diff --git a/lib/Target/AMDGPU/SIISelLowering.cpp b/lib/Target/AMDGPU/SIISelLowering.cpp
index 6fde731..0c706d5 100644
--- a/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -16062,7 +16062,7 @@
       if (HasSystemScope)
         return AtomicExpansionKind::CmpXChg;
 
-      if ((AS == AMDGPUAS::GLOBAL_ADDRESS ||
+      if ((AMDGPU::isExtendedGlobalAddrSpace(AS) ||
            AS == AMDGPUAS::BUFFER_FAT_POINTER) &&
           Ty->isFloatTy()) {
         // global/buffer atomic fadd f32 no-rtn: gfx908, gfx90a, gfx940, gfx11+.
diff --git a/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll b/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll
index a5d4c32..9010dec 100644
--- a/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll
+++ b/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll
@@ -59,6 +59,112 @@
   ret void
 }
 
+define void @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe(ptr addrspace(7) %ptr, float %value) #0 {
+; CI-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe(
+; CI-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; CI-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; CI:       atomicrmw.start:
+; CI-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; CI-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; CI-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; CI-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; CI-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; CI-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; CI-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; CI-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; CI-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; CI:       atomicrmw.end:
+; CI-NEXT:    ret void
+;
+; GFX9-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe(
+; GFX9-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; GFX9-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX9:       atomicrmw.start:
+; GFX9-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX9-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX9-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX9-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX9-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; GFX9-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX9-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX9-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX9-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX9:       atomicrmw.end:
+; GFX9-NEXT:    ret void
+;
+; GFX908-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe(
+; GFX908-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX908-NEXT:    ret void
+;
+; GFX90A-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe(
+; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX90A-NEXT:    ret void
+;
+; GFX940-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe(
+; GFX940-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX940-NEXT:    ret void
+;
+; GFX11-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe(
+; GFX11-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX11-NEXT:    ret void
+;
+  %res = atomicrmw fadd ptr addrspace(7) %ptr, float %value syncscope("wavefront") monotonic
+  ret void
+}
+
+define void @test_atomicrmw_fadd_f32_as999_no_use_unsafe(ptr addrspace(999) %ptr, float %value) #0 {
+; CI-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe(
+; CI-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; CI-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; CI:       atomicrmw.start:
+; CI-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; CI-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; CI-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; CI-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; CI-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; CI-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; CI-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; CI-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; CI-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; CI:       atomicrmw.end:
+; CI-NEXT:    ret void
+;
+; GFX9-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe(
+; GFX9-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; GFX9-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX9:       atomicrmw.start:
+; GFX9-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX9-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX9-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX9-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX9-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; GFX9-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX9-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX9-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX9-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX9:       atomicrmw.end:
+; GFX9-NEXT:    ret void
+;
+; GFX908-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe(
+; GFX908-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX908-NEXT:    ret void
+;
+; GFX90A-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe(
+; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX90A-NEXT:    ret void
+;
+; GFX940-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe(
+; GFX940-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX940-NEXT:    ret void
+;
+; GFX11-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe(
+; GFX11-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX11-NEXT:    ret void
+;
+  %res = atomicrmw fadd ptr addrspace(999) %ptr, float %value syncscope("wavefront") monotonic
+  ret void
+}
+
 define float @test_atomicrmw_fadd_f32_global_unsafe(ptr addrspace(1) %ptr, float %value) #0 {
 ; CI-LABEL: @test_atomicrmw_fadd_f32_global_unsafe(
 ; CI-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4
@@ -124,6 +230,136 @@
   ret float %res
 }
 
+define float @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe(ptr addrspace(7) %ptr, float %value) #0 {
+; CI-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe(
+; CI-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; CI-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; CI:       atomicrmw.start:
+; CI-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; CI-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; CI-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; CI-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; CI-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; CI-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; CI-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; CI-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; CI-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; CI:       atomicrmw.end:
+; CI-NEXT:    ret float [[TMP5]]
+;
+; GFX9-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe(
+; GFX9-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; GFX9-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX9:       atomicrmw.start:
+; GFX9-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX9-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX9-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX9-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX9-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; GFX9-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX9-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX9-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX9-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX9:       atomicrmw.end:
+; GFX9-NEXT:    ret float [[TMP5]]
+;
+; GFX908-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe(
+; GFX908-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; GFX908-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX908:       atomicrmw.start:
+; GFX908-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX908-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX908-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX908-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX908-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; GFX908-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX908-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX908-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX908-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX908:       atomicrmw.end:
+; GFX908-NEXT:    ret float [[TMP5]]
+;
+; GFX90A-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe(
+; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX90A-NEXT:    ret float [[RES]]
+;
+; GFX940-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe(
+; GFX940-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX940-NEXT:    ret float [[RES]]
+;
+; GFX11-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe(
+; GFX11-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX11-NEXT:    ret float [[RES]]
+;
+  %res = atomicrmw fadd ptr addrspace(7) %ptr, float %value syncscope("wavefront") monotonic
+  ret float %res
+}
+
+define float @test_atomicrmw_fadd_f32_as999_unsafe(ptr addrspace(999) %ptr, float %value) #0 {
+; CI-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe(
+; CI-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; CI-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; CI:       atomicrmw.start:
+; CI-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; CI-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; CI-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; CI-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; CI-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; CI-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; CI-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; CI-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; CI-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; CI:       atomicrmw.end:
+; CI-NEXT:    ret float [[TMP5]]
+;
+; GFX9-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe(
+; GFX9-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; GFX9-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX9:       atomicrmw.start:
+; GFX9-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX9-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX9-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX9-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX9-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; GFX9-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX9-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX9-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX9-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX9:       atomicrmw.end:
+; GFX9-NEXT:    ret float [[TMP5]]
+;
+; GFX908-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe(
+; GFX908-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; GFX908-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX908:       atomicrmw.start:
+; GFX908-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX908-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX908-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX908-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX908-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4
+; GFX908-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX908-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX908-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX908-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX908:       atomicrmw.end:
+; GFX908-NEXT:    ret float [[TMP5]]
+;
+; GFX90A-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe(
+; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX90A-NEXT:    ret float [[RES]]
+;
+; GFX940-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe(
+; GFX940-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX940-NEXT:    ret float [[RES]]
+;
+; GFX11-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe(
+; GFX11-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
+; GFX11-NEXT:    ret float [[RES]]
+;
+  %res = atomicrmw fadd ptr addrspace(999) %ptr, float %value syncscope("wavefront") monotonic
+  ret float %res
+}
+
 define double @test_atomicrmw_fadd_f64_global_unsafe(ptr addrspace(1) %ptr, double %value) #0 {
 ; CI-LABEL: @test_atomicrmw_fadd_f64_global_unsafe(
 ; CI-NEXT:    [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8
@@ -545,6 +781,184 @@
   ret float %res
 }
 
+define float @test_atomicrmw_fadd_f32_buffer_fat_ptr(ptr addrspace(7) %ptr, float %value) {
+; CI-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr(
+; CI-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; CI-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; CI:       atomicrmw.start:
+; CI-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; CI-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; CI-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; CI-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; CI-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; CI-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; CI-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; CI-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; CI-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; CI:       atomicrmw.end:
+; CI-NEXT:    ret float [[TMP5]]
+;
+; GFX9-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr(
+; GFX9-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; GFX9-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX9:       atomicrmw.start:
+; GFX9-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX9-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX9-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX9-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX9-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; GFX9-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX9-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX9-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX9-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX9:       atomicrmw.end:
+; GFX9-NEXT:    ret float [[TMP5]]
+;
+; GFX908-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr(
+; GFX908-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; GFX908-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX908:       atomicrmw.start:
+; GFX908-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX908-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX908-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX908-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX908-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; GFX908-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX908-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX908-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX908-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX908:       atomicrmw.end:
+; GFX908-NEXT:    ret float [[TMP5]]
+;
+; GFX90A-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr(
+; GFX90A-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; GFX90A-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX90A:       atomicrmw.start:
+; GFX90A-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX90A-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX90A-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX90A-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX90A-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; GFX90A-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX90A-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX90A-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX90A-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX90A:       atomicrmw.end:
+; GFX90A-NEXT:    ret float [[TMP5]]
+;
+; GFX940-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr(
+; GFX940-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4
+; GFX940-NEXT:    ret float [[RES]]
+;
+; GFX11-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr(
+; GFX11-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4
+; GFX11-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX11:       atomicrmw.start:
+; GFX11-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX11-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX11-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX11-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX11-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; GFX11-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX11-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX11-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX11-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX11:       atomicrmw.end:
+; GFX11-NEXT:    ret float [[TMP5]]
+;
+  %res = atomicrmw fadd ptr addrspace(7) %ptr, float %value seq_cst
+  ret float %res
+}
+
+define float @test_atomicrmw_fadd_f32_as999(ptr addrspace(999) %ptr, float %value) {
+; CI-LABEL: @test_atomicrmw_fadd_f32_as999(
+; CI-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; CI-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; CI:       atomicrmw.start:
+; CI-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; CI-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; CI-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; CI-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; CI-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; CI-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; CI-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; CI-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; CI-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; CI:       atomicrmw.end:
+; CI-NEXT:    ret float [[TMP5]]
+;
+; GFX9-LABEL: @test_atomicrmw_fadd_f32_as999(
+; GFX9-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; GFX9-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX9:       atomicrmw.start:
+; GFX9-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX9-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX9-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX9-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX9-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; GFX9-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX9-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX9-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX9-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX9:       atomicrmw.end:
+; GFX9-NEXT:    ret float [[TMP5]]
+;
+; GFX908-LABEL: @test_atomicrmw_fadd_f32_as999(
+; GFX908-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; GFX908-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX908:       atomicrmw.start:
+; GFX908-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX908-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX908-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX908-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX908-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; GFX908-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX908-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX908-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX908-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX908:       atomicrmw.end:
+; GFX908-NEXT:    ret float [[TMP5]]
+;
+; GFX90A-LABEL: @test_atomicrmw_fadd_f32_as999(
+; GFX90A-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; GFX90A-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX90A:       atomicrmw.start:
+; GFX90A-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX90A-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX90A-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX90A-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX90A-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; GFX90A-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX90A-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX90A-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX90A-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX90A:       atomicrmw.end:
+; GFX90A-NEXT:    ret float [[TMP5]]
+;
+; GFX940-LABEL: @test_atomicrmw_fadd_f32_as999(
+; GFX940-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4
+; GFX940-NEXT:    ret float [[RES]]
+;
+; GFX11-LABEL: @test_atomicrmw_fadd_f32_as999(
+; GFX11-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4
+; GFX11-NEXT:    br label [[ATOMICRMW_START:%.*]]
+; GFX11:       atomicrmw.start:
+; GFX11-NEXT:    [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
+; GFX11-NEXT:    [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]]
+; GFX11-NEXT:    [[TMP2:%.*]] = bitcast float [[NEW]] to i32
+; GFX11-NEXT:    [[TMP3:%.*]] = bitcast float [[LOADED]] to i32
+; GFX11-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
+; GFX11-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
+; GFX11-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
+; GFX11-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to float
+; GFX11-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
+; GFX11:       atomicrmw.end:
+; GFX11-NEXT:    ret float [[TMP5]]
+;
+  %res = atomicrmw fadd ptr addrspace(999) %ptr, float %value seq_cst
+  ret float %res
+}
+
 define void @test_atomicrmw_fadd_f32_global_no_use_ieee(ptr addrspace(1) %ptr, float %value) {
 ; CI-LABEL: @test_atomicrmw_fadd_f32_global_no_use_ieee(
 ; CI-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4