AMDGPU: Fixes for 'LOD bias' operand in ISelDAG path and GobalISel path when A16-bit is 'ON'

[Ravi]

Summary

Background: https://reviews.llvm.org/D74314

The LOD bias operand is of type 'half' when the A16-bit is ON' for MIMG instructions. 'bias' is only 16-bit but occupies 32-bits with upper 16-bits containing junk. The patch fixes both the paths(ISelDAG and GlobalISel) for proper encoding of LOD bias operand. The fix could have been done in one of 2 approaches.

Approaches:

1. First approach is to add 'bias' operand index for each MIMG instruction in the table-gen generated image intrinsic info to later identify the 'bias' operand with it and fix it as 2 packed 16-bit operands with the upper 16-bit being undefined.
2. The 'bias' operand is the only operand in the MIMG intrinsics that would be a 16-bit incoming operand below the index for gradients. The other image address operands 'offset' and 'z-compare' that come before gradients are always 32-bit irrespective of the A16-bit. The patch implements this logic.

Testing:
Multiple tests with A16 ON and OFF in both GlobalISel and ISelDAG path are checked and updated. Especially the SAMPLE and GATHER4 instruction tests. Few tests with
SAMPLE were missing in both the paths with A16 'OFF'. But the GATHER4 tests have covered this case. So no additional tests are added.
All the lit tests have passed.

Observations:

1. The ISelDAG path generates the same code as earlier without any inefficiency. But the GlobalISel path adds explicit instructions to fill the upper 16-bit with junk. This has to be probably analysed as a new optimization issue and identify the path that's introducing them.
2. Occuring with earlier code as well as with theis patch. The image resource constant(a group of 8 registers) are being copied from set of contiguous registers to another set of contiguous registers to take care of alignment. These copies can be avoided with a custom lowering of formal arguments. And the specific register info could be reported back to the driver/encoded in the code objects.

[sebastian-ne]

This also needs fixing in the combiner (which was introduced in D85887). I.e. checking that the bias can be losslessly converted to f16 and converting it when converting the address.
(I guess this doesn’t need to be part of this change.)

[arsenm]

LGTM although I'm not the most familiar person with images

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

4242

Capitalize

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

6286

Capitalize

[foad]

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.sample.a16.ll

9

Why has the indentation changed in almost every test?

[foad]

But the GlobalISel path adds explicit instructions to fill the upper 16-bit with junk.

I think D112064 should clean this up.

[foad]

@Ravi are you still working on this?

[Ravi]

In D111754#3137210, @foad wrote:

@Ravi are you still working on this?

I don't have commit permissions, so checked with @cdevadas and we decided to check-in after D112064 goes in. The suggestions from @arsenm are taken care and the indent changes have also been removed.

[foad]

we decided to check-in after D112064 goes in.

OK. I have updated D112064.

[Ravi]

fixed comments and rebased with latest main.

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

if (!SrcOp.isReg())
       continue; // _L to _LZ may have eliminated this.
 
     Register AddrReg = SrcOp.getReg();
 
     if ((I < Intr->GradientStart) ||
         (I >= Intr->GradientStart && I < Intr->CoordStart && !IsG16) ||
         (I >= Intr->CoordStart && !IsA16)) {
       // Handle any gradient or coordinate operands that should not be packed
+      if ((I < Intr->GradientStart) && IsA16 &&
+          (B.getMRI()->getType(AddrReg) == S16)) {
+        // Special handling of bias when A16 is on. Bias is of type half but
+        // occupies full 32-bit.
+        PackedAddrs.push_back(
+            B.buildBuildVector(V2S16, {AddrReg, B.buildUndef(S16).getReg(0)})
+                .getReg(0));
+      } else {
         AddrReg = B.buildBitcast(V2S16, AddrReg).getReg(0);
         PackedAddrs.push_back(AddrReg);
+      }
     } else {
       // Dz/dh, dz/dv and the last odd coord are packed with undef. Also, in 1D,
       // derivatives dx/dh and dx/dv are packed with undef.
       if (((I + 1) >= EndIdx) ||
           ((Intr->NumGradients / 2) % 2 == 1 &&
            (I == static_cast<unsigned>(Intr->GradientStart +
                                        (Intr->NumGradients / 2) - 1) ||
             I == static_cast<unsigned>(Intr->GradientStart +

