diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
--- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
@@ -1688,7 +1688,7 @@
         NeedWaitStates = SMFMA4x4WriteVgprVALUMemExpReadWaitStates;
         break;
       case 4:
-        assert(isDGEMM(MFMA->getOpcode()));
+        assert(isDGEMM(MFMA->getOpcode()) || ST.hasGFX940Insts());
         NeedWaitStates =
             IsMemOrExport ? DMFMA4x4WriteVgprMemExpReadWaitStates
                           : DMFMA4x4WriteVgprVALUReadWaitStates;
diff --git a/llvm/lib/Target/AMDGPU/GCNProcessors.td b/llvm/lib/Target/AMDGPU/GCNProcessors.td
--- a/llvm/lib/Target/AMDGPU/GCNProcessors.td
+++ b/llvm/lib/Target/AMDGPU/GCNProcessors.td
@@ -192,7 +192,7 @@
   FeatureISAVersion9_0_C.Features
 >;
 
-def : ProcessorModel<"gfx940", SIDPFullSpeedModel,
+def : ProcessorModel<"gfx940", SIDPGFX940FullSpeedModel,
   FeatureISAVersion9_4_0.Features
 >;
 
diff --git a/llvm/lib/Target/AMDGPU/SISchedule.td b/llvm/lib/Target/AMDGPU/SISchedule.td
--- a/llvm/lib/Target/AMDGPU/SISchedule.td
+++ b/llvm/lib/Target/AMDGPU/SISchedule.td
@@ -59,6 +59,7 @@
 
 // mAI multipass instructions.
 def Write2PassMAI  : SchedWrite;
+def Write4PassMAI  : SchedWrite;
 def Write8PassMAI  : SchedWrite;
 def Write16PassMAI : SchedWrite;
 def Write4PassDGEMM : SchedWrite;
@@ -86,6 +87,7 @@
 def SIFullSpeedModel : SISchedMachineModel;
 def SIQuarterSpeedModel : SISchedMachineModel;
 def SIDPFullSpeedModel : SISchedMachineModel;
+def SIDPGFX940FullSpeedModel : SISchedMachineModel;
 def GFX10SpeedModel : SISchedMachineModel;
 
 // XXX: Are the resource counts correct?
@@ -156,6 +158,8 @@
 
   let ResourceCycles = [2] in
   def : HWWriteRes<Write2PassMAI,  [HWXDL], 2>;
+  let ResourceCycles = [4] in
+  def : HWWriteRes<Write4PassMAI,  [HWXDL], 4>;
   let ResourceCycles = [8] in
   def : HWWriteRes<Write8PassMAI,  [HWXDL], 8>;
   let ResourceCycles = [16] in
@@ -244,6 +248,33 @@
 
 } // End SchedModel = SIDPFullSpeedModel
 
+let SchedModel = SIDPGFX940FullSpeedModel in {
+
+defm : SICommonWriteRes;
+
+def : HWVALUWriteRes<WriteFloatFMA,    1>;
+def : HWVALUWriteRes<WriteDouble,      1>;
+def : HWVALUWriteRes<WriteDoubleAdd,   1>;
+def : HWVALUWriteRes<WriteDoubleCvt,   1>;
+def : HWVALUWriteRes<WriteTrans64,     4>;
+def : HWVALUWriteRes<WriteIntMul,      1>;
+def : HWVALUWriteRes<Write64Bit,       1>;
+
+def : InstRW<[WriteCopy], (instrs COPY)>;
+def : InstRW<[Write64Bit], (instregex "^V_ACCVGPR_WRITE_B32_e64$")>;
+def : InstRW<[Write2PassMAI,   MIMFMARead], (instregex "^V_MFMA_.32_4X4X")>;
+
+def : InstRW<[Write4PassMAI,   MIMFMARead], (instregex "^V_MFMA_.32_16X16X16")>;
+def : InstRW<[Write8PassMAI,   MIMFMARead], (instregex "^V_MFMA_.32_16X16X[14][FBI]")>;
+
+def : InstRW<[Write8PassMAI,   MIMFMARead], (instregex "^V_MFMA_.32_32X32X8")>;
+def : InstRW<[Write16PassMAI,  MIMFMARead], (instregex "^V_MFMA_.32_32X32X[124][FBI]")>;
+
+def : InstRW<[Write4PassDGEMM, MIMFMARead], (instregex "^V_MFMA_.64_4X4X")>;
+def : InstRW<[Write8PassDGEMM, MIMFMARead], (instregex "^V_MFMA_.64_16X16X")>;
+
+} // End SchedModel = SIDPGFX940FullSpeedModel
+
 let SchedModel = GFX10SpeedModel in {
 
 // The latency values are 1 / (operations / cycle).