[AMDGPU] Introduce a new generic target gfx9-4-generic

shiltian · shiltian · commit 2fa6523a0fdf · 2024-11-07T23:42:51.000-05:00
diff --git a/clang/include/clang/Basic/Cuda.h b/clang/include/clang/Basic/Cuda.h
@@ -103,6 +103,7 @@ enum class OffloadArch {
   GFX909,
   GFX90a,
   GFX90c,
+  GFX9_4_GENERIC,
   GFX940,
   GFX941,
   GFX942,
diff --git a/clang/lib/Basic/Cuda.cpp b/clang/lib/Basic/Cuda.cpp
@@ -121,6 +121,7 @@ static const OffloadArchToStringMap arch_names[] = {
     GFX(909),  // gfx909
     GFX(90a),  // gfx90a
     GFX(90c),  // gfx90c
+    {OffloadArch::GFX9_4_GENERIC, "gfx9-4-generic", "compute_amdgcn"},
     GFX(940),  // gfx940
     GFX(941),  // gfx941
     GFX(942),  // gfx942
diff --git a/clang/lib/Basic/Targets/NVPTX.cpp b/clang/lib/Basic/Targets/NVPTX.cpp
@@ -205,6 +205,7 @@ void NVPTXTargetInfo::getTargetDefines(const LangOptions &Opts,
       case OffloadArch::GFX909:
       case OffloadArch::GFX90a:
       case OffloadArch::GFX90c:
+      case OffloadArch::GFX9_4_GENERIC:
       case OffloadArch::GFX940:
       case OffloadArch::GFX941:
       case OffloadArch::GFX942:
diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -2301,6 +2301,7 @@ void CGOpenMPRuntimeGPU::processRequiresDirective(const OMPRequiresDecl *D) {
       case OffloadArch::GFX909:
       case OffloadArch::GFX90a:
       case OffloadArch::GFX90c:
+      case OffloadArch::GFX9_4_GENERIC:
       case OffloadArch::GFX940:
       case OffloadArch::GFX941:
       case OffloadArch::GFX942:
diff --git a/clang/test/CodeGenOpenCL/amdgpu-features.cl b/clang/test/CodeGenOpenCL/amdgpu-features.cl
@@ -56,6 +56,8 @@
 
 // RUN: %clang_cc1 -triple amdgcn -target-cpu gfx1103 -target-feature +wavefrontsize64 -emit-llvm -o - %s | FileCheck --check-prefix=GFX1103-W64 %s
 
+// RUN: %clang_cc1 -triple amdgcn -target-cpu gfx9-4-generic -emit-llvm -o - %s | FileCheck --check-prefix=GFX9_4_Generic %s
+
 // NOCPU-NOT: "target-features"
 // NOCPU-WAVE32: "target-features"="+wavefrontsize32"
 // NOCPU-WAVE64: "target-features"="+wavefrontsize64"
@@ -85,6 +87,7 @@
 // GFX940: "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+fp8-conversion-insts,+fp8-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
 // GFX941: "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+fp8-conversion-insts,+fp8-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
 // GFX942: "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+fp8-conversion-insts,+fp8-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
+// GFX9_4_Generic: "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+fp8-conversion-insts,+fp8-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64"
 // GFX1010: "target-features"="+16-bit-insts,+ci-insts,+dl-insts,+dpp,+gfx10-insts,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize32"
 // GFX1011: "target-features"="+16-bit-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx10-insts,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize32"
 // GFX1012: "target-features"="+16-bit-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx10-insts,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize32"
diff --git a/clang/test/Driver/amdgpu-macros.cl b/clang/test/Driver/amdgpu-macros.cl
@@ -133,6 +133,7 @@
 // RUN: %clang -E -dM -target amdgcn -mcpu=gfx1201 %s 2>&1 | FileCheck --check-prefixes=ARCH-GCN,FAST_FMAF %s -DWAVEFRONT_SIZE=32 -DCPU=gfx1201 -DFAMILY=GFX12
 