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

if (auto *ConstantLod = dyn_cast<ConstantSDNode>(
             Op.getOperand(ArgOffset + Intr->MipIndex))) {
       if (ConstantLod->isZero()) {
         IntrOpcode = MIPMappingInfo->NONMIP;  // set new opcode to variant without _mip
         VAddrEnd--;                           // remove 'mip'
       }
     }
   }
 
-  // Push back extra arguments.
-  for (unsigned I = Intr->VAddrStart; I < Intr->GradientStart; I++)
-    VAddrs.push_back(Op.getOperand(ArgOffset + I));
-
   // Check for 16 bit addresses or derivatives and pack if true.
   MVT VAddrVT =
       Op.getOperand(ArgOffset + Intr->GradientStart).getSimpleValueType();
   MVT VAddrScalarVT = VAddrVT.getScalarType();
   MVT GradPackVectorVT = VAddrScalarVT == MVT::f16 ? MVT::v2f16 : MVT::v2i16;
   IsG16 = VAddrScalarVT == MVT::f16 || VAddrScalarVT == MVT::i16;
 
   VAddrVT = Op.getOperand(ArgOffset + Intr->CoordStart).getSimpleValueType();
   VAddrScalarVT = VAddrVT.getScalarType();
   MVT AddrPackVectorVT = VAddrScalarVT == MVT::f16 ? MVT::v2f16 : MVT::v2i16;
   IsA16 = VAddrScalarVT == MVT::f16 || VAddrScalarVT == MVT::i16;
 
+  // Push back extra arguments.
+  for (unsigned I = Intr->VAddrStart; I < Intr->GradientStart; I++) {
+    if (IsA16 && (Op.getOperand(ArgOffset + I).getValueType() == MVT::f16)) {
+      // Special handling of bias when A16 is on. Bias is of type half but
+      // occupies full 32-bit.
+      SDValue bias = DAG.getBuildVector( MVT::v2f16, DL, {Op.getOperand(ArgOffset + I), DAG.getUNDEF(MVT::f16)});
+      VAddrs.push_back(bias);
+    } else
+      VAddrs.push_back(Op.getOperand(ArgOffset + I));
+  }
+
   if (BaseOpcode->Gradients && !ST->hasG16() && (IsA16 != IsG16)) {
     // 16 bit gradients are supported, but are tied to the A16 control
     // so both gradients and addresses must be 16 bit
     LLVM_DEBUG(
         dbgs() << "Failed to lower image intrinsic: 16 bit addresses "
                   "require 16 bit args for both gradients and addresses");
     return Op;
   }

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.sample.a16.ll

 ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX9 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX10 %s
 
 define amdgpu_ps <4 x float> @sample_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s) {
   ; GFX9-LABEL: name: sample_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
define amdgpu_ps <4 x float> @sample_c_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t, half %clamp) {
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s) {
+define amdgpu_ps <4 x float> @sample_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s) {
   ; GFX9-LABEL: name: sample_b_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
-  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
-  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
-  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: sample_b_1d
define amdgpu_ps <4 x float> @sample_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s) {
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
-  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
-  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
-  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f32.f16(i32 15, float %bias, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f16.f16(i32 15, half %bias, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t) {
+define amdgpu_ps <4 x float> @sample_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t) {
   ; GFX9-LABEL: name: sample_b_2d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
-  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: sample_b_2d
define amdgpu_ps <4 x float> @sample_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t) {
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
-  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f32.f16(i32 15, float %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f16.f16(i32 15, half %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_c_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s) {
+define amdgpu_ps <4 x float> @sample_c_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s) {
   ; GFX9-LABEL: name: sample_c_b_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
-  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
-  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
-  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.1d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: sample_c_b_1d
define amdgpu_ps <4 x float> @sample_c_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s) {
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
-  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
-  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.1d), 15, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t) {
+define amdgpu_ps <4 x float> @sample_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t) {
   ; GFX9-LABEL: name: sample_c_b_2d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
-  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
-  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
-  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.2d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: sample_c_b_2d
define amdgpu_ps <4 x float> @sample_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t) {
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
-  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
-  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.2d), 15, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %clamp) {
+define amdgpu_ps <4 x float> @sample_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %clamp) {
   ; GFX9-LABEL: name: sample_b_cl_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
-  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: sample_b_cl_1d
define amdgpu_ps <4 x float> @sample_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %clamp) {
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
-  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f32.f16(i32 15, float %bias, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f16.f16(i32 15, half %bias, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t, half %clamp) {
+define amdgpu_ps <4 x float> @sample_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t, half %clamp) {
   ; GFX9-LABEL: name: sample_b_cl_2d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
-  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
-  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
-  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
   ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.2d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: sample_b_cl_2d
define amdgpu_ps <4 x float> @sample_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t, half %clamp) {
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
-  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
-  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f32.f16(i32 15, float %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f16.f16(i32 15, half %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_c_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %clamp) {
+define amdgpu_ps <4 x float> @sample_c_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %clamp) {
   ; GFX9-LABEL: name: sample_c_b_cl_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
-  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
-  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
-  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
   ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.1d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: sample_c_b_cl_1d
define amdgpu_ps <4 x float> @sample_c_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %clamp) {
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
-  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
-  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.1d), 15, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t, half %clamp) {
+define amdgpu_ps <4 x float> @sample_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t, half %clamp) {
   ; GFX9-LABEL: name: sample_c_b_cl_2d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
-  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
-  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
-  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
-  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
   ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.2d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: sample_c_b_cl_2d
define amdgpu_ps <4 x float> @sample_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t, half %clamp) {
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
-  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
-  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
-  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
-  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load (<4 x s32>) from custom "ImageResource")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, half %s) {
   ; GFX9-LABEL: name: sample_d_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2

