[AMDGPU] Use a target feature to enable __builtin_amdgcn_global_load_lds on gfx9/10 (llvm#133055)

jmmartinez · David Salinas · commit 4f5b6142e421 · 2025-04-17T18:02:06.000Z
This patch introduces the `vmem-to-lds-load-insts` target feature, which
can be used to enable builtins `__builtin_amdgcn_global_load_lds` and
`__builtin_amdgcn_raw_ptr_buffer_load_lds` on platforms which have this
feature.

This feature is only available on gfx9/10.

A limitation of using a common target feature for both builtins is that
we could have made `__builtin_amdgcn_raw_ptr_buffer_load_lds` available
on gfx6,7,8.

Change-Id: Iaca77f588cccc1074916e5a9452fea0acbfb4df5
diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -256,7 +256,7 @@ TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_v2bf16, "V2sV2s*0V2s", "t", "at
 TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_v2bf16, "V2sV2s*1V2s", "t", "atomic-global-pk-add-bf16-inst")
 TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_v2bf16, "V2sV2s*3V2s", "t", "atomic-ds-pk-add-16-insts")
 TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_v2f16, "V2hV2h*3V2h", "t", "atomic-ds-pk-add-16-insts")
-TARGET_BUILTIN(__builtin_amdgcn_global_load_lds, "vv*1v*3IUiIiIUi", "t", "gfx940-insts")
+TARGET_BUILTIN(__builtin_amdgcn_global_load_lds, "vv*1v*3IUiIiIUi", "t", "vmem-to-lds-load-insts")
 
 //===----------------------------------------------------------------------===//
 // Deep learning builtins.
diff --git a/clang/lib/Basic/Targets/AMDGPU.cpp b/clang/lib/Basic/Targets/AMDGPU.cpp
@@ -253,7 +253,7 @@ AMDGPUTargetInfo::AMDGPUTargetInfo(const llvm::Triple &Triple,
 
   MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 64;
   CUMode = !(GPUFeatures & llvm::AMDGPU::FEATURE_WGP);
-  for (auto F : {"image-insts", "gws"})
+  for (auto F : {"image-insts", "gws", "vmem-to-lds-load-insts"})
     ReadOnlyFeatures.insert(F);
   HalfArgsAndReturns = true;
 }
diff --git a/clang/test/CodeGen/link-builtin-bitcode.c b/clang/test/CodeGen/link-builtin-bitcode.c
@@ -44,6 +44,6 @@ int bar() { return no_attr() + attr_in_target() + attr_not_in_target() + attr_in
 // CHECK-SAME: () #[[ATTR_INCOMPATIBLE:[0-9]+]] {
 