 // RUN: %clang -E -dM -target amdgcn -mcpu=gfx9-generic %s 2>&1 | FileCheck --check-prefixes=ARCH-GCN,FAST_FMAF %s -DWAVEFRONT_SIZE=64 -DCPU=gfx9_generic -DFAMILY=GFX9
+// RUN: %clang -E -dM -target amdgcn -mcpu=gfx9-4-generic %s 2>&1 | FileCheck --check-prefixes=ARCH-GCN,FAST_FMAF %s -DWAVEFRONT_SIZE=64 -DCPU=gfx9_generic -DFAMILY=GFX9
 // RUN: %clang -E -dM -target amdgcn -mcpu=gfx10-1-generic %s 2>&1 | FileCheck --check-prefixes=ARCH-GCN,FAST_FMAF %s -DWAVEFRONT_SIZE=32 -DCPU=gfx10_1_generic -DFAMILY=GFX10
 // RUN: %clang -E -dM -target amdgcn -mcpu=gfx10-3-generic %s 2>&1 | FileCheck --check-prefixes=ARCH-GCN,FAST_FMAF %s -DWAVEFRONT_SIZE=32 -DCPU=gfx10_3_generic -DFAMILY=GFX10
 // RUN: %clang -E -dM -target amdgcn -mcpu=gfx11-generic %s 2>&1 | FileCheck --check-prefixes=ARCH-GCN,FAST_FMAF %s -DWAVEFRONT_SIZE=32 -DCPU=gfx11_generic -DFAMILY=GFX11
diff --git a/clang/test/Driver/amdgpu-mcpu.cl b/clang/test/Driver/amdgpu-mcpu.cl
@@ -118,6 +118,7 @@
 // RUN: %clang -### -target amdgcn -mcpu=gfx1201 %s 2>&1 | FileCheck --check-prefix=GFX1201 %s
 
 // RUN: %clang -### -target amdgcn -mcpu=gfx9-generic %s 2>&1 | FileCheck --check-prefix=GFX9_GENERIC %s
+// RUN: %clang -### -target amdgcn -mcpu=gfx9-4-generic %s 2>&1 | FileCheck --check-prefix=GFX9_4_GENERIC %s
 // RUN: %clang -### -target amdgcn -mcpu=gfx10-1-generic %s 2>&1 | FileCheck --check-prefix=GFX10_1_GENERIC %s
 // RUN: %clang -### -target amdgcn -mcpu=gfx10-3-generic %s 2>&1 | FileCheck --check-prefix=GFX10_3_GENERIC %s
 // RUN: %clang -### -target amdgcn -mcpu=gfx11-generic %s 2>&1 | FileCheck --check-prefix=GFX11_GENERIC %s
@@ -172,6 +173,7 @@
 // GFX1201:   "-target-cpu" "gfx1201"
 
 // GFX9_GENERIC:      "-target-cpu" "gfx9-generic"
+// GFX9_4_GENERIC:    "-target-cpu" "gfx9-4-generic"
 // GFX10_1_GENERIC:   "-target-cpu" "gfx10-1-generic"
 // GFX10_3_GENERIC:   "-target-cpu" "gfx10-3-generic"
 // GFX11_GENERIC:     "-target-cpu" "gfx11-generic"
diff --git a/clang/test/Misc/target-invalid-cpu-note/amdgcn.c b/clang/test/Misc/target-invalid-cpu-note/amdgcn.c
@@ -70,6 +70,7 @@
 // CHECK-SAME: {{^}}, gfx1200
 // CHECK-SAME: {{^}}, gfx1201
 // CHECK-SAME: {{^}}, gfx9-generic
+// CHECK-SAME: {{^}}, gfx9-4-generic
 // CHECK-SAME: {{^}}, gfx10-1-generic
 // CHECK-SAME: {{^}}, gfx10-3-generic
 // CHECK-SAME: {{^}}, gfx11-generic
diff --git a/clang/test/Misc/target-invalid-cpu-note/nvptx.c b/clang/test/Misc/target-invalid-cpu-note/nvptx.c
@@ -50,6 +50,7 @@
 // CHECK-SAME: {{^}}, gfx909
 // CHECK-SAME: {{^}}, gfx90a
 // CHECK-SAME: {{^}}, gfx90c
+// CHECK-SAME: {{^}}, gfx9-4-generic
 // CHECK-SAME: {{^}}, gfx940
 // CHECK-SAME: {{^}}, gfx941
 // CHECK-SAME: {{^}}, gfx942
diff --git a/llvm/docs/AMDGPUUsage.rst b/llvm/docs/AMDGPUUsage.rst
@@ -576,6 +576,12 @@ Generic processor code objects are versioned. See :ref:`amdgpu-generic-processor
                                                                                                   - ``v_dot2_f32_f16``
 
