[AMDGPU] Promote uniform i16 ops to i32 ops for targets that have 16 bit instructions

Differential Revision: https://reviews.llvm.org/D24125

llvm-svn: 282624

5 files changed, 1101 insertions, 147 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll
index d78c75165be..d78c75165be 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-i16-to-i32.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-i16-to-i32.ll
new file mode 100644
index 00000000000..ed512b4ddd9
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-i16-to-i32.ll
@@ -0,0 +1,856 @@
+; RUN: opt -S -mtriple=amdgcn-- -amdgpu-codegenprepare %s | FileCheck -check-prefix=SI %s
+; RUN: opt -S -mtriple=amdgcn-- -mcpu=tonga -amdgpu-codegenprepare %s | FileCheck -check-prefix=VI %s
+
+; SI-NOT: zext
+; SI-NOT: sext
+; SI-NOT: trunc
+
+; VI-LABEL: @add_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = add i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @add_i16(i16 %a, i16 %b) {
+  %r = add i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @add_nsw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = add nsw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @add_nsw_i16(i16 %a, i16 %b) {
+  %r = add nsw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @add_nuw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = add nuw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @add_nuw_i16(i16 %a, i16 %b) {
+  %r = add nuw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @add_nuw_nsw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = add nuw nsw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @add_nuw_nsw_i16(i16 %a, i16 %b) {
+  %r = add nuw nsw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @sub_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = sub i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @sub_i16(i16 %a, i16 %b) {
+  %r = sub i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @sub_nsw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = sub nsw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @sub_nsw_i16(i16 %a, i16 %b) {
+  %r = sub nsw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @sub_nuw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = sub nuw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @sub_nuw_i16(i16 %a, i16 %b) {
+  %r = sub nuw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @sub_nuw_nsw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = sub nuw nsw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @sub_nuw_nsw_i16(i16 %a, i16 %b) {
+  %r = sub nuw nsw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @mul_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = mul i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @mul_i16(i16 %a, i16 %b) {
+  %r = mul i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @mul_nsw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = mul nsw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @mul_nsw_i16(i16 %a, i16 %b) {
+  %r = mul nsw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @mul_nuw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = mul nuw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @mul_nuw_i16(i16 %a, i16 %b) {
+  %r = mul nuw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @mul_nuw_nsw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = mul nuw nsw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @mul_nuw_nsw_i16(i16 %a, i16 %b) {
+  %r = mul nuw nsw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @urem_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = urem i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @urem_i16(i16 %a, i16 %b) {
+  %r = urem i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @srem_i16(
+; VI: %[[A_32:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = sext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = srem i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @srem_i16(i16 %a, i16 %b) {
+  %r = srem i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @shl_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = shl i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @shl_i16(i16 %a, i16 %b) {
+  %r = shl i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @shl_nsw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = shl nsw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @shl_nsw_i16(i16 %a, i16 %b) {
+  %r = shl nsw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @shl_nuw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = shl nuw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @shl_nuw_i16(i16 %a, i16 %b) {
+  %r = shl nuw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @shl_nuw_nsw_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = shl nuw nsw i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @shl_nuw_nsw_i16(i16 %a, i16 %b) {
+  %r = shl nuw nsw i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @lshr_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = lshr i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @lshr_i16(i16 %a, i16 %b) {
+  %r = lshr i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @lshr_exact_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = lshr exact i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @lshr_exact_i16(i16 %a, i16 %b) {
+  %r = lshr exact i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @ashr_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = ashr i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @ashr_i16(i16 %a, i16 %b) {
+  %r = ashr i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @ashr_exact_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = ashr exact i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @ashr_exact_i16(i16 %a, i16 %b) {
+  %r = ashr exact i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @and_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = and i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @and_i16(i16 %a, i16 %b) {
+  %r = and i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @or_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = or i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @or_i16(i16 %a, i16 %b) {
+  %r = or i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @xor_i16(
+; VI: %[[A_32:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32:[0-9]+]] = zext i16 %b to i32
+; VI: %[[R_32:[0-9]+]] = xor i32 %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc i32 %[[R_32]] to i16
+; VI: ret i16 %[[R_16]]
+define i16 @xor_i16(i16 %a, i16 %b) {
+  %r = xor i16 %a, %b
+  ret i16 %r
+}
+
+; VI-LABEL: @select_eq_i16(
+; VI: %[[A_32_0:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = zext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp eq i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = zext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_eq_i16(i16 %a, i16 %b) {
+  %cmp = icmp eq i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_ne_i16(
+; VI: %[[A_32_0:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = zext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp ne i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = zext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_ne_i16(i16 %a, i16 %b) {
+  %cmp = icmp ne i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_ugt_i16(
+; VI: %[[A_32_0:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = zext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp ugt i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = zext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_ugt_i16(i16 %a, i16 %b) {
+  %cmp = icmp ugt i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_uge_i16(
+; VI: %[[A_32_0:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = zext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp uge i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = zext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_uge_i16(i16 %a, i16 %b) {
+  %cmp = icmp uge i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_ult_i16(
+; VI: %[[A_32_0:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = zext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp ult i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = zext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_ult_i16(i16 %a, i16 %b) {
+  %cmp = icmp ult i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_ule_i16(
+; VI: %[[A_32_0:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = zext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp ule i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = zext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_ule_i16(i16 %a, i16 %b) {
+  %cmp = icmp ule i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_sgt_i16(
+; VI: %[[A_32_0:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = sext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp sgt i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = sext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_sgt_i16(i16 %a, i16 %b) {
+  %cmp = icmp sgt i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_sge_i16(
+; VI: %[[A_32_0:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = sext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp sge i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = sext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_sge_i16(i16 %a, i16 %b) {
+  %cmp = icmp sge i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_slt_i16(
+; VI: %[[A_32_0:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = sext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp slt i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = sext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_slt_i16(i16 %a, i16 %b) {
+  %cmp = icmp slt i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @select_sle_i16(
+; VI: %[[A_32_0:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32_0:[0-9]+]] = sext i16 %b to i32
+; VI: %[[CMP:[0-9]+]] = icmp sle i32 %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = sext i16 %a to i32
+; VI: %[[B_32_1:[0-9]+]] = sext i16 %b to i32
