Index: llvm/lib/Target/AMDGPU/AMDGPUPropagateAttributes.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPUPropagateAttributes.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUPropagateAttributes.cpp
@@ -32,6 +32,7 @@
 #include "Utils/AMDGPUBaseInfo.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/SmallVector.h"
+#include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Target/TargetMachine.h"
@@ -56,8 +57,10 @@
 };
 
 // Attributes to propagate.
+// TODO: Support conservative min/max merging instead of cloning.
 static constexpr const char* AttributeNames[] = {
-  "amdgpu-waves-per-eu"
+  "amdgpu-waves-per-eu",
+  "amdgpu-flat-work-group-size"
 };
 
 static constexpr unsigned NumAttr =
@@ -371,15 +374,28 @@
 }
 
 bool AMDGPUPropagateAttributesEarly::runOnFunction(Function &F) {
-  if (!TM || !AMDGPU::isEntryFunctionCC(F.getCallingConv()))
+  if (!TM) {
+    auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
+    if (!TPC)
+      return false;
+
+    TM = &TPC->getTM<TargetMachine>();
+  }
+
+  if (!AMDGPU::isEntryFunctionCC(F.getCallingConv()))
     return false;
 
   return AMDGPUPropagateAttributes(TM, false).process(F);
 }
 
 bool AMDGPUPropagateAttributesLate::runOnModule(Module &M) {
-  if (!TM)
-    return false;
+  if (!TM) {
+    auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
+    if (!TPC)
+      return false;
+
+    TM = &TPC->getTM<TargetMachine>();
+  }
 
   return AMDGPUPropagateAttributes(TM, true).process(M);
 }
Index: llvm/test/CodeGen/AMDGPU/propagate-attributes-flat-work-group-size.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/AMDGPU/propagate-attributes-flat-work-group-size.ll
@@ -0,0 +1,48 @@
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-propagate-attributes-late %s | FileCheck %s
+
+; CHECK: define internal void @max_flat_1_1024() #0 {
+define internal void @max_flat_1_1024() #0 {
+  ret void
+}
+
+; CHECK: define internal void @max_flat_1_256() #1 {
+define internal void @max_flat_1_256() #1 {
+  ret void
+}
+
+; CHECK: define amdgpu_kernel void @kernel_1_256_call_default() #1 {
+define amdgpu_kernel void @kernel_1_256_call_default() #1 {
+  call void @default()
+  ret void
+}
+
+; CHECK: define amdgpu_kernel void @kernel_1_256_call_1_256() #1 {
+define amdgpu_kernel void @kernel_1_256_call_1_256() #1 {
+  call void @max_flat_1_256()
+  ret void
+}
+
+; CHECK: define amdgpu_kernel void @kernel_1_256_call_64_64() #1 {
+define amdgpu_kernel void @kernel_1_256_call_64_64() #1 {
+  call void @max_flat_64_64()
+  ret void
+}
+
+; CHECK: define internal void @max_flat_64_64() #2 {
+define internal void @max_flat_64_64() #2 {
+  ret void
+}
+
+; CHECK: define internal void @default() #2 {
+define internal void @default() #3 {
+  ret void
+}
+
+attributes #0 = { noinline "amdgpu-flat-work-group-size"="1,1024" }
+attributes #1 = { noinline "amdgpu-flat-work-group-size"="1,256" }
+attributes #2 = { noinline "amdgpu-flat-work-group-size"="64,64" }
+attributes #3 = { noinline }
+
+; CHECK: attributes #0 = { noinline "amdgpu-flat-work-group-size"="1,1024"
+; CHECK-NEXT: attributes #1 = { noinline "amdgpu-flat-work-group-size"="1,256"
+; CHECK-NEXT: attributes #2 = { noinline "amdgpu-flat-work-group-size"="1,256"