 
+     ``gfx9-4-generic``   ``amdgcn``     - ``gfx940``      - xnack            - Absolute flat   FP8 and BF8 instructions,
+                                         - ``gfx941``      - sramecc            scratch         FP8 and BF8 conversion instructions,
+                                         - ``gfx942``                                           as well as instructions with XF32 format support
+                                                                                                are not available.
+
+
      ``gfx10-1-generic``  ``amdgcn``     - ``gfx1010``     - xnack            - Absolute flat   - The following instructions are
                                          - ``gfx1011``     - wavefrontsize64    scratch           not available on ``gfx1011``
                                          - ``gfx1012``     - cumode                               and ``gfx1012``
@@ -2180,6 +2186,7 @@ The AMDGPU backend uses the following ELF header:
      *reserved*                                 0x057      Reserved.
      ``EF_AMDGPU_MACH_AMDGCN_GFX1153``          0x058      ``gfx1153``.
      ``EF_AMDGPU_MACH_AMDGCN_GFX12_GENERIC``    0x059      ``gfx12-generic``
+     ``EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC``   0x05f      ``gfx9-4-generic``
      ========================================== ========== =============================
 
 Sections
diff --git a/llvm/include/llvm/BinaryFormat/ELF.h b/llvm/include/llvm/BinaryFormat/ELF.h
@@ -822,11 +822,12 @@ enum : unsigned {
   EF_AMDGPU_MACH_AMDGCN_RESERVED_0X57   = 0x057,
   EF_AMDGPU_MACH_AMDGCN_GFX1153         = 0x058,
   EF_AMDGPU_MACH_AMDGCN_GFX12_GENERIC   = 0x059,
+  EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC  = 0x05f,
   // clang-format on
 
   // First/last AMDGCN-based processors.
   EF_AMDGPU_MACH_AMDGCN_FIRST = EF_AMDGPU_MACH_AMDGCN_GFX600,
-  EF_AMDGPU_MACH_AMDGCN_LAST = EF_AMDGPU_MACH_AMDGCN_GFX12_GENERIC,
+  EF_AMDGPU_MACH_AMDGCN_LAST = EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC,
 
   // Indicates if the "xnack" target feature is enabled for all code contained
   // in the object.
diff --git a/llvm/include/llvm/TargetParser/TargetParser.h b/llvm/include/llvm/TargetParser/TargetParser.h
@@ -119,9 +119,10 @@ enum GPUKind : uint32_t {
   GK_GFX10_3_GENERIC = 194,
   GK_GFX11_GENERIC = 195,
   GK_GFX12_GENERIC = 196,
+  GK_GFX9_4_GENERIC = 197,
 
   GK_AMDGCN_GENERIC_FIRST = GK_GFX9_GENERIC,
-  GK_AMDGCN_GENERIC_LAST = GK_GFX12_GENERIC,
+  GK_AMDGCN_GENERIC_LAST = GK_GFX9_4_GENERIC,
 };
 
 /// Instruction set architecture version.
diff --git a/llvm/lib/Object/ELFObjectFile.cpp b/llvm/lib/Object/ELFObjectFile.cpp
@@ -602,6 +602,8 @@ StringRef ELFObjectFileBase::getAMDGPUCPUName() const {
   // Generic AMDGCN targets
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX9_GENERIC:
     return "gfx9-generic";
+  case ELF::EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC:
+    return "gfx9-4-generic";
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX10_1_GENERIC:
     return "gfx10-1-generic";
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX10_3_GENERIC:
diff --git a/llvm/lib/ObjectYAML/ELFYAML.cpp b/llvm/lib/ObjectYAML/ELFYAML.cpp
@@ -631,6 +631,7 @@ void ScalarBitSetTraits<ELFYAML::ELF_EF>::bitset(IO &IO,
     BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1200, EF_AMDGPU_MACH);
     BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1201, EF_AMDGPU_MACH);
     BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX9_GENERIC, EF_AMDGPU_MACH);
+    BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC, EF_AMDGPU_MACH);
     BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX10_1_GENERIC, EF_AMDGPU_MACH);
     BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX10_3_GENERIC, EF_AMDGPU_MACH);
     BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX11_GENERIC, EF_AMDGPU_MACH);
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -1451,11 +1451,7 @@ def FeatureISAVersion9_4_Common : FeatureSet<
    FeatureDPALU_DPP,
    FeaturePackedFP32Ops,
    FeatureMAIInsts,
-   FeatureFP8Insts,
-   FeatureFP8ConversionInsts,
-   FeatureCvtFP8VOP1Bug,
    FeaturePkFmacF16Inst,
-   FeatureXF32Insts,
    FeatureAtomicFaddRtnInsts,
    FeatureAtomicFaddNoRtnInsts,
    FeatureAtomicBufferGlobalPkAddF16Insts,
@@ -1476,15 +1472,36 @@ def FeatureISAVersion9_4_Common : FeatureSet<
 