+; VI: %[[SEL_32:[0-9]+]] = select i1 %[[CMP]], i32 %[[A_32_1]], i32 %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc i32 %[[SEL_32]] to i16
+; VI: ret i16 %[[SEL_16]]
+define i16 @select_sle_i16(i16 %a, i16 %b) {
+  %cmp = icmp sle i16 %a, %b
+  %sel = select i1 %cmp, i16 %a, i16 %b
+  ret i16 %sel
+}
+
+; VI-LABEL: @add_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = add <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @add_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = add <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @add_nsw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = add nsw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @add_nsw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = add nsw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @add_nuw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = add nuw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @add_nuw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = add nuw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @add_nuw_nsw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = add nuw nsw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @add_nuw_nsw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = add nuw nsw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @sub_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = sub <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @sub_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = sub <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @sub_nsw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = sub nsw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @sub_nsw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = sub nsw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @sub_nuw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = sub nuw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @sub_nuw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = sub nuw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @sub_nuw_nsw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = sub nuw nsw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @sub_nuw_nsw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = sub nuw nsw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @mul_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = mul <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @mul_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = mul <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @mul_nsw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = mul nsw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @mul_nsw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = mul nsw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @mul_nuw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = mul nuw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @mul_nuw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = mul nuw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @mul_nuw_nsw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = mul nuw nsw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @mul_nuw_nsw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = mul nuw nsw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @urem_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = urem <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @urem_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = urem <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @srem_3xi16(
+; VI: %[[A_32:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = srem <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @srem_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = srem <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @shl_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = shl <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @shl_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = shl <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @shl_nsw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = shl nsw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @shl_nsw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = shl nsw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @shl_nuw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = shl nuw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @shl_nuw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = shl nuw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @shl_nuw_nsw_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = shl nuw nsw <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @shl_nuw_nsw_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = shl nuw nsw <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @lshr_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = lshr <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @lshr_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = lshr <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @lshr_exact_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = lshr exact <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @lshr_exact_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = lshr exact <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @ashr_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = ashr <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @ashr_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = ashr <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @ashr_exact_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = ashr exact <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @ashr_exact_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = ashr exact <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @and_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = and <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @and_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = and <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @or_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = or <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @or_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = or <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @xor_3xi16(
+; VI: %[[A_32:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[R_32:[0-9]+]] = xor <3 x i32> %[[A_32]], %[[B_32]]
+; VI: %[[R_16:[0-9]+]] = trunc <3 x i32> %[[R_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[R_16]]
+define <3 x i16> @xor_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %r = xor <3 x i16> %a, %b
+  ret <3 x i16> %r
+}
+
+; VI-LABEL: @select_eq_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp eq <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_eq_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp eq <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_ne_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp ne <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_ne_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp ne <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_ugt_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp ugt <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_ugt_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp ugt <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_uge_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp uge <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_uge_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp uge <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_ult_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp ult <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_ult_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp ult <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_ule_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp ule <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = zext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = zext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_ule_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp ule <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_sgt_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp sgt <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_sgt_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp sgt <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_sge_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp sge <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_sge_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp sge <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_slt_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp slt <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_slt_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp slt <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
+
+; VI-LABEL: @select_sle_3xi16(
+; VI: %[[A_32_0:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_0:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[CMP:[0-9]+]] = icmp sle <3 x i32> %[[A_32_0]], %[[B_32_0]]
+; VI: %[[A_32_1:[0-9]+]] = sext <3 x i16> %a to <3 x i32>
+; VI: %[[B_32_1:[0-9]+]] = sext <3 x i16> %b to <3 x i32>
+; VI: %[[SEL_32:[0-9]+]] = select <3 x i1> %[[CMP]], <3 x i32> %[[A_32_1]], <3 x i32> %[[B_32_1]]
+; VI: %[[SEL_16:[0-9]+]] = trunc <3 x i32> %[[SEL_32]] to <3 x i16>
+; VI: ret <3 x i16> %[[SEL_16]]
+define <3 x i16> @select_sle_3xi16(<3 x i16> %a, <3 x i16> %b) {
+  %cmp = icmp sle <3 x i16> %a, %b
+  %sel = select <3 x i1> %cmp, <3 x i16> %a, <3 x i16> %b
+  ret <3 x i16> %sel
+}
diff --git a/llvm/test/CodeGen/AMDGPU/ctlz.ll b/llvm/test/CodeGen/AMDGPU/ctlz.ll
index e9d26a225e3..6a8666f9ad4 100644
--- a/llvm/test/CodeGen/AMDGPU/ctlz.ll
+++ b/llvm/test/CodeGen/AMDGPU/ctlz.ll
@@ -1,5 +1,5 @@
-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI -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
@@ -17,13 +17,13 @@ declare <4 x i64> @llvm.ctlz.v4i64(<4 x i64>, i1) nounwind readnone
 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
 
