Index: lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -184,9 +184,11 @@
 }
 
 bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
+  const AMDGPUMachineFunction *MFI = MF.getInfo<AMDGPUMachineFunction>();
 
   // The starting address of all shader programs must be 256 bytes aligned.
-  MF.setAlignment(8);
+  // Regular functions just need the basic required instruction alignment.
+  MF.setAlignment(MFI->isEntryFunction() ? 8 : 2);
 
   SetupMachineFunction(MF);
 
Index: test/CodeGen/AMDGPU/hsa-func-align.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AMDGPU/hsa-func-align.ll
@@ -0,0 +1,18 @@
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri < %s | FileCheck -check-prefix=HSA %s
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -filetype=obj < %s | llvm-readobj -symbols -s -sd | FileCheck -check-prefix=ELF %s
+
+; ELF: Section {
+; ELF: Name: .text
+; ELF: SHF_ALLOC (0x2)
+; ELF: SHF_EXECINSTR (0x4)
+; ELF: AddressAlignment: 32
+; ELF: }
+
+; HSA: .globl simple_align16
+; HSA: .p2align 5
+define void @simple_align16(i32 addrspace(1)* addrspace(2)* %ptr.out) align 32 {
+entry:
+  %out = load i32 addrspace(1)*, i32 addrspace(1)* addrspace(2)* %ptr.out
+  store i32 0, i32 addrspace(1)* %out
+  ret void
+}
Index: test/CodeGen/AMDGPU/hsa-func.ll
===================================================================
--- test/CodeGen/AMDGPU/hsa-func.ll
+++ test/CodeGen/AMDGPU/hsa-func.ll
@@ -14,6 +14,7 @@
 ; ELF: Flags [ (0x6)
 ; ELF: SHF_ALLOC (0x2)
 ; ELF: SHF_EXECINSTR (0x4)
+; ELF: AddressAlignment: 4
 ; ELF: }
 
 ; ELF: SHT_NOTE
@@ -36,6 +37,8 @@
 ; HSA-VI: .hsa_code_object_isa 8,0,1,"AMD","AMDGPU"
 
 ; HSA-NOT: .amdgpu_hsa_kernel simple
+; HSA: .globl simple
+; HSA: .p2align 2
 ; HSA: {{^}}simple:
 ; HSA: .amd_kernel_code_t
 ; HSA: enable_sgpr_private_segment_buffer = 0
@@ -59,3 +62,13 @@
   store i32 0, i32 addrspace(1)* %out
   ret void
 }
+
+; Ignore explicit alignment that is too low.
+; HSA: .globl simple_align2
+; HSA: .p2align 2
+define void @simple_align2(i32 addrspace(1)* addrspace(2)* %ptr.out) align 2 {
+entry:
+  %out = load i32 addrspace(1)*, i32 addrspace(1)* addrspace(2)* %ptr.out
+  store i32 0, i32 addrspace(1)* %out
+  ret void
+}