 def FeatureISAVersion9_4_0 : FeatureSet<
   !listconcat(FeatureISAVersion9_4_Common.Features,
-    [FeatureForceStoreSC0SC1])>;
+    [
+      FeatureForceStoreSC0SC1,
+      FeatureFP8Insts,
+      FeatureFP8ConversionInsts,
+      FeatureCvtFP8VOP1Bug,
+      FeatureXF32Insts
+    ])>;
 
 def FeatureISAVersion9_4_1 : FeatureSet<
   !listconcat(FeatureISAVersion9_4_Common.Features,
-    [FeatureForceStoreSC0SC1])>;
+    [
+      FeatureForceStoreSC0SC1,
+      FeatureFP8Insts,
+      FeatureFP8ConversionInsts,
+      FeatureCvtFP8VOP1Bug,
+      FeatureXF32Insts
+    ])>;
 
 def FeatureISAVersion9_4_2 : FeatureSet<
   !listconcat(FeatureISAVersion9_4_Common.Features,
-    [])>;
+    [
+      FeatureFP8Insts,
+      FeatureFP8ConversionInsts,
+      FeatureCvtFP8VOP1Bug,
+      FeatureXF32Insts
+    ])>;
+
+def FeatureISAVersion9_4_Generic : FeatureSet<
+  !listconcat(FeatureISAVersion9_4_Common.Features,
+    [FeatureRequiresCOV6])>;
 
 def FeatureISAVersion10_Common : FeatureSet<
   [FeatureGFX10,
diff --git a/llvm/lib/Target/AMDGPU/GCNProcessors.td b/llvm/lib/Target/AMDGPU/GCNProcessors.td
@@ -209,6 +209,11 @@ def : ProcessorModel<"gfx9-generic", SIQuarterSpeedModel,
   FeatureISAVersion9_Generic.Features
 >;
 
+// [gfx940, gfx941, gfx942]
+def : ProcessorModel<"gfx9-4-generic", SIDPGFX940FullSpeedModel,
+  FeatureISAVersion9_4_Generic.Features
+>;
+
 //===----------------------------------------------------------------------===//
 // GCN GFX10.
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp b/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp
@@ -119,6 +119,7 @@ StringRef AMDGPUTargetStreamer::getArchNameFromElfMach(unsigned ElfMach) {
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1200: AK = GK_GFX1200; break;
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1201: AK = GK_GFX1201; break;
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX9_GENERIC:     AK = GK_GFX9_GENERIC; break;
+  case ELF::EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC:   AK = GK_GFX9_4_GENERIC; break;
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX10_1_GENERIC:  AK = GK_GFX10_1_GENERIC; break;
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX10_3_GENERIC:  AK = GK_GFX10_3_GENERIC; break;
   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX11_GENERIC:    AK = GK_GFX11_GENERIC; break;
@@ -204,6 +205,7 @@ unsigned AMDGPUTargetStreamer::getElfMach(StringRef GPU) {
   case GK_GFX1200: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1200;
   case GK_GFX1201: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1201;
   case GK_GFX9_GENERIC:     return ELF::EF_AMDGPU_MACH_AMDGCN_GFX9_GENERIC;
+  case GK_GFX9_4_GENERIC:   return ELF::EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC;
   case GK_GFX10_1_GENERIC:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX10_1_GENERIC;
   case GK_GFX10_3_GENERIC:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX10_3_GENERIC;
   case GK_GFX11_GENERIC:    return ELF::EF_AMDGPU_MACH_AMDGCN_GFX11_GENERIC;
@@ -821,6 +823,9 @@ unsigned AMDGPUTargetELFStreamer::getEFlagsV6() {
     case AMDGPU::GK_GFX9_GENERIC:
       Version = GenericVersion::GFX9;
       break;
+    case AMDGPU::GK_GFX9_4_GENERIC:
+      Version = GenericVersion::GFX9_4;
+      break;
     case AMDGPU::GK_GFX10_1_GENERIC:
       Version = GenericVersion::GFX10_1;
       break;
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
@@ -46,6 +46,7 @@ struct IsaVersion;
 /// within a generic family.
 namespace GenericVersion {
 static constexpr unsigned GFX9 = 1;
+static constexpr unsigned GFX9_4 = 1;
 static constexpr unsigned GFX10_1 = 1;
 static constexpr unsigned GFX10_3 = 1;
 static constexpr unsigned GFX11 = 1;
diff --git a/llvm/lib/TargetParser/TargetParser.cpp b/llvm/lib/TargetParser/TargetParser.cpp
@@ -130,6 +130,7 @@ constexpr GPUInfo AMDGCNGPUs[] = {
     {{"gfx1201"},   {"gfx1201"}, GK_GFX1201, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
 
