[AMDGPU] Add support for point sample accel out of order returns (llvm#796)

Add target feature for point sample acceleration and enable it for
relevant targets.

Also add support to insert waitcnts where required when point sample
accel may have occurred. This has implications for out of order returns,
which is why extra waitcnts are required.

Add a VMEM_NOSAMPLER bit in the register masks to determine when waitcnt
is required.

Author: dstutt

llvm/lib/Target/AMDGPU/AMDGPU.td

@@ -1082,6 +1082,12 @@ def FeaturePrngInst : SubtargetFeature<"prng-inst",
   "Has v_prng_b32 instruction"
 >;
 
+def FeaturePointSampleAccel : SubtargetFeature<"point-sample-accel",
+  "HasPointSampleAccel",
+  "true",
+  "Has point sample acceleration feature"
+>;
+
 //===------------------------------------------------------------===//
 // Subtarget Features (options and debugging)
 //===------------------------------------------------------------===//
@@ -1769,20 +1775,23 @@ def FeatureISAVersion11_5_0 : FeatureSet<
   !listconcat(FeatureISAVersion11_Common.Features,
     [FeatureSALUFloatInsts,
      FeatureDPPSrc1SGPR,
-     FeatureRequiredExportPriority])>;
+     FeatureRequiredExportPriority,
+     FeaturePointSampleAccel])>;
 
 def FeatureISAVersion11_5_1 : FeatureSet<
   !listconcat(FeatureISAVersion11_Common.Features,
     [FeatureSALUFloatInsts,
      FeatureDPPSrc1SGPR,
      Feature1_5xVGPRs,
-     FeatureRequiredExportPriority])>;
+     FeatureRequiredExportPriority,
+     FeaturePointSampleAccel])>;
 
 def FeatureISAVersion11_5_2 : FeatureSet<
   !listconcat(FeatureISAVersion11_Common.Features,
     [FeatureSALUFloatInsts,
      FeatureDPPSrc1SGPR,
-     FeatureRequiredExportPriority])>;
+     FeatureRequiredExportPriority,
+     FeaturePointSampleAccel])>;
 
 def FeatureISAVersion11_5_3 : FeatureSet<
   !listconcat(FeatureISAVersion11_Common.Features,

llvm/lib/Target/AMDGPU/GCNSubtarget.h

@@ -252,6 +252,7 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
   bool HasMinimum3Maximum3F32 = false;
   bool HasMinimum3Maximum3F16 = false;
   bool HasMinimum3Maximum3PKF16 = false;
+  bool HasPointSampleAccel = false;
 
   bool RequiresCOV6 = false;
 
@@ -1352,6 +1353,8 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
     return HasMinimum3Maximum3PKF16;
   }
 
+  bool hasPointSampleAccel() const { return HasPointSampleAccel; }
+
   /// \returns The maximum number of instructions that can be enclosed in an
   /// S_CLAUSE on the given subtarget, or 0 for targets that do not support that
   /// instruction.

llvm/lib/Target/AMDGPU/MIMGInstructions.td

@@ -52,6 +52,7 @@ class MIMGBaseOpcode : PredicateControl {
   bit BVH = 0;
   bit A16 = 0;
   bit NoReturn = 0;
+  bit PointSampleAccel = 0; // Opcode eligible for gfx11.5 point sample acceleration
 }
 
 def MIMGBaseOpcode : GenericEnum {
@@ -63,7 +64,8 @@ def MIMGBaseOpcodesTable : GenericTable {
   let CppTypeName = "MIMGBaseOpcodeInfo";
   let Fields = ["BaseOpcode", "Store", "Atomic", "AtomicX2", "Sampler",
                 "Gather4", "NumExtraArgs", "Gradients", "G16", "Coordinates",
-                "LodOrClampOrMip", "HasD16", "MSAA", "BVH", "A16", "NoReturn"];
+                "LodOrClampOrMip", "HasD16", "MSAA", "BVH", "A16", "NoReturn",
+                "PointSampleAccel"];
   string TypeOf_BaseOpcode = "MIMGBaseOpcode";
 
   let PrimaryKey = ["BaseOpcode"];
@@ -1458,13 +1460,14 @@ multiclass MIMG_Sampler_NoReturn <mimgopc op, AMDGPUSampleVariant sample, bit wq
   }
 }
 