 // CHECK: attributes #[[ATTR_BAR]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64" }
-// CHECK: attributes #[[ATTR_COMPATIBLE]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64" }
-// CHECK: attributes #[[ATTR_EXTEND]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+extended-image-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64" }
-// CHECK: attributes #[[ATTR_INCOMPATIBLE]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-gfx9-insts" }
+// CHECK: attributes #[[ATTR_COMPATIBLE]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+vmem-to-lds-load-insts,+wavefrontsize64" }
+// CHECK: attributes #[[ATTR_EXTEND]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+extended-image-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+vmem-to-lds-load-insts,+wavefrontsize64" }
+// CHECK: attributes #[[ATTR_INCOMPATIBLE]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+vmem-to-lds-load-insts,+wavefrontsize64,-gfx9-insts" }
diff --git a/clang/test/CodeGenCXX/dynamic-cast-address-space.cpp b/clang/test/CodeGenCXX/dynamic-cast-address-space.cpp
@@ -112,9 +112,9 @@ const B& f(A *a) {
 // CHECK: attributes #[[ATTR3]] = { nounwind }
 // CHECK: attributes #[[ATTR4]] = { noreturn }
 //.
-// WITH-NONZERO-DEFAULT-AS: attributes #[[ATTR0]] = { mustprogress noinline optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+16-bit-insts,+ashr-pk-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-buffer-pk-add-bf16-inst,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+bf8-cvt-scale-insts,+bitop3-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot11-insts,+dot12-insts,+dot13-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dot8-insts,+dot9-insts,+dpp,+f16bf16-to-fp6bf6-cvt-scale-insts,+f32-to-f16bf16-cvt-sr-insts,+fp4-cvt-scale-insts,+fp6bf6-cvt-scale-insts,+fp8-conversion-insts,+fp8-cvt-scale-insts,+fp8-insts,+gfx10-3-insts,+gfx10-insts,+gfx11-insts,+gfx12-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+gfx950-insts,+gws,+image-insts,+mai-insts,+permlane16-swap,+permlane32-swap,+prng-inst,+s-memrealtime,+s-memtime-inst,+wavefrontsize32,+wavefrontsize64" }
+// WITH-NONZERO-DEFAULT-AS: attributes #[[ATTR0]] = { mustprogress noinline optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+16-bit-insts,+ashr-pk-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-buffer-pk-add-bf16-inst,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+bf8-cvt-scale-insts,+bitop3-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot11-insts,+dot12-insts,+dot13-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dot8-insts,+dot9-insts,+dpp,+f16bf16-to-fp6bf6-cvt-scale-insts,+f32-to-f16bf16-cvt-sr-insts,+fp4-cvt-scale-insts,+fp6bf6-cvt-scale-insts,+fp8-conversion-insts,+fp8-cvt-scale-insts,+fp8-insts,+gfx10-3-insts,+gfx10-insts,+gfx11-insts,+gfx12-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+gfx950-insts,+gws,+image-insts,+mai-insts,+permlane16-swap,+permlane32-swap,+prng-inst,+s-memrealtime,+s-memtime-inst,+vmem-to-lds-load-insts,+wavefrontsize32,+wavefrontsize64" }
 // WITH-NONZERO-DEFAULT-AS: attributes #[[ATTR1:[0-9]+]] = { nounwind willreturn memory(read) }
-// WITH-NONZERO-DEFAULT-AS: attributes #[[ATTR2:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+16-bit-insts,+ashr-pk-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-buffer-pk-add-bf16-inst,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+bf8-cvt-scale-insts,+bitop3-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot11-insts,+dot12-insts,+dot13-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dot8-insts,+dot9-insts,+dpp,+f16bf16-to-fp6bf6-cvt-scale-insts,+f32-to-f16bf16-cvt-sr-insts,+fp4-cvt-scale-insts,+fp6bf6-cvt-scale-insts,+fp8-conversion-insts,+fp8-cvt-scale-insts,+fp8-insts,+gfx10-3-insts,+gfx10-insts,+gfx11-insts,+gfx12-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+gfx950-insts,+gws,+image-insts,+mai-insts,+permlane16-swap,+permlane32-swap,+prng-inst,+s-memrealtime,+s-memtime-inst,+wavefrontsize32,+wavefrontsize64" }
+// WITH-NONZERO-DEFAULT-AS: attributes #[[ATTR2:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+16-bit-insts,+ashr-pk-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-buffer-pk-add-bf16-inst,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+bf8-cvt-scale-insts,+bitop3-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot11-insts,+dot12-insts,+dot13-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dot8-insts,+dot9-insts,+dpp,+f16bf16-to-fp6bf6-cvt-scale-insts,+f32-to-f16bf16-cvt-sr-insts,+fp4-cvt-scale-insts,+fp6bf6-cvt-scale-insts,+fp8-conversion-insts,+fp8-cvt-scale-insts,+fp8-insts,+gfx10-3-insts,+gfx10-insts,+gfx11-insts,+gfx12-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+gfx950-insts,+gws,+image-insts,+mai-insts,+permlane16-swap,+permlane32-swap,+prng-inst,+s-memrealtime,+s-memtime-inst,+vmem-to-lds-load-insts,+wavefrontsize32,+wavefrontsize64" }
 // WITH-NONZERO-DEFAULT-AS: attributes #[[ATTR3]] = { nounwind }
 // WITH-NONZERO-DEFAULT-AS: attributes #[[ATTR4]] = { noreturn }
 //.
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-global-load-lds.cl b/clang/test/CodeGenOpenCL/builtins-amdgcn-global-load-lds.cl
@@ -1,5 +1,7 @@
 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
-// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx940 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx900 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx942 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx1010 -emit-llvm -o - %s | FileCheck %s
 // REQUIRES: amdgpu-registered-target
 