     {{"gfx9-generic"},      {"gfx9-generic"},    GK_GFX9_GENERIC,    FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_XNACK},
+    {{"gfx9-4-generic"},    {"gfx9-4-generic"},  GK_GFX9_4_GENERIC,  FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_XNACK|FEATURE_SRAMECC},
     {{"gfx10-1-generic"},   {"gfx10-1-generic"}, GK_GFX10_1_GENERIC, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_XNACK|FEATURE_WGP},
     {{"gfx10-3-generic"},   {"gfx10-3-generic"}, GK_GFX10_3_GENERIC, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
     {{"gfx11-generic"},     {"gfx11-generic"},   GK_GFX11_GENERIC,   FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
@@ -155,6 +156,7 @@ const GPUInfo *getArchEntry(AMDGPU::GPUKind AK, ArrayRef<GPUInfo> Table) {
 StringRef llvm::AMDGPU::getArchFamilyNameAMDGCN(GPUKind AK) {
   switch (AK) {
   case AMDGPU::GK_GFX9_GENERIC:
+  case AMDGPU::GK_GFX9_4_GENERIC:
     return "gfx9";
   case AMDGPU::GK_GFX10_1_GENERIC:
   case AMDGPU::GK_GFX10_3_GENERIC:
@@ -296,6 +298,7 @@ AMDGPU::IsaVersion AMDGPU::getIsaVersion(StringRef GPU) {
   // TODO: Split up this API depending on its caller so
   // generic target handling is more obvious and less risky.
   case GK_GFX9_GENERIC:    return {9, 0, 0};
+  case GK_GFX9_4_GENERIC:  return {9, 4, 0};
   case GK_GFX10_1_GENERIC: return {10, 1, 0};
   case GK_GFX10_3_GENERIC: return {10, 3, 0};
   case GK_GFX11_GENERIC:   return {11, 0, 3};
@@ -466,9 +469,12 @@ void AMDGPU::fillAMDGPUFeatureMap(StringRef GPU, const Triple &T,
     case GK_GFX942:
     case GK_GFX941:
     case GK_GFX940:
-      Features["gfx940-insts"] = true;
       Features["fp8-insts"] = true;
       Features["fp8-conversion-insts"] = true;
+      Features["xf32-insts"] = true;
+      [[fallthrough]];
+    case GK_GFX9_4_GENERIC:
+      Features["gfx940-insts"] = true;
       Features["atomic-ds-pk-add-16-insts"] = true;
       Features["atomic-flat-pk-add-16-insts"] = true;
       Features["atomic-global-pk-add-bf16-inst"] = true;
diff --git a/llvm/test/CodeGen/AMDGPU/directive-amdgcn-target.ll b/llvm/test/CodeGen/AMDGPU/directive-amdgcn-target.ll
@@ -112,6 +112,8 @@
 
 ; RUN: llc --amdhsa-code-object-version=6 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx9-generic -mattr=-xnack < %s | FileCheck --check-prefixes=GFX9_GENERIC_NOXNACK %s
 ; RUN: llc --amdhsa-code-object-version=6 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx9-generic -mattr=+xnack < %s | FileCheck --check-prefixes=GFX9_GENERIC_XNACK %s
+; RUN: llc --amdhsa-code-object-version=6 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx9-4-generic -mattr=-xnack < %s | FileCheck --check-prefixes=GFX9_4_GENERIC_NOXNACK %s
+; RUN: llc --amdhsa-code-object-version=6 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx9-4-generic -mattr=+xnack < %s | FileCheck --check-prefixes=GFX9_4_GENERIC_XNACK %s
 ; RUN: llc --amdhsa-code-object-version=6 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx10-1-generic -mattr=-xnack < %s | FileCheck --check-prefixes=GFX10_1_GENERIC_NOXNACK %s
 ; RUN: llc --amdhsa-code-object-version=6 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx10-1-generic -mattr=+xnack < %s | FileCheck --check-prefixes=GFX10_1_GENERIC_XNACK %s
 ; RUN: llc --amdhsa-code-object-version=6 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx10-3-generic < %s | FileCheck --check-prefixes=GFX10_3_GENERIC %s
@@ -210,6 +212,8 @@
 