-multiclass MIMG_Sampler <mimgopc op, AMDGPUSampleVariant sample, bit wqm = 0,
-                         bit isG16 = 0, bit isGetLod = 0,
+multiclass MIMG_Sampler <mimgopc op, AMDGPUSampleVariant sample, bit isPointSampleAccel = 0,
+                         bit wqm = 0, bit isG16 = 0, bit isGetLod = 0,
                          string asm = "image_sample"#sample.LowerCaseMod#!if(isG16, "_g16", ""),
                          bit ExtendedImageInst = !ne(sample.LowerCaseMod, "")> {
   def "" : MIMG_Sampler_BaseOpcode<sample> {
     let HasD16 = !not(isGetLod);
     let G16 = isG16;
+    let PointSampleAccel = isPointSampleAccel;
   }
 
   let BaseOpcode = !cast<MIMGBaseOpcode>(NAME), WQM = wqm,
@@ -1485,8 +1488,8 @@ multiclass MIMG_Sampler <mimgopc op, AMDGPUSampleVariant sample, bit wqm = 0,
   defm "_nortn" : MIMG_Sampler_NoReturn <op, sample, wqm, isG16, asm>;
 }
 
-multiclass MIMG_Sampler_WQM <mimgopc op, AMDGPUSampleVariant sample>
-    : MIMG_Sampler<op, sample, 1>;
+multiclass MIMG_Sampler_WQM <mimgopc op, AMDGPUSampleVariant sample, bit isPointSampleAccel = 0>
+    : MIMG_Sampler<op, sample, isPointSampleAccel, 1>;
 
 multiclass MIMG_Gather <mimgopc op, AMDGPUSampleVariant sample, bit wqm = 0,
                         string asm = "image_gather4"#sample.LowerCaseMod> {
@@ -1670,15 +1673,15 @@ let AssemblerPredicate = isGFX12Plus in {
   def : AMDGPUMnemonicAlias<"image_atomic_fmax", "image_atomic_max_flt">;
 }
 
-defm IMAGE_SAMPLE               : MIMG_Sampler_WQM <mimgopc<0x1b, 0x1b, 0x20>, AMDGPUSample>;
+defm IMAGE_SAMPLE               : MIMG_Sampler_WQM <mimgopc<0x1b, 0x1b, 0x20>, AMDGPUSample, 1>;
 let OtherPredicates = [HasImageInsts, HasExtendedImageInsts] in {
 defm IMAGE_SAMPLE_CL            : MIMG_Sampler_WQM <mimgopc<0x40, 0x40, 0x21>, AMDGPUSample_cl>;
 defm IMAGE_SAMPLE_D             : MIMG_Sampler <mimgopc<0x1c, 0x1c, 0x22>, AMDGPUSample_d>;
 defm IMAGE_SAMPLE_D_CL          : MIMG_Sampler <mimgopc<0x41, 0x41, 0x23>, AMDGPUSample_d_cl>;
-defm IMAGE_SAMPLE_L             : MIMG_Sampler <mimgopc<0x1d, 0x1d, 0x24>, AMDGPUSample_l>;
+defm IMAGE_SAMPLE_L             : MIMG_Sampler <mimgopc<0x1d, 0x1d, 0x24>, AMDGPUSample_l, 1>;
 defm IMAGE_SAMPLE_B             : MIMG_Sampler_WQM <mimgopc<0x1e, 0x1e, 0x25>, AMDGPUSample_b>;
 defm IMAGE_SAMPLE_B_CL          : MIMG_Sampler_WQM <mimgopc<0x42, 0x42, 0x26>, AMDGPUSample_b_cl>;
-defm IMAGE_SAMPLE_LZ            : MIMG_Sampler <mimgopc<0x1f, 0x1f, 0x27>, AMDGPUSample_lz>;
+defm IMAGE_SAMPLE_LZ            : MIMG_Sampler <mimgopc<0x1f, 0x1f, 0x27>, AMDGPUSample_lz, 1>;
 defm IMAGE_SAMPLE_C             : MIMG_Sampler_WQM <mimgopc<0x20, 0x20, 0x28>, AMDGPUSample_c>;
 defm IMAGE_SAMPLE_C_CL          : MIMG_Sampler_WQM <mimgopc<0x43, 0x43, 0x29>, AMDGPUSample_c_cl>;
 defm IMAGE_SAMPLE_C_D           : MIMG_Sampler <mimgopc<0x21, 0x21, 0x2a>, AMDGPUSample_c_d>;
@@ -1731,7 +1734,7 @@ defm IMAGE_GATHER4_C_LZ_O       : MIMG_Gather <mimgopc<0x37, 0x37, 0x5f>, AMDGPU
 let OtherPredicates = [HasImageInsts, HasExtendedImageInsts, isGFX9Plus] in
 defm IMAGE_GATHER4H             : MIMG_Gather <mimgopc<0x90, 0x90, 0x61, 0x42>, AMDGPUSample, 1, "image_gather4h">;
 