 
 declare <4 x float> @llvm.amdgcn.image.sample.c.1d.v4f32.f16(i32, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.2d.v4f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.cl.1d.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.cl.2d.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.cl.1d.v4f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
-declare <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f32.f16(i32, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f32.f16(i32, float, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f32.f16(i32, float, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f32.f16(i32, float, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f32.f16(i32, float, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f16.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f16.f16(i32, half, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f16.f16(i32, half, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f16.f16(i32, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f16.f16(i32, half, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f16.f16(i32, half, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
 declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f16(i32, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f16(i32, half, half, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f32.f16(i32, float, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f16(i32, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f16(i32, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.image.gather4.a16.dim.ll

 ; GFX10NSA-NEXT:    image_gather4_c_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10NSA-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10NSA-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.cl.2d.v4f32.f16(i32 1, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @gather4_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t) {
+define amdgpu_ps <4 x float> @gather4_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t) {
 ; GFX9-LABEL: gather4_b_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[14:15], exec
 ; GFX9-NEXT:    s_mov_b32 s0, s2
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
+; GFX9-NEXT:    s_mov_b32 s2, s4
+; GFX9-NEXT:    s_mov_b32 s4, s6
+; GFX9-NEXT:    s_mov_b32 s6, s8
+; GFX9-NEXT:    s_mov_b32 s8, s10
+; GFX9-NEXT:    s_mov_b32 s10, s12
 ; GFX9-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s12, s0, 16
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX9-NEXT:    s_mov_b32 s1, s3
-; GFX9-NEXT:    s_mov_b32 s2, s4
 ; GFX9-NEXT:    s_mov_b32 s3, s5
-; GFX9-NEXT:    s_mov_b32 s4, s6
 ; GFX9-NEXT:    s_mov_b32 s5, s7
-; GFX9-NEXT:    s_mov_b32 s6, s8
 ; GFX9-NEXT:    s_mov_b32 s7, s9
-; GFX9-NEXT:    s_mov_b32 s8, s10
 ; GFX9-NEXT:    s_mov_b32 s9, s11
-; GFX9-NEXT:    s_mov_b32 s10, s12
 ; GFX9-NEXT:    s_mov_b32 s11, s13
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v3, s12
 ; GFX9-NEXT:    v_and_or_b32 v1, v1, v3, v2
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[14:15]
 ; GFX9-NEXT:    image_gather4_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10NSA-LABEL: gather4_b_2d:
 ; GFX10NSA:       ; %bb.0: ; %main_body
 ; GFX10NSA-NEXT:    s_mov_b32 s14, exec_lo
 ; GFX10NSA-NEXT:    s_mov_b32 s0, s2
 ; GFX10NSA-NEXT:    s_wqm_b32 exec_lo, exec_lo
+; GFX10NSA-NEXT:    v_mov_b32_e32 v3, 0xffff
 ; GFX10NSA-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
-; GFX10NSA-NEXT:    s_mov_b32 s1, s3
 ; GFX10NSA-NEXT:    s_mov_b32 s2, s4
-; GFX10NSA-NEXT:    s_mov_b32 s3, s5
 ; GFX10NSA-NEXT:    s_mov_b32 s4, s6
-; GFX10NSA-NEXT:    s_mov_b32 s5, s7
 ; GFX10NSA-NEXT:    s_mov_b32 s6, s8
-; GFX10NSA-NEXT:    s_mov_b32 s7, s9
 ; GFX10NSA-NEXT:    s_mov_b32 s8, s10
-; GFX10NSA-NEXT:    s_mov_b32 s9, s11
 ; GFX10NSA-NEXT:    s_mov_b32 s10, s12
+; GFX10NSA-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10NSA-NEXT:    s_mov_b32 s1, s3
+; GFX10NSA-NEXT:    s_mov_b32 s3, s5
+; GFX10NSA-NEXT:    s_mov_b32 s5, s7
+; GFX10NSA-NEXT:    s_mov_b32 s7, s9
+; GFX10NSA-NEXT:    s_mov_b32 s9, s11
 ; GFX10NSA-NEXT:    s_mov_b32 s11, s13
-; GFX10NSA-NEXT:    v_and_or_b32 v1, 0xffff, v1, v2
+; GFX10NSA-NEXT:    v_and_or_b32 v0, v0, v3, s12
+; GFX10NSA-NEXT:    v_and_or_b32 v1, v1, v3, v2
 ; GFX10NSA-NEXT:    s_and_b32 exec_lo, exec_lo, s14
