Index: lib/Target/AMDGPU/AMDGPUISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -2521,6 +2521,27 @@
   return Opc == ISD::CTLZ || Opc == ISD::CTLZ_ZERO_UNDEF;
 }
 
+// Get FFBH node if the incoming op may have been type legalized from a smaller
+// type VT.
+// Need to match pre-legalized type because the generic legalization inserts the
+// add/sub between the select and compare.
+static SDValue getFFBH_U32(const TargetLowering &TLI,
+                           SelectionDAG &DAG, SDLoc SL, SDValue Op) {
+  EVT VT = Op.getValueType();
+  EVT LegalVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
+  if (LegalVT != MVT::i32)
+    return SDValue();
+
+  if (VT != MVT::i32)
+    Op = DAG.getNode(ISD::ZERO_EXTEND, SL, MVT::i32, Op);
+
+  SDValue FFBH = DAG.getNode(AMDGPUISD::FFBH_U32, SL, MVT::i32, Op);
+  if (VT != MVT::i32)
+    FFBH = DAG.getNode(ISD::TRUNCATE, SL, VT, FFBH);
+
+  return FFBH;
+}
+
 // The native instructions return -1 on 0 input. Optimize out a select that
 // produces -1 on 0.
 //
@@ -2546,7 +2567,7 @@
       isCtlzOpc(RHS.getOpcode()) &&
       RHS.getOperand(0) == CmpLHS &&
       isNegativeOne(LHS)) {
-    return DAG.getNode(AMDGPUISD::FFBH_U32, SL, MVT::i32, CmpLHS);
+    return getFFBH_U32(*this, DAG, SL, CmpLHS);
   }
 
   // select (setcc x, 0, ne), (ctlz_zero_undef x), -1 -> ffbh_u32 x
@@ -2554,7 +2575,7 @@
       isCtlzOpc(LHS.getOpcode()) &&
       LHS.getOperand(0) == CmpLHS &&
       isNegativeOne(RHS)) {
-    return DAG.getNode(AMDGPUISD::FFBH_U32, SL, MVT::i32, CmpLHS);
+    return getFFBH_U32(*this, DAG, SL, CmpLHS);
   }
 
   return SDValue();
@@ -2578,10 +2599,7 @@
     return CombineFMinMaxLegacy(SDLoc(N), VT, LHS, RHS, True, False, CC, DCI);
 
   // There's no reason to not do this if the condition has other uses.
-  if (VT == MVT::i32)
-    return performCtlzCombine(SDLoc(N), Cond, True, False, DCI);
-
-  return SDValue();
+  return performCtlzCombine(SDLoc(N), Cond, True, False, DCI);
 }
 
 SDValue AMDGPUTargetLowering::PerformDAGCombine(SDNode *N,
Index: test/CodeGen/AMDGPU/ctlz.ll
===================================================================
--- test/CodeGen/AMDGPU/ctlz.ll
+++ test/CodeGen/AMDGPU/ctlz.ll
@@ -2,6 +2,10 @@
 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 
+declare i7 @llvm.ctlz.i7(i7, i1) nounwind readnone
+declare i8 @llvm.ctlz.i8(i8, i1) nounwind readnone
+declare i16 @llvm.ctlz.i16(i16, i1) nounwind readnone
+
 declare i32 @llvm.ctlz.i32(i32, i1) nounwind readnone
 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1) nounwind readnone
 declare <4 x i32> @llvm.ctlz.v4i32(<4 x i32>, i1) nounwind readnone
@@ -92,6 +96,20 @@
   ret void
 }
 