 ; GFX9_GENERIC_NOXNACK:     .amdgcn_target "amdgcn-amd-amdhsa--gfx9-generic:xnack-"
 ; GFX9_GENERIC_XNACK:       .amdgcn_target "amdgcn-amd-amdhsa--gfx9-generic:xnack+"
+; GFX9_4_GENERIC_NOXNACK:   .amdgcn_target "amdgcn-amd-amdhsa--gfx9-4-generic:xnack-"
+; GFX9_4_GENERIC_XNACK:     .amdgcn_target "amdgcn-amd-amdhsa--gfx9-4-generic:xnack+"
 ; GFX10_1_GENERIC_NOXNACK:  .amdgcn_target "amdgcn-amd-amdhsa--gfx10-1-generic:xnack-"
 ; GFX10_1_GENERIC_XNACK:    .amdgcn_target "amdgcn-amd-amdhsa--gfx10-1-generic:xnack+"
 ; GFX10_3_GENERIC:          .amdgcn_target "amdgcn-amd-amdhsa--gfx10-3-generic"
diff --git a/llvm/test/CodeGen/AMDGPU/div-rem-by-constant-64.ll b/llvm/test/CodeGen/AMDGPU/div-rem-by-constant-64.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -o - %s | FileCheck -check-prefixes=GFX942 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx9-4-generic --amdhsa-code-object-version=6 -o - %s | FileCheck -check-prefixes=GFX942 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1030 -o - %s | FileCheck -check-prefixes=GFX1030 %s
 
 ; Sample test to check how we deal with division/modulos by 64 bit constants.
diff --git a/llvm/test/CodeGen/AMDGPU/dst-sel-hazard.mir b/llvm/test/CodeGen/AMDGPU/dst-sel-hazard.mir
@@ -1,5 +1,6 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
 # RUN: llc -mtriple=amdgcn -mcpu=gfx942 -run-pass post-RA-hazard-rec -o - %s | FileCheck -check-prefix=HAZARD %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx9-4-generic -run-pass post-RA-hazard-rec -o - %s | FileCheck -check-prefix=HAZARD %s
 # RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass post-RA-hazard-rec -o - %s | FileCheck -check-prefix=NOHAZARD %s
 
 ---
diff --git a/llvm/test/CodeGen/AMDGPU/elf-header-flags-mach.ll b/llvm/test/CodeGen/AMDGPU/elf-header-flags-mach.ll
diff --git a/llvm/test/CodeGen/AMDGPU/generic-targets-require-v6.ll b/llvm/test/CodeGen/AMDGPU/generic-targets-require-v6.ll
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.exp.large.mir b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.exp.large.mir
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.exp.small.mir b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.exp.small.mir
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll
diff --git a/llvm/test/CodeGen/AMDGPU/no-corresponding-integer-type.ll b/llvm/test/CodeGen/AMDGPU/no-corresponding-integer-type.ll
diff --git a/llvm/test/MC/AMDGPU/gfx9_4_generic_unsupported.s b/llvm/test/MC/AMDGPU/gfx9_4_generic_unsupported.s
diff --git a/llvm/test/Object/AMDGPU/elf-header-flags-mach.yaml b/llvm/test/Object/AMDGPU/elf-header-flags-mach.yaml
diff --git a/llvm/test/tools/llvm-objdump/ELF/AMDGPU/subtarget.ll b/llvm/test/tools/llvm-objdump/ELF/AMDGPU/subtarget.ll
diff --git a/llvm/test/tools/llvm-readobj/ELF/AMDGPU/elf-headers.test b/llvm/test/tools/llvm-readobj/ELF/AMDGPU/elf-headers.test
diff --git a/llvm/tools/llvm-readobj/ELFDumper.cpp b/llvm/tools/llvm-readobj/ELFDumper.cpp

-Original file line number
+Diff line change
   GFX909,
   GFX90a,
   GFX90c,
 +  GFX9_4_GENERIC,
   GFX940,
   GFX941,
   GFX942,