 ; GFX10NSA-NEXT:    image_gather4_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10NSA-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10NSA-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f32.f16(i32 1, float %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f16.f16(i32 1, half %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @gather4_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t) {
+define amdgpu_ps <4 x float> @gather4_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: gather4_c_b_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[14:15], exec
 ; GFX9-NEXT:    s_mov_b32 s0, s2
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
+; GFX9-NEXT:    s_mov_b32 s2, s4
+; GFX9-NEXT:    s_mov_b32 s4, s6
+; GFX9-NEXT:    s_mov_b32 s6, s8
+; GFX9-NEXT:    s_mov_b32 s8, s10
+; GFX9-NEXT:    s_mov_b32 s10, s12
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s12, s0, 16
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; GFX9-NEXT:    s_mov_b32 s1, s3
-; GFX9-NEXT:    s_mov_b32 s2, s4
 ; GFX9-NEXT:    s_mov_b32 s3, s5
-; GFX9-NEXT:    s_mov_b32 s4, s6
 ; GFX9-NEXT:    s_mov_b32 s5, s7
-; GFX9-NEXT:    s_mov_b32 s6, s8
 ; GFX9-NEXT:    s_mov_b32 s7, s9
-; GFX9-NEXT:    s_mov_b32 s8, s10
 ; GFX9-NEXT:    s_mov_b32 s9, s11
-; GFX9-NEXT:    s_mov_b32 s10, s12
 ; GFX9-NEXT:    s_mov_b32 s11, s13
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v4, s12
 ; GFX9-NEXT:    v_and_or_b32 v2, v2, v4, v3
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[14:15]
 ; GFX9-NEXT:    image_gather4_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10NSA-LABEL: gather4_c_b_2d:
 ; GFX10NSA:       ; %bb.0: ; %main_body
 ; GFX10NSA-NEXT:    s_mov_b32 s14, exec_lo
 ; GFX10NSA-NEXT:    s_mov_b32 s0, s2
 ; GFX10NSA-NEXT:    s_wqm_b32 exec_lo, exec_lo
+; GFX10NSA-NEXT:    v_mov_b32_e32 v4, 0xffff
 ; GFX10NSA-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
-; GFX10NSA-NEXT:    s_mov_b32 s1, s3
 ; GFX10NSA-NEXT:    s_mov_b32 s2, s4
-; GFX10NSA-NEXT:    s_mov_b32 s3, s5
 ; GFX10NSA-NEXT:    s_mov_b32 s4, s6
-; GFX10NSA-NEXT:    s_mov_b32 s5, s7
 ; GFX10NSA-NEXT:    s_mov_b32 s6, s8
-; GFX10NSA-NEXT:    s_mov_b32 s7, s9
 ; GFX10NSA-NEXT:    s_mov_b32 s8, s10
-; GFX10NSA-NEXT:    s_mov_b32 s9, s11
 ; GFX10NSA-NEXT:    s_mov_b32 s10, s12
+; GFX10NSA-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10NSA-NEXT:    s_mov_b32 s1, s3
+; GFX10NSA-NEXT:    s_mov_b32 s3, s5
+; GFX10NSA-NEXT:    s_mov_b32 s5, s7
+; GFX10NSA-NEXT:    s_mov_b32 s7, s9
+; GFX10NSA-NEXT:    s_mov_b32 s9, s11
 ; GFX10NSA-NEXT:    s_mov_b32 s11, s13
-; GFX10NSA-NEXT:    v_and_or_b32 v2, 0xffff, v2, v3
+; GFX10NSA-NEXT:    v_and_or_b32 v0, v0, v4, s12
+; GFX10NSA-NEXT:    v_and_or_b32 v2, v2, v4, v3
 ; GFX10NSA-NEXT:    s_and_b32 exec_lo, exec_lo, s14
 ; GFX10NSA-NEXT:    image_gather4_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10NSA-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10NSA-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f32.f16(i32 1, float %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f16.f16(i32 1, half %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @gather4_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t, half %clamp) {
+define amdgpu_ps <4 x float> @gather4_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: gather4_b_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[14:15], exec
 ; GFX9-NEXT:    s_mov_b32 s0, s2
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    s_mov_b32 s2, s4
 ; GFX9-NEXT:    s_mov_b32 s4, s6
 ; GFX9-NEXT:    s_mov_b32 s6, s8
 ; GFX9-NEXT:    s_mov_b32 s8, s10
 ; GFX9-NEXT:    s_mov_b32 s10, s12
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0xffff
-; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX9-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX9-NEXT:    s_mov_b32 s1, s3
 ; GFX9-NEXT:    s_mov_b32 s3, s5
 ; GFX9-NEXT:    s_mov_b32 s5, s7
 ; GFX9-NEXT:    s_mov_b32 s7, s9
 ; GFX9-NEXT:    s_mov_b32 s9, s11
 ; GFX9-NEXT:    s_mov_b32 s11, s13
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v4, s12
 ; GFX9-NEXT:    v_and_or_b32 v1, v1, v4, v2
 ; GFX9-NEXT:    v_and_or_b32 v2, v3, v4, s12
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[14:15]
 ; GFX9-NEXT:    image_gather4_b_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10NSA-LABEL: gather4_b_cl_2d:

