VI: v_max_f16_e64 [[REG1:v[0-9]+]], {{s[0-9]+}}, {{s[0-9]+}}
 ; VI-NOT: v_and_b32
 
-; GFX9: v_pk_mul_f16 [[REG:v[0-9]+]], 1.0, {{s[0-9]+}} op_sel_hi:[0,1]{{$}}
+; GFX9: v_pk_max_f16 [[REG:v[0-9]+]], {{s[0-9]+}}, {{s[0-9]+$}}
 ; GFX9: buffer_store_dword [[REG]]
 define amdgpu_kernel void @s_test_canonicalize_var_v2f16(<2 x half> addrspace(1)* %out, i32 zeroext %val.arg) #1 {
   %val = bitcast i32 %val.arg to <2 x half>
diff --git a/test/CodeGen/AMDGPU/fcanonicalize.ll b/test/CodeGen/AMDGPU/fcanonicalize.ll
index 028766ebd60..c0ac6599e4a 100644
--- a/test/CodeGen/AMDGPU/fcanonicalize.ll
+++ b/test/CodeGen/AMDGPU/fcanonicalize.ll
@@ -5,6 +5,7 @@ declare float @llvm.canonicalize.f32(float) #0
 declare double @llvm.fabs.f64(double) #0
 declare double @llvm.canonicalize.f64(double) #0
 declare half @llvm.canonicalize.f16(half) #0
+declare <2 x half> @llvm.canonicalize.v2f16(<2 x half>) #0
 declare i32 @llvm.amdgcn.workitem.id.x() #0
 
 ; GCN-LABEL: {{^}}v_test_canonicalize_var_f32:
@@ -453,6 +454,32 @@ define amdgpu_kernel void @test_canonicalize_value_f16_flush(half addrspace(1)*
   ret void
 }
 
+; GCN-LABEL:  {{^}}test_canonicalize_value_v2f16_flush_gfx8:
+; GCN:     v_mov_b32_e32 [[ONE:v[0-9]+]], 0x3c00
+; GCN-DAG: v_mul_f16_sdwa v{{[0-9]+}}, [[ONE]], v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GCN-DAG: v_mul_f16_e32 v{{[0-9]+}}, 1.0, v{{[0-9]+}}
+define amdgpu_kernel void @test_canonicalize_value_v2f16_flush_gfx8(<2 x half> addrspace(1)* %arg, <2 x half> addrspace(1)* %out) #4 {
+  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
+  %gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %arg, i32 %id
+  %v = load <2 x half>, <2 x half> addrspace(1)* %gep, align 4
+  %canonicalized = tail call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %v)
+  %gep2 = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i32 %id
+  store <2 x half> %canonicalized, <2 x half> addrspace(1)* %gep2, align 2
+  ret void
+}
+
+; GCN-LABEL:  {{^}}test_canonicalize_value_v2f16_flush_gfx9:
+; GCN-DAG: v_pk_mul_f16 v{{[0-9]+}}, 1.0, v{{[0-9]+}}
+define amdgpu_kernel void @test_canonicalize_value_v2f16_flush_gfx9(<2 x half> addrspace(1)* %arg, <2 x half> addrspace(1)* %out) #6 {
+  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
+  %gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %arg, i32 %id
+  %v = load <2 x half>, <2 x half> addrspace(1)* %gep, align 4
+  %canonicalized = tail call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %v)
+  %gep2 = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i32 %id
+  store <2 x half> %canonicalized, <2 x half> addrspace(1)* %gep2, align 2
+  ret void
+}
+
 ; GCN-LABEL:  {{^}}test_canonicalize_value_f64_denorm:
 ; GCN: v_max_f64 v[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}]
 define amdgpu_kernel void @test_canonicalize_value_f64_denorm(double addrspace(1)* %arg, double addrspace(1)* %out) #5 {
@@ -489,9 +516,22 @@ define amdgpu_kernel void @test_canonicalize_value_f16_denorm(half addrspace(1)*
   ret void
 }
 
+; GCN-LABEL:  {{^}}test_canonicalize_value_v2f16_denorm:
+; GCN: v_pk_max_f16 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
+define amdgpu_kernel void @test_canonicalize_value_v2f16_denorm(<2 x half> addrspace(1)* %arg, <2 x half> addrspace(1)* %out) #5 {
+  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
+  %gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %arg, i32 %id
+  %v = load <2 x half>, <2 x half> addrspace(1)* %gep, align 4
+  %canonicalized = tail call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %v)
+  %gep2 = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i32 %id
+  store <2 x half> %canonicalized, <2 x half> addrspace(1)* %gep2, align 2
+  ret void
+}
+
 attributes #0 = { nounwind readnone }
 attributes #1 = { nounwind }
 attributes #2 = { nounwind "target-features"="-fp32-denormals,-fp64-fp16-denormals" }
 attributes #3 = { nounwind "target-features"="+fp32-denormals,+fp64-fp16-denormals" }
 attributes #4 = { nounwind "target-features"="-fp32-denormals,-fp64-fp16-denormals" "target-cpu"="tonga" }
 attributes #5 = { nounwind "target-features"="+fp32-denormals,+fp64-fp16-denormals" "target-cpu"="gfx900" }
+attributes #6 = { nounwind "target-features"="-fp32-denormals,-fp64-fp16-denormals" "target-cpu"="gfx900" }