 typedef unsigned int u32;
diff --git a/clang/test/SemaOpenCL/builtins-amdgcn-global-load-lds-err.cl b/clang/test/SemaOpenCL/builtins-amdgcn-global-load-lds-err.cl
@@ -1,4 +1,6 @@
-// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx940 -S -verify=gfx940,expected -o - %s
+// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx900 -S -verify=gfx,expected -o - %s
+// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx942 -S -verify=gfx,expected -o - %s
+// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx1010 -S -verify=gfx,expected -o - %s
 // RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx950 -S -verify=gfx950,expected -o - %s
 // REQUIRES: amdgpu-registered-target
 
@@ -8,12 +10,12 @@ void test_global_load_lds_unsupported_size(global u32* src, local u32 *dst, u32
   __builtin_amdgcn_global_load_lds(src, dst, size, /*offset=*/0, /*aux=*/0); // expected-error{{argument to '__builtin_amdgcn_global_load_lds' must be a constant integer}}
   __builtin_amdgcn_global_load_lds(src, dst, /*size=*/4, offset, /*aux=*/0); // expected-error{{argument to '__builtin_amdgcn_global_load_lds' must be a constant integer}}
   __builtin_amdgcn_global_load_lds(src, dst, /*size=*/4, /*offset=*/0, aux); // expected-error{{argument to '__builtin_amdgcn_global_load_lds' must be a constant integer}}
-  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/5, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx940-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
-  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/0, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx940-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
-  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/3, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx940-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
-  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/12, /*offset=*/0, /*aux=*/0); // gfx940-error{{invalid size value}} gfx940-note {{size must be 1, 2, or 4}}
-  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/16, /*offset=*/0, /*aux=*/0); // gfx940-error{{invalid size value}} gfx940-note {{size must be 1, 2, or 4}}
-  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/-1, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx940-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
+  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/5, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
+  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/0, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
+  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/3, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
+  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/12, /*offset=*/0, /*aux=*/0); // gfx-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}}
+  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/16, /*offset=*/0, /*aux=*/0); // gfx-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}}
+  __builtin_amdgcn_global_load_lds(src, dst, /*size=*/-1, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
 }
 
 __attribute__((target("gfx950-insts")))
diff --git a/flang/test/Lower/OpenMP/target_cpu_features.f90 b/flang/test/Lower/OpenMP/target_cpu_features.f90
@@ -12,7 +12,7 @@
 !AMDGCN-SAME: "+dl-insts", "+dot1-insts", "+dot10-insts", "+dot2-insts", "+dot3-insts",
 !AMDGCN-SAME: "+dot4-insts", "+dot5-insts", "+dot6-insts", "+dot7-insts", "+dpp",
 !AMDGCN-SAME: "+gfx8-insts", "+gfx9-insts", "+gws", "+image-insts", "+mai-insts",
-!AMDGCN-SAME: "+s-memrealtime", "+s-memtime-inst", "+wavefrontsize64"]>
+!AMDGCN-SAME: "+s-memrealtime", "+s-memtime-inst", "+vmem-to-lds-load-insts", "+wavefrontsize64"]>
 