 ; GFX10NSA-NEXT:    s_mov_b32 s10, s12
 ; GFX10NSA-NEXT:    s_lshl_b32 s12, s0, 16
 ; GFX10NSA-NEXT:    s_mov_b32 s1, s3
 ; GFX10NSA-NEXT:    s_mov_b32 s3, s5
 ; GFX10NSA-NEXT:    s_mov_b32 s5, s7
 ; GFX10NSA-NEXT:    s_mov_b32 s7, s9
 ; GFX10NSA-NEXT:    s_mov_b32 s9, s11
 ; GFX10NSA-NEXT:    s_mov_b32 s11, s13
+; GFX10NSA-NEXT:    v_and_or_b32 v0, v0, v4, s12
 ; GFX10NSA-NEXT:    v_and_or_b32 v1, v1, v4, v2
 ; GFX10NSA-NEXT:    v_and_or_b32 v2, v3, v4, s12
 ; GFX10NSA-NEXT:    s_and_b32 exec_lo, exec_lo, s14
 ; GFX10NSA-NEXT:    image_gather4_b_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10NSA-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10NSA-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f32.f16(i32 1, float %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f16.f16(i32 1, half %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @gather4_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t, half %clamp) {
+define amdgpu_ps <4 x float> @gather4_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: gather4_c_b_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[14:15], exec
 ; GFX9-NEXT:    s_mov_b32 s0, s2
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    s_mov_b32 s2, s4
 ; GFX9-NEXT:    s_mov_b32 s4, s6
 ; GFX9-NEXT:    s_mov_b32 s6, s8
 ; GFX9-NEXT:    s_mov_b32 s8, s10
 ; GFX9-NEXT:    s_mov_b32 s10, s12
 ; GFX9-NEXT:    v_mov_b32_e32 v5, 0xffff
-; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; GFX9-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; GFX9-NEXT:    s_mov_b32 s1, s3
 ; GFX9-NEXT:    s_mov_b32 s3, s5
 ; GFX9-NEXT:    s_mov_b32 s5, s7
 ; GFX9-NEXT:    s_mov_b32 s7, s9
 ; GFX9-NEXT:    s_mov_b32 s9, s11
 ; GFX9-NEXT:    s_mov_b32 s11, s13
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v5, s12
 ; GFX9-NEXT:    v_and_or_b32 v2, v2, v5, v3
 ; GFX9-NEXT:    v_and_or_b32 v3, v4, v5, s12
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[14:15]
 ; GFX9-NEXT:    image_gather4_c_b_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10NSA-LABEL: gather4_c_b_cl_2d:

 ; GFX10NSA-NEXT:    s_mov_b32 s10, s12
 ; GFX10NSA-NEXT:    s_lshl_b32 s12, s0, 16
 ; GFX10NSA-NEXT:    s_mov_b32 s1, s3
 ; GFX10NSA-NEXT:    s_mov_b32 s3, s5
 ; GFX10NSA-NEXT:    s_mov_b32 s5, s7
 ; GFX10NSA-NEXT:    s_mov_b32 s7, s9
 ; GFX10NSA-NEXT:    s_mov_b32 s9, s11
 ; GFX10NSA-NEXT:    s_mov_b32 s11, s13
+; GFX10NSA-NEXT:    v_and_or_b32 v0, v0, v5, s12
 ; GFX10NSA-NEXT:    v_and_or_b32 v2, v2, v5, v3
 ; GFX10NSA-NEXT:    v_and_or_b32 v3, v4, v5, s12
 ; GFX10NSA-NEXT:    s_and_b32 exec_lo, exec_lo, s14
 ; GFX10NSA-NEXT:    image_gather4_c_b_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10NSA-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10NSA-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f32.f16(i32 1, float %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f16.f16(i32 1, half %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %lod) {
 ; GFX9-LABEL: gather4_l_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b32 s0, s2
 ; GFX9-NEXT:    s_mov_b32 s2, s4
main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.lz.2d.v4f32.f16(i32 1, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 declare <4 x float> @llvm.amdgcn.image.gather4.2d.v4f32.f16(i32 immarg, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.cube.v4f32.f16(i32 immarg, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.2darray.v4f32.f16(i32 immarg, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.cl.2d.v4f32.f16(i32 immarg, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
-declare <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f32.f16(i32 immarg, float, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
-declare <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f32.f16(i32 immarg, float, float, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
-declare <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f32.f16(i32 immarg, float, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
-declare <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f32.f16(i32 immarg, float, float, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
+declare <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f16.f16(i32 immarg, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
+declare <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f16.f16(i32 immarg, half, float, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
+declare <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f16.f16(i32 immarg, half, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
+declare <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f16.f16(i32 immarg, half, float, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.l.2d.v4f32.f16(i32 immarg, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.lz.2d.v4f32.f16(i32 immarg, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.c.2d.v4f32.f16(i32 immarg, float, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.c.cl.2d.v4f32.f16(i32 immarg, float, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.c.l.2d.v4f32.f16(i32 immarg, float, half, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.gather4.c.lz.2d.v4f32.f16(i32 immarg, float, half, half, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 
 attributes #0 = { nounwind readonly }

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.gather4.a16.dim.ll

 ; GFX10-NEXT:    image_gather4_c_cl v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.cl.2d.v4f32.f32(i32 1, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @gather4_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t) {
+define amdgpu_ps <4 x float> @gather4_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t) {
 ; GFX9-LABEL: gather4_b_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_b_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f32.f16(i32 1, float %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f16.f16(i32 1, half %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @gather4_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t) {
+define amdgpu_ps <4 x float> @gather4_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: gather4_c_b_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX9-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_c_b_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f32.f16(i32 1, float %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f16.f16(i32 1, half %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @gather4_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t, half %clamp) {
+define amdgpu_ps <4 x float> @gather4_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: gather4_b_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
 ; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
 ; GFX9-NEXT:    v_lshl_or_b32 v4, v2, 16, v0
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_b_cl v[0:3], v[3:5], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_b_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_b_cl v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f32.f16(i32 1, float %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f16.f16(i32 1, half %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @gather4_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t, half %clamp) {
+define amdgpu_ps <4 x float> @gather4_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: gather4_c_b_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_mov_b32_e32 v7, v4
 ; GFX9-NEXT:    v_mov_b32_e32 v4, v0
 ; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v1