-defm IMAGE_GET_LOD              : MIMG_Sampler <mimgopc<0x38, 0x38, 0x60>, AMDGPUSample, 1, 0, 1, "image_get_lod">;
+defm IMAGE_GET_LOD              : MIMG_Sampler <mimgopc<0x38, 0x38, 0x60>, AMDGPUSample, 0, 1, 0, 1, "image_get_lod">;
 
 defm IMAGE_SAMPLE_CD            : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x68>, AMDGPUSample_cd>;
 defm IMAGE_SAMPLE_CD_CL         : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x69>, AMDGPUSample_cd_cl>;
@@ -1744,22 +1747,22 @@ defm IMAGE_SAMPLE_C_CD_CL_O     : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x6f
 } // End OtherPredicates = [HasImageInsts, HasExtendedImageInsts]
 
 let OtherPredicates = [HasImageInsts, HasExtendedImageInsts, HasG16] in {
-defm IMAGE_SAMPLE_D_G16         : MIMG_Sampler <mimgopc<0x39, 0x39, 0xa2>, AMDGPUSample_d, 0, 1>;
-defm IMAGE_SAMPLE_D_CL_G16      : MIMG_Sampler <mimgopc<0x5f, 0x5f, 0xa3>, AMDGPUSample_d_cl, 0, 1>;
-defm IMAGE_SAMPLE_C_D_G16       : MIMG_Sampler <mimgopc<0x3a, 0x3a, 0xaa>, AMDGPUSample_c_d, 0, 1>;
-defm IMAGE_SAMPLE_C_D_CL_G16    : MIMG_Sampler <mimgopc<0x54, 0x54, 0xab>, AMDGPUSample_c_d_cl, 0, 1>;
-defm IMAGE_SAMPLE_D_O_G16       : MIMG_Sampler <mimgopc<0x3b, 0x3b, 0xb2>, AMDGPUSample_d_o, 0, 1>;
-defm IMAGE_SAMPLE_D_CL_O_G16    : MIMG_Sampler <mimgopc<0x55, 0x55, 0xb3>, AMDGPUSample_d_cl_o, 0, 1>;
-defm IMAGE_SAMPLE_C_D_O_G16     : MIMG_Sampler <mimgopc<0x3c, 0x3c, 0xba>, AMDGPUSample_c_d_o, 0, 1>;
-defm IMAGE_SAMPLE_C_D_CL_O_G16  : MIMG_Sampler <mimgopc<0x56, 0x56, 0xbb>, AMDGPUSample_c_d_cl_o, 0, 1>;
-defm IMAGE_SAMPLE_CD_G16        : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xe8>, AMDGPUSample_cd, 0, 1>;
-defm IMAGE_SAMPLE_CD_CL_G16     : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xe9>, AMDGPUSample_cd_cl, 0, 1>;
-defm IMAGE_SAMPLE_C_CD_G16      : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xea>, AMDGPUSample_c_cd, 0, 1>;
-defm IMAGE_SAMPLE_C_CD_CL_G16   : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xeb>, AMDGPUSample_c_cd_cl, 0, 1>;
-defm IMAGE_SAMPLE_CD_O_G16      : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xec>, AMDGPUSample_cd_o, 0, 1>;
-defm IMAGE_SAMPLE_CD_CL_O_G16   : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xed>, AMDGPUSample_cd_cl_o, 0, 1>;
-defm IMAGE_SAMPLE_C_CD_O_G16    : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xee>, AMDGPUSample_c_cd_o, 0, 1>;
-defm IMAGE_SAMPLE_C_CD_CL_O_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xef>, AMDGPUSample_c_cd_cl_o, 0, 1>;