 ; FUNC-LABEL: {{^}}s_ctlz_i32:
-; SI: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
-; SI-DAG: s_flbit_i32_b32 [[CTLZ:s[0-9]+]], [[VAL]]
-; SI-DAG: v_cmp_eq_i32_e64 [[CMPZ:s\[[0-9]+:[0-9]+\]]], [[VAL]], 0{{$}}
-; SI-DAG: v_mov_b32_e32 [[VCTLZ:v[0-9]+]], [[CTLZ]]
-; SI: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], [[VCTLZ]], 32, [[CMPZ]]
-; SI: buffer_store_dword [[RESULT]]
-; SI: s_endpgm
+; GCN: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
+; GCN-DAG: s_flbit_i32_b32 [[CTLZ:s[0-9]+]], [[VAL]]
+; GCN-DAG: v_cmp_eq_i32_e64 [[CMPZ:s\[[0-9]+:[0-9]+\]]], [[VAL]], 0{{$}}
+; GCN-DAG: v_mov_b32_e32 [[VCTLZ:v[0-9]+]], [[CTLZ]]
+; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], [[VCTLZ]], 32, [[CMPZ]]
+; GCN: buffer_store_dword [[RESULT]]
+; GCN: s_endpgm
 
 ; EG: FFBH_UINT
 ; EG: CNDE_INT
@@ -34,12 +34,12 @@ define void @s_ctlz_i32(i32 addrspace(1)* noalias %out, i32 %val) nounwind {
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_i32:
-; SI: buffer_load_dword [[VAL:v[0-9]+]],
-; SI-DAG: v_ffbh_u32_e32 [[CTLZ:v[0-9]+]], [[VAL]]
-; SI-DAG: v_cmp_eq_i32_e32 vcc, 0, [[CTLZ]]
-; SI: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], [[CTLZ]], 32, vcc
-; SI: buffer_store_dword [[RESULT]],
-; SI: s_endpgm
+; GCN: buffer_load_dword [[VAL:v[0-9]+]],
+; GCN-DAG: v_ffbh_u32_e32 [[CTLZ:v[0-9]+]], [[VAL]]
+; GCN-DAG: v_cmp_eq_i32_e32 vcc, 0, [[CTLZ]]
+; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], [[CTLZ]], 32, vcc
+; GCN: buffer_store_dword [[RESULT]],
+; GCN: s_endpgm
 
 ; EG: FFBH_UINT
 ; EG: CNDE_INT
@@ -51,11 +51,11 @@ define void @v_ctlz_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalia
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_v2i32:
-; SI: buffer_load_dwordx2
-; SI: v_ffbh_u32_e32
-; SI: v_ffbh_u32_e32
-; SI: buffer_store_dwordx2
-; SI: s_endpgm
+; GCN: buffer_load_dwordx2
+; GCN: v_ffbh_u32_e32
+; GCN: v_ffbh_u32_e32
+; GCN: buffer_store_dwordx2
+; GCN: s_endpgm
 