+; FUNC-LABEL: {{^}}v_ctlz_i8:
+; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
+; SI-DAG: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; SI-DAG: v_cmp_eq_i32_e32 vcc, 0, [[CTLZ]]
+; SI-DAG: v_cndmask_b32_e64 [[CORRECTED_FFBH:v[0-9]+]], [[FFBH]], 32, vcc
+; SI: v_add_i32_e32 [[RESULT:v[0-9]+]], vcc, 0xffffffe8, [[CORRECTED_FFBH]]
+; SI: buffer_store_byte [[RESULT]],
+define void @v_ctlz_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
+  %val = load i8, i8 addrspace(1)* %valptr
+  %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 false) nounwind readnone
+  store i8 %ctlz, i8 addrspace(1)* %out
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}s_ctlz_i64:
 ; SI: s_load_dwordx2 s{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
 ; SI-DAG: v_cmp_eq_i32_e64 vcc, 0, s[[HI]]
@@ -209,3 +227,43 @@
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
+
+; FUNC-LABEL: {{^}}v_ctlz_i8_sel_eq_neg1:
+; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
+; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; SI: buffer_store_byte [[FFBH]],
+ define void @v_ctlz_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
+  %val = load i8, i8 addrspace(1)* %valptr
+  %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 false) nounwind readnone
+  %cmp = icmp eq i8 %val, 0
+  %sel = select i1 %cmp, i8 -1, i8 %ctlz
+  store i8 %sel, i8 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v_ctlz_i16_sel_eq_neg1:
+; SI: buffer_load_ushort [[VAL:v[0-9]+]],
+; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; SI: buffer_store_short [[FFBH]],
+ define void @v_ctlz_i16_sel_eq_neg1(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %valptr) nounwind {
+  %val = load i16, i16 addrspace(1)* %valptr
+  %ctlz = call i16 @llvm.ctlz.i16(i16 %val, i1 false) nounwind readnone
+  %cmp = icmp eq i16 %val, 0
+  %sel = select i1 %cmp, i16 -1, i16 %ctlz
+  store i16 %sel, i16 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v_ctlz_i7_sel_eq_neg1:
+; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
+; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; SI: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0x7f, [[FFBH]]
+; SI: buffer_store_byte [[TRUNC]],
+ define void @v_ctlz_i7_sel_eq_neg1(i7 addrspace(1)* noalias %out, i7 addrspace(1)* noalias %valptr) nounwind {
+  %val = load i7, i7 addrspace(1)* %valptr
+  %ctlz = call i7 @llvm.ctlz.i7(i7 %val, i1 false) nounwind readnone
+  %cmp = icmp eq i7 %val, 0
+  %sel = select i1 %cmp, i7 -1, i7 %ctlz
+  store i7 %sel, i7 addrspace(1)* %out
+  ret void
+}
Index: test/CodeGen/AMDGPU/ctlz_zero_undef.ll
===================================================================
--- test/CodeGen/AMDGPU/ctlz_zero_undef.ll
+++ test/CodeGen/AMDGPU/ctlz_zero_undef.ll
@@ -78,6 +78,18 @@
   ret void
 }
 
+; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i8:
+; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
+; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; SI: v_add_i32_e32 [[RESULT:v[0-9]+]], vcc, 0xffffffe8, [[FFBH]]
+; SI: buffer_store_byte [[RESULT]],
+define void @v_ctlz_zero_undef_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
+  %val = load i8, i8 addrspace(1)* %valptr
+  %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 true) nounwind readnone
+  store i8 %ctlz, i8 addrspace(1)* %out
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}s_ctlz_zero_undef_i64:
 ; SI: s_load_dwordx2 s{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
 ; SI-DAG: v_cmp_eq_i32_e64 vcc, 0, s[[HI]]
@@ -160,6 +172,19 @@
   ret void
 }
 
+; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i8_sel_eq_neg1:
+; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
+; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; SI: buffer_store_byte [[FFBH]],
+ define void @v_ctlz_zero_undef_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
+  %val = load i8, i8 addrspace(1)* %valptr
+  %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 true) nounwind readnone
+  %cmp = icmp eq i8 %val, 0
+  %sel = select i1 %cmp, i8 -1, i8 %ctlz
+  store i8 %sel, i8 addrspace(1)* %out
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32_sel_eq_neg1_two_use:
 ; SI: buffer_load_dword [[VAL:v[0-9]+]],
 ; SI-DAG: v_ffbh_u32_e32 [[RESULT0:v[0-9]+]], [[VAL]]
@@ -241,15 +266,3 @@
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
-
-; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i8:
-; SI: buffer_load_dword [[VAL:v[0-9]+]],
-; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
-; SI: v_add_i32_e32 [[RESULT:v[0-9]+]], vcc, 0xffffffe8, [[FFBH]]
-; SI: buffer_store_dword [[RESULT]],
-define void @v_ctlz_zero_undef_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
-  %val = load i8, i8 addrspace(1)* %valptr
-  %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 true) nounwind readnone
-  store i8 %ctlz, i8 addrspace(1)* %out
-  ret void
-}