+defm IMAGE_SAMPLE_D_G16         : MIMG_Sampler <mimgopc<0x39, 0x39, 0xa2>, AMDGPUSample_d, 0, 0, 1>;
+defm IMAGE_SAMPLE_D_CL_G16      : MIMG_Sampler <mimgopc<0x5f, 0x5f, 0xa3>, AMDGPUSample_d_cl, 0, 0, 1>;
+defm IMAGE_SAMPLE_C_D_G16       : MIMG_Sampler <mimgopc<0x3a, 0x3a, 0xaa>, AMDGPUSample_c_d, 0, 0, 1>;
+defm IMAGE_SAMPLE_C_D_CL_G16    : MIMG_Sampler <mimgopc<0x54, 0x54, 0xab>, AMDGPUSample_c_d_cl, 0, 0, 1>;
+defm IMAGE_SAMPLE_D_O_G16       : MIMG_Sampler <mimgopc<0x3b, 0x3b, 0xb2>, AMDGPUSample_d_o, 0, 0, 1>;
+defm IMAGE_SAMPLE_D_CL_O_G16    : MIMG_Sampler <mimgopc<0x55, 0x55, 0xb3>, AMDGPUSample_d_cl_o, 0, 0, 1>;
+defm IMAGE_SAMPLE_C_D_O_G16     : MIMG_Sampler <mimgopc<0x3c, 0x3c, 0xba>, AMDGPUSample_c_d_o, 0, 0, 1>;
+defm IMAGE_SAMPLE_C_D_CL_O_G16  : MIMG_Sampler <mimgopc<0x56, 0x56, 0xbb>, AMDGPUSample_c_d_cl_o, 0, 0, 1>;
+defm IMAGE_SAMPLE_CD_G16        : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xe8>, AMDGPUSample_cd, 0, 0, 1>;
+defm IMAGE_SAMPLE_CD_CL_G16     : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xe9>, AMDGPUSample_cd_cl, 0, 0, 1>;
+defm IMAGE_SAMPLE_C_CD_G16      : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xea>, AMDGPUSample_c_cd, 0, 0, 1>;
+defm IMAGE_SAMPLE_C_CD_CL_G16   : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xeb>, AMDGPUSample_c_cd_cl, 0, 0, 1>;
+defm IMAGE_SAMPLE_CD_O_G16      : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xec>, AMDGPUSample_cd_o, 0, 0, 1>;
+defm IMAGE_SAMPLE_CD_CL_O_G16   : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xed>, AMDGPUSample_cd_cl_o, 0, 0, 1>;
+defm IMAGE_SAMPLE_C_CD_O_G16    : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xee>, AMDGPUSample_c_cd_o, 0, 0, 1>;
+defm IMAGE_SAMPLE_C_CD_CL_O_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xef>, AMDGPUSample_c_cd_cl_o, 0, 0, 1>;
 } // End OtherPredicates = [HasImageInsts, HasExtendedImageInsts, HasG16]
 
 //def IMAGE_RSRC256 : MIMG_NoPattern_RSRC256 <"image_rsrc256", mimgopc<0x7e>>;

llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp

@@ -384,6 +384,10 @@ class WaitcntBrackets {
     return LDSDMAStores;
   }
 
+  bool hasPointSampleAccel(const MachineInstr &MI) const;
+  bool hasPointSamplePendingVmemTypes(const MachineInstr &MI,
+                                      RegInterval Interval) const;
+
   void print(raw_ostream &);
   void dump() { print(dbgs()); }
 
@@ -808,6 +812,34 @@ void WaitcntBrackets::setScoreByOperand(const MachineInstr *MI,
   setScoreByInterval(Interval, CntTy, Score);
 }
 
