Index: lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -763,6 +763,11 @@
   header.reserved_vgpr_first = KernelInfo.ReservedVGPRFirst;
   header.reserved_vgpr_count = KernelInfo.ReservedVGPRCount;
 
+  // These alignment values are specified in powers of two, so alignment =
+  // 2^n.  The minimum alignment is 2^4 = 16.
+  header.kernarg_segment_alignment = std::max((size_t)4,
+      countTrailingZeros(MFI->getMaxKernArgAlign()));
+
   if (STM.debuggerEmitPrologue()) {
     header.debug_wavefront_private_segment_offset_sgpr =
       KernelInfo.DebuggerWavefrontPrivateSegmentOffsetSGPR;
Index: lib/Target/AMDGPU/AMDGPUMachineFunction.h
===================================================================
--- lib/Target/AMDGPU/AMDGPUMachineFunction.h
+++ lib/Target/AMDGPU/AMDGPUMachineFunction.h
@@ -50,6 +50,10 @@
     return KernArgSize;
   }
 
+  uint64_t getMaxKernArgAlign() const {
+    return MaxKernArgAlign;
+  }
+
   void setABIArgOffset(unsigned NewOffset) {
     ABIArgOffset = NewOffset;
   }
Index: test/CodeGen/AMDGPU/kernel-args.ll
===================================================================
--- test/CodeGen/AMDGPU/kernel-args.ll
+++ test/CodeGen/AMDGPU/kernel-args.ll
@@ -5,6 +5,7 @@
 ; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s --check-prefix=EG --check-prefix=FUNC
 
 ; FUNC-LABEL: {{^}}i8_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: AND_INT {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
 ; SI: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xb
 ; MESA-VI: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0x2c
@@ -24,6 +25,7 @@
 }
 
 ; FUNC-LABEL: {{^}}i8_zext_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
 ; SI: s_load_dword s{{[0-9]}}, s[0:1], 0xb
 ; MESA-VI: s_load_dword s{{[0-9]}}, s[0:1], 0x2c
@@ -42,6 +44,7 @@
 }
 
 ; FUNC-LABEL: {{^}}i8_sext_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
 ; SI: s_load_dword s{{[0-9]}}, s[0:1], 0xb
 ; MESA-VI: s_load_dword s{{[0-9]}}, s[0:1], 0x2c
@@ -60,6 +63,7 @@
 }
 
 ; FUNC-LABEL: {{^}}i16_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: AND_INT {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
 ; SI: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xb
 ; MESA-VI: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0x2c
@@ -79,6 +83,7 @@
 }
 
 ; FUNC-LABEL: {{^}}i16_zext_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
 ; SI: s_load_dword s{{[0-9]}}, s[0:1], 0xb
 ; MESA-VI: s_load_dword s{{[0-9]}}, s[0:1], 0x2c
@@ -97,6 +102,7 @@
 }
 
 ; FUNC-LABEL: {{^}}i16_sext_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
 ; SI: s_load_dword s{{[0-9]}}, s[0:1], 0xb
 ; MESA-VI: s_load_dword s{{[0-9]}}, s[0:1], 0x2c
@@ -115,6 +121,7 @@
 }
 
 ; FUNC-LABEL: {{^}}i32_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: T{{[0-9]\.[XYZW]}}, KC0[2].Z
 ; SI: s_load_dword s{{[0-9]}}, s[0:1], 0xb
 ; MESA-VI: s_load_dword s{{[0-9]}}, s[0:1], 0x2c
@@ -126,6 +133,7 @@
 }
 
 ; FUNC-LABEL: {{^}}f32_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: T{{[0-9]\.[XYZW]}}, KC0[2].Z
 ; SI: s_load_dword s{{[0-9]}}, s[0:1], 0xb
 ; MESA-VI: s_load_dword s{{[0-9]}}, s[0:1], 0x2c
@@ -137,6 +145,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v2i8_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: VTX_READ_8
 ; EG: VTX_READ_8
 ; MESA-GCN: buffer_load_ubyte
@@ -150,6 +159,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v2i16_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: VTX_READ_16
 ; EG: VTX_READ_16
 ; MESA-GCN: buffer_load_ushort
@@ -163,6 +173,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v2i32_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].X
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[2].W
 ; SI: s_load_dwordx2 s{{\[[0-9]:[0-9]\]}}, s[0:1], 0xb
@@ -175,6 +186,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v2f32_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].X
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[2].W
 ; SI: s_load_dwordx2 s{{\[[0-9]:[0-9]\]}}, s[0:1], 0xb
@@ -187,6 +199,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v3i8_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG-DAG: VTX_READ_8 T{{[0-9]}}.X, T{{[0-9]}}.X, 40
 ; EG-DAG: VTX_READ_8 T{{[0-9]}}.X, T{{[0-9]}}.X, 41
 ; EG-DAG: VTX_READ_8 T{{[0-9]}}.X, T{{[0-9]}}.X, 42
@@ -203,6 +216,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v3i16_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG-DAG: VTX_READ_16 T{{[0-9]}}.X, T{{[0-9]}}.X, 44
 ; EG-DAG: VTX_READ_16 T{{[0-9]}}.X, T{{[0-9]}}.X, 46
 ; EG-DAG: VTX_READ_16 T{{[0-9]}}.X, T{{[0-9]}}.X, 48
@@ -218,6 +232,7 @@
   ret void
 }
 ; FUNC-LABEL: {{^}}v3i32_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Y
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Z
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].W
@@ -231,6 +246,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v3f32_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Y
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Z
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].W
@@ -244,6 +260,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v4i8_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: VTX_READ_8
 ; EG: VTX_READ_8
 ; EG: VTX_READ_8
@@ -263,6 +280,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v4i16_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: VTX_READ_16
 ; EG: VTX_READ_16
 ; EG: VTX_READ_16
@@ -282,6 +300,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v4i32_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Y
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Z
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].W
@@ -296,6 +315,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v4f32_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Y
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Z
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].W
@@ -310,6 +330,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v8i8_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: VTX_READ_8
 ; EG: VTX_READ_8
 ; EG: VTX_READ_8
@@ -340,6 +361,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v8i16_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: VTX_READ_16
 ; EG: VTX_READ_16
 ; EG: VTX_READ_16
@@ -371,6 +393,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v8i32_arg:
+; HSA-VI: kernarg_segment_alignment = 5
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].Y
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].Z
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].W
@@ -389,6 +412,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v8f32_arg:
+; HSA-VI: kernarg_segment_alignment = 5
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].Y
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].Z
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].W
@@ -405,6 +429,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v16i8_arg:
+; HSA-VI: kernarg_segment_alignment = 4
 ; EG: VTX_READ_8
 ; EG: VTX_READ_8
 ; EG: VTX_READ_8
@@ -460,6 +485,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v16i16_arg:
+; HSA-VI: kernarg_segment_alignment = 5
 ; EG: VTX_READ_16
 ; EG: VTX_READ_16
 ; EG: VTX_READ_16
@@ -515,6 +541,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v16i32_arg:
+; HSA-VI: kernarg_segment_alignment = 6
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].Y
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].Z
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].W
@@ -541,6 +568,7 @@
 }
 
 ; FUNC-LABEL: {{^}}v16f32_arg:
+; HSA-VI: kernarg_segment_alignment = 6
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].Y
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].Z
 ; EG-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].W