 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_c_b_cl v[0:3], [v0, v1, v2, v4], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f32.f16(i32 1, float %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f16.f16(i32 1, half %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %lod) {
 ; GFX9-LABEL: gather4_l_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0

 declare <4 x float> @llvm.amdgcn.image.gather4.2d.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.gather4.cube.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.gather4.2darray.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
 declare <4 x float> @llvm.amdgcn.image.gather4.c.2d.v4f32.f32(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.gather4.cl.2d.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.gather4.c.cl.2d.v4f32.f32(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
-declare <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f32.f16(i32, float, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f32.f16(i32, float, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f16.f16(i32, half, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f16.f16(i32, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f16.f16(i32, half, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
 declare <4 x float> @llvm.amdgcn.image.gather4.l.2d.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.gather4.c.l.2d.v4f32.f32(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
 declare <4 x float> @llvm.amdgcn.image.gather4.lz.2d.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.gather4.c.lz.2d.v4f32.f32(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readonly }
 attributes #2 = { nounwind readnone }

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.a16.dim.ll

 ; GFX10-NEXT:    image_sample_c_cl v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s) {
+define amdgpu_ps <4 x float> @sample_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s) {
 ; GFX9-LABEL: sample_b_1d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_b_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f32.f16(i32 15, float %bias, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f16.f16(i32 15, half %bias, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t) {
+define amdgpu_ps <4 x float> @sample_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t) {
 ; GFX9-LABEL: sample_b_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_b_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f32.f16(i32 15, float %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f16.f16(i32 15, half %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_c_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s) {
+define amdgpu_ps <4 x float> @sample_c_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s) {
 ; GFX9-LABEL: sample_c_b_1d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_b_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t) {
+define amdgpu_ps <4 x float> @sample_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: sample_c_b_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX9-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_b_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %clamp) {
+define amdgpu_ps <4 x float> @sample_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %clamp) {
 ; GFX9-LABEL: sample_b_cl_1d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_b_cl v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_b_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_b_cl v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f32.f16(i32 15, float %bias, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f16.f16(i32 15, half %bias, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t, half %clamp) {
+define amdgpu_ps <4 x float> @sample_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_b_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
 ; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
 ; GFX9-NEXT:    v_lshl_or_b32 v4, v2, 16, v0
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_b_cl v[0:3], v[3:5], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_b_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_b_cl v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f32.f16(i32 15, float %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f16.f16(i32 15, half %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_c_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %clamp) {
+define amdgpu_ps <4 x float> @sample_c_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %clamp) {
 ; GFX9-LABEL: sample_c_b_cl_1d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX9-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c_b_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_b_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_b_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
-define amdgpu_ps <4 x float> @sample_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t, half %clamp) {
+define amdgpu_ps <4 x float> @sample_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_c_b_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
 ; GFX9-NEXT:    v_mov_b32_e32 v7, v4
 ; GFX9-NEXT:    v_mov_b32_e32 v4, v0
 ; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v1

 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
 ; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_b_cl v[0:3], [v0, v1, v2, v4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
-  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, half %s) {
 ; GFX9-LABEL: sample_d_1d:
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    image_sample_d v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)

 
 declare <4 x float> @llvm.amdgcn.image.sample.c.1d.v4f32.f16(i32, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.2d.v4f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.cl.1d.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.cl.2d.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.cl.1d.v4f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
-declare <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f32.f16(i32, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f32.f16(i32, float, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f32.f16(i32, float, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f32.f16(i32, float, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
-declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f32.f16(i32, float, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f16.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f16.f16(i32, half, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f16.f16(i32, half, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f16.f16(i32, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f16.f16(i32, half, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f16.f16(i32, half, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
 declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f16(i32, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f16(i32, half, half, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f32.f16(i32, float, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f16(i32, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 declare <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f16(i32, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1