+// Return true if the subtarget is one that enables Point Sample Acceleration
+// and the MachineInstr passed in is one to which it might be applied (the
+// hardware makes this decision based on several factors, but we can't determine
+// this at compile time, so we have to assume it might be applied if the
+// instruction supports it).
+bool WaitcntBrackets::hasPointSampleAccel(const MachineInstr &MI) const {
+  if (!ST->hasPointSampleAccel() || !SIInstrInfo::isMIMG(MI))
+    return false;
+
+  const AMDGPU::MIMGInfo *Info = AMDGPU::getMIMGInfo(MI.getOpcode());
+  const AMDGPU::MIMGBaseOpcodeInfo *BaseInfo =
+      AMDGPU::getMIMGBaseOpcodeInfo(Info->BaseOpcode);
+  return BaseInfo->PointSampleAccel;
+}
+
+// Return true if the subtarget enables Point Sample Acceleration, the supplied
+// MachineInstr is one to which it might be applied and the supplied interval is
+// one that has outstanding writes to vmem-types different than VMEM_NOSAMPLER
+// (this is the type that a point sample accelerated instruction effectively
+// becomes)
+bool WaitcntBrackets::hasPointSamplePendingVmemTypes(
+    const MachineInstr &MI, RegInterval Interval) const {
+  if (!hasPointSampleAccel(MI))
+    return false;
+
+  return hasOtherPendingVmemTypes(Interval, VMEM_NOSAMPLER);
+}
+
 void WaitcntBrackets::updateByEvent(const SIInstrInfo *TII,
                                     const SIRegisterInfo *TRI,
                                     const MachineRegisterInfo *MRI,
@@ -924,8 +956,13 @@ void WaitcntBrackets::updateByEvent(const SIInstrInfo *TII,
           // defs. That's required for a sane index into `VgprMemTypes` below
           assert(TRI->isVectorRegister(*MRI, Op.getReg()));
           VmemType V = getVmemType(Inst);
+          unsigned char TypesMask = 1 << V;
+          // If instruction can have Point Sample Accel applied, we have to flag
+          // this with another potential dependency
+          if (hasPointSampleAccel(Inst))
+            TypesMask |= 1 << VMEM_NOSAMPLER;
           for (int RegNo = Interval.first; RegNo < Interval.second; ++RegNo)
-            VgprVmemTypes[RegNo] |= 1 << V;
+            VgprVmemTypes[RegNo] |= TypesMask;
         }
       }
       setScoreByInterval(Interval, T, CurrScore);
@@ -1787,9 +1824,12 @@ bool SIInsertWaitcnts::generateWaitcntInstBefore(MachineInstr &MI,
           // previous write and this write are the same type of VMEM
           // instruction, in which case they are (in some architectures)
           // guaranteed to write their results in order anyway.
+          // Additionally check instructions where Point Sample Acceleration
+          // might be applied.
           if (Op.isUse() || !updateVMCntOnly(MI) ||
               ScoreBrackets.hasOtherPendingVmemTypes(Interval,
                                                      getVmemType(MI)) ||
+              ScoreBrackets.hasPointSamplePendingVmemTypes(MI, Interval) ||
               !ST->hasVmemWriteVgprInOrder()) {
             ScoreBrackets.determineWait(LOAD_CNT, Interval, Wait);
             ScoreBrackets.determineWait(SAMPLE_CNT, Interval, Wait);

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h

@@ -430,6 +430,7 @@ struct MIMGBaseOpcodeInfo {
   bool BVH;
   bool A16;
   bool NoReturn;
+  bool PointSampleAccel;
 };
 