 !NVPTX: module attributes {
 !NVPTX-SAME: fir.target_cpu = "sm_80"
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -1245,6 +1245,12 @@ def FeatureXF32Insts : SubtargetFeature<"xf32-insts",
    "v_mfma_f32_16x16x8_xf32 and v_mfma_f32_32x32x4_xf32"
  >;
 
+def FeatureMemToLDSLoad : SubtargetFeature<"vmem-to-lds-load-insts",
+  "HasVMemToLDSLoad",
+  "true",
+  "The platform has memory to lds instructions (global_load w/lds bit set, buffer_load w/lds bit set or global_load_lds. This does not include scratch_load_lds."
+>;
+
 // Dummy feature used to disable assembler instructions.
 def FeatureDisable : SubtargetFeature<"",
   "FeatureDisable","true",
@@ -1315,7 +1321,7 @@ def FeatureGFX9 : GCNSubtargetFeatureGeneration<"GFX9",
    FeatureA16, FeatureSMemTimeInst, FeatureFastDenormalF32, FeatureSupportsXNACK,
    FeatureUnalignedBufferAccess, FeatureUnalignedScratchAccess,
    FeatureUnalignedDSAccess, FeatureNegativeScratchOffsetBug, FeatureGWS,
-   FeatureDefaultComponentZero,FeatureVmemWriteVgprInOrder
+   FeatureDefaultComponentZero,FeatureVmemWriteVgprInOrder, FeatureMemToLDSLoad
   ]
 >;
 
@@ -1339,7 +1345,7 @@ def FeatureGFX10 : GCNSubtargetFeatureGeneration<"GFX10",
    FeatureDefaultComponentZero, FeatureMaxHardClauseLength63,
    FeatureAtomicFMinFMaxF32GlobalInsts, FeatureAtomicFMinFMaxF64GlobalInsts,
    FeatureAtomicFMinFMaxF32FlatInsts, FeatureAtomicFMinFMaxF64FlatInsts,
-   FeatureVmemWriteVgprInOrder
+   FeatureVmemWriteVgprInOrder, FeatureMemToLDSLoad
   ]
 >;
 
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -3337,7 +3337,8 @@ bool AMDGPUInstructionSelector::selectG_INSERT_VECTOR_ELT(
 }
 
 bool AMDGPUInstructionSelector::selectBufferLoadLds(MachineInstr &MI) const {
-  assert(!AMDGPU::isGFX12Plus(STI));
+  if (!Subtarget->hasVMemToLDSLoad())
+    return false;
   unsigned Opc;
   unsigned Size = MI.getOperand(3).getImm();
 
@@ -3473,6 +3474,9 @@ static Register matchZeroExtendFromS32(MachineRegisterInfo &MRI, Register Reg) {
 }
 
 bool AMDGPUInstructionSelector::selectGlobalLoadLds(MachineInstr &MI) const{
+  if (!Subtarget->hasVMemToLDSLoad())
+    return false;
+
   unsigned Opc;
   unsigned Size = MI.getOperand(3).getImm();
 
diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -190,6 +190,7 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
   /// indicates a lack of S_CLAUSE support.
   unsigned MaxHardClauseLength = 0;
   bool SupportsSRAMECC = false;
+  bool HasVMemToLDSLoad = false;
 
   // This should not be used directly. 'TargetID' tracks the dynamic settings
   // for SRAMECC.
@@ -1314,6 +1315,8 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
     return hasGFX950Insts();
   }
 