 LLVM_READONLY

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.waterfall.ll

@@ -273,7 +273,7 @@ define amdgpu_ps <4 x float> @test_waterfall_non_uniform_img(<8 x i32> addrspace
 ; GFX1150-NEXT:    s_add_u32 s10, s0, s10
 ; GFX1150-NEXT:    s_addc_u32 s11, s1, s11
 ; GFX1150-NEXT:    s_load_b256 s[12:19], s[10:11], 0x0
-; GFX1150-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1150-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1150-NEXT:    image_sample v[0:3], v4, s[12:19], s[4:7] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX1150-NEXT:    ; implicit-def: $vgpr4
 ; GFX1150-NEXT:    s_xor_b64 exec, exec, s[20:21]
@@ -509,6 +509,7 @@ define amdgpu_ps <4 x float> @test_waterfall_non_uniform_img_single_read(<8 x i3
 ; GFX1150-NEXT:    v_readfirstlane_b32 s19, v14
 ; GFX1150-NEXT:    ; implicit-def: $vgpr4
 ; GFX1150-NEXT:    ; implicit-def: $vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14
+; GFX1150-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1150-NEXT:    image_sample v[0:3], v6, s[12:19], s[4:7] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX1150-NEXT:    ; implicit-def: $vgpr6
 ; GFX1150-NEXT:    s_xor_b64 exec, exec, s[2:3]
@@ -896,7 +897,7 @@ define amdgpu_ps <4 x float> @test_waterfall_non_uniform_img_multi_rl(<8 x i32>
 ; GFX1150-NEXT:    s_addc_u32 s17, s3, s11
 ; GFX1150-NEXT:    s_load_b256 s[8:15], s[8:9], 0x0
 ; GFX1150-NEXT:    s_load_b128 s[16:19], s[16:17], 0x0
-; GFX1150-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1150-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1150-NEXT:    image_sample v[0:3], v4, s[8:15], s[16:19] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX1150-NEXT:    ; implicit-def: $vgpr4
 ; GFX1150-NEXT:    s_xor_b64 exec, exec, s[20:21]
@@ -1110,7 +1111,7 @@ define amdgpu_ps <4 x float> @test_waterfall_non_uni_img_2_idx(<8 x i32> addrspa
 ; GFX1150-NEXT:    s_addc_u32 s17, s3, s11
 ; GFX1150-NEXT:    s_load_b256 s[8:15], s[8:9], 0x0
 ; GFX1150-NEXT:    s_load_b128 s[16:19], s[16:17], 0x0
-; GFX1150-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1150-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1150-NEXT:    image_sample v[0:3], v6, s[8:15], s[16:19] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX1150-NEXT:    ; implicit-def: $vgpr6
 ; GFX1150-NEXT:    s_xor_b64 exec, exec, s[20:21]
@@ -1644,7 +1645,7 @@ define amdgpu_ps <4 x float> @test_keep_waterfall_multi_rl(<8 x i32> addrspace(4
 ; GFX1150-NEXT:    s_addc_u32 s21, s3, s13
 ; GFX1150-NEXT:    s_load_b256 s[12:19], s[10:11], 0x0
 ; GFX1150-NEXT:    s_load_b128 s[20:23], s[20:21], 0x0
-; GFX1150-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1150-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1150-NEXT:    image_sample v[0:3], v4, s[12:19], s[20:23] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX1150-NEXT:    ; implicit-def: $vgpr4
 ; GFX1150-NEXT:    s_xor_b64 exec, exec, s[24:25]
@@ -1888,7 +1889,7 @@ define amdgpu_ps void @test_waterfall_sample_with_kill(<8 x i32> addrspace(4)* i
 ; GFX1150-NEXT:    s_addc_u32 s21, s3, s13
 ; GFX1150-NEXT:    s_load_b256 s[12:19], s[10:11], 0x0
 ; GFX1150-NEXT:    s_load_b128 s[20:23], s[20:21], 0x0
-; GFX1150-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX1150-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1150-NEXT:    image_sample v[2:5], v1, s[12:19], s[20:23] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX1150-NEXT:    ; implicit-def: $vgpr1
 ; GFX1150-NEXT:    s_xor_b64 exec, exec, s[24:25]
@@ -4002,6 +4003,7 @@ define amdgpu_ps {<4 x float>,<4 x float>} @test_waterfall_multi_begin_uniform_i
 ; GFX1150-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1150-NEXT:    ; implicit-def: $vgpr5
 ; GFX1150-NEXT:    ; implicit-def: $vgpr6
+; GFX1150-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1150-NEXT:    image_sample v[0:3], [v4, v4], s[8:15], s[16:19] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX1150-NEXT:    s_xor_b64 exec, exec, s[2:3]
 ; GFX1150-NEXT:    s_cbranch_execnz .LBB17_1