+  bool hasVMemToLDSLoad() const { return HasVMemToLDSLoad; }
+
   bool hasSALUFloatInsts() const { return HasSALUFloatInsts; }
 
   bool hasPseudoScalarTrans() const { return HasPseudoScalarTrans; }
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -10001,7 +10001,8 @@ SDValue SITargetLowering::LowerINTRINSIC_VOID(SDValue Op,
   case Intrinsic::amdgcn_raw_ptr_buffer_load_lds:
   case Intrinsic::amdgcn_struct_buffer_load_lds:
   case Intrinsic::amdgcn_struct_ptr_buffer_load_lds: {
-    assert(!AMDGPU::isGFX12Plus(*Subtarget));
+    if (!Subtarget->hasVMemToLDSLoad())
+      return SDValue();
     unsigned Opc;
     bool HasVIndex =
         IntrinsicID == Intrinsic::amdgcn_struct_buffer_load_lds ||
@@ -10108,6 +10109,9 @@ SDValue SITargetLowering::LowerINTRINSIC_VOID(SDValue Op,
     return SDValue(Load, 0);
   }
   case Intrinsic::amdgcn_global_load_lds: {
+    if (!Subtarget->hasVMemToLDSLoad())
+      return SDValue();
+
     unsigned Opc;
     unsigned Size = Op->getConstantOperandVal(4);
     switch (Size) {
diff --git a/llvm/lib/TargetParser/TargetParser.cpp b/llvm/lib/TargetParser/TargetParser.cpp
@@ -378,6 +378,7 @@ void AMDGPU::fillAMDGPUFeatureMap(StringRef GPU, const Triple &T,
     Features["prng-inst"] = true;
     Features["wavefrontsize32"] = true;
     Features["wavefrontsize64"] = true;
+    Features["vmem-to-lds-load-insts"] = true;
   } else if (T.isAMDGCN()) {
     AMDGPU::GPUKind Kind = parseArchAMDGCN(GPU);
     switch (Kind) {
@@ -463,6 +464,7 @@ void AMDGPU::fillAMDGPUFeatureMap(StringRef GPU, const Triple &T,
       Features["s-memrealtime"] = true;
       Features["s-memtime-inst"] = true;
       Features["gws"] = true;
+      Features["vmem-to-lds-load-insts"] = true;
       break;
     case GK_GFX1012:
     case GK_GFX1011:
@@ -487,6 +489,7 @@ void AMDGPU::fillAMDGPUFeatureMap(StringRef GPU, const Triple &T,
       Features["s-memrealtime"] = true;
       Features["s-memtime-inst"] = true;
       Features["gws"] = true;
+      Features["vmem-to-lds-load-insts"] = true;
       break;
     case GK_GFX950:
       Features["bitop3-insts"] = true;
@@ -539,6 +542,7 @@ void AMDGPU::fillAMDGPUFeatureMap(StringRef GPU, const Triple &T,
       Features["ci-insts"] = true;
       Features["s-memtime-inst"] = true;
       Features["gws"] = true;
+      Features["vmem-to-lds-load-insts"] = true;
       break;
     case GK_GFX90A:
       Features["gfx90a-insts"] = true;
@@ -591,6 +595,7 @@ void AMDGPU::fillAMDGPUFeatureMap(StringRef GPU, const Triple &T,
       Features["image-insts"] = true;
       Features["s-memtime-inst"] = true;
       Features["gws"] = true;
+      Features["vmem-to-lds-load-insts"] = true;
       break;
     case GK_NONE:
       break;
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.load.lds.err.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.load.lds.err.ll
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.global.load.lds.err.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.global.load.lds.err.ll

Original file line number	Diff line number	Diff line change
`@@ -253,7 +253,7 @@ AMDGPUTargetInfo::AMDGPUTargetInfo(const llvm::Triple &Triple,`
`253`	`253`
`254`	`254`	`MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 64;`
`255`	`255`	`CUMode = !(GPUFeatures & llvm::AMDGPU::FEATURE_WGP);`
`256`		`- for (auto F : {"image-insts", "gws"})`
	`256`	`+ for (auto F : {"image-insts", "gws", "vmem-to-lds-load-insts"})`
`257`	`257`	`ReadOnlyFeatures.insert(F);`
`258`	`258`	`HalfArgsAndReturns = true;`
`259`	`259`	`}`