 ; EG: FFBH_UINT
 ; EG: CNDE_INT
@@ -69,13 +69,13 @@ define void @v_ctlz_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrsp
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_v4i32:
-; SI: buffer_load_dwordx4
-; SI: v_ffbh_u32_e32
-; SI: v_ffbh_u32_e32
-; SI: v_ffbh_u32_e32
-; SI: v_ffbh_u32_e32
-; SI: buffer_store_dwordx4
-; SI: s_endpgm
+; GCN: buffer_load_dwordx4
+; GCN: v_ffbh_u32_e32
+; GCN: v_ffbh_u32_e32
+; GCN: v_ffbh_u32_e32
+; GCN: v_ffbh_u32_e32
+; GCN: buffer_store_dwordx4
+; GCN: s_endpgm
 
 
 ; EG-DAG: FFBH_UINT
@@ -97,12 +97,12 @@ define void @v_ctlz_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrsp
 }
 
 ; 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]],
+; GCN: buffer_load_ubyte [[VAL:v[0-9]+]],
+; GCN-DAG: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; GCN-DAG: v_cmp_eq_i32_e32 vcc, 0, [[CTLZ]]
+; GCN-DAG: v_cndmask_b32_e64 [[CORRECTED_FFBH:v[0-9]+]], [[FFBH]], 32, vcc
+; GCN: v_add_i32_e32 [[RESULT:v[0-9]+]], vcc, 0xffffffe8, [[CORRECTED_FFBH]]
+; GCN: 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
@@ -111,16 +111,16 @@ define void @v_ctlz_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %
 }
 
 ; 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, s[[HI]], 0{{$}}
-; SI-DAG: s_flbit_i32_b32 [[FFBH_LO:s[0-9]+]], s[[LO]]
-; SI-DAG: s_add_i32 [[ADD:s[0-9]+]], [[FFBH_LO]], 32
-; SI-DAG: s_flbit_i32_b32 [[FFBH_HI:s[0-9]+]], s[[HI]]
-; SI-DAG: v_mov_b32_e32 [[VFFBH_LO:v[0-9]+]], [[ADD]]
-; SI-DAG: v_mov_b32_e32 [[VFFBH_HI:v[0-9]+]], [[FFBH_HI]]
-; SI-DAG: v_cndmask_b32_e32 v[[CTLZ:[0-9]+]], [[VFFBH_HI]], [[VFFBH_LO]]
-; SI-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
-; SI: {{buffer|flat}}_store_dwordx2 {{.*}}v{{\[}}[[CTLZ]]:[[CTLZ_HI]]{{\]}}
+; GCN: s_load_dwordx2 s{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
+; GCN-DAG: v_cmp_eq_i32_e64 vcc, s[[HI]], 0{{$}}
+; GCN-DAG: s_flbit_i32_b32 [[FFBH_LO:s[0-9]+]], s[[LO]]
+; GCN-DAG: s_add_i32 [[ADD:s[0-9]+]], [[FFBH_LO]], 32
+; GCN-DAG: s_flbit_i32_b32 [[FFBH_HI:s[0-9]+]], s[[HI]]
+; GCN-DAG: v_mov_b32_e32 [[VFFBH_LO:v[0-9]+]], [[ADD]]
+; GCN-DAG: v_mov_b32_e32 [[VFFBH_HI:v[0-9]+]], [[FFBH_HI]]
+; GCN-DAG: v_cndmask_b32_e32 v[[CTLZ:[0-9]+]], [[VFFBH_HI]], [[VFFBH_LO]]
+; GCN-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
+; GCN: {{buffer|flat}}_store_dwordx2 {{.*}}v{{\[}}[[CTLZ]]:[[CTLZ_HI]]{{\]}}
 define void @s_ctlz_i64(i64 addrspace(1)* noalias %out, i64 %val) nounwind {
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
   store i64 %ctlz, i64 addrspace(1)* %out
@@ -136,17 +136,17 @@ define void @s_ctlz_i64_trunc(i32 addrspace(1)* noalias %out, i64 %val) nounwind
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_i64:
-; SI-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
-; SI-DAG: {{buffer|flat}}_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
-; SI-DAG: v_cmp_eq_i32_e64 [[CMPHI:s\[[0-9]+:[0-9]+\]]], 0, v[[HI]]
-; SI-DAG: v_ffbh_u32_e32 [[FFBH_LO:v[0-9]+]], v[[LO]]
-; SI-DAG: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, 32, [[FFBH_LO]]
-; SI-DAG: v_ffbh_u32_e32 [[FFBH_HI:v[0-9]+]], v[[HI]]
-; SI-DAG: v_cndmask_b32_e64 v[[CTLZ:[0-9]+]], [[FFBH_HI]], [[ADD]], [[CMPHI]]
-; SI-DAG: v_or_b32_e32 [[OR:v[0-9]+]], v[[HI]], v[[LO]]
-; SI-DAG: v_cmp_eq_i32_e32 vcc, 0, [[OR]]
-; SI-DAG: v_cndmask_b32_e64 v[[CLTZ_LO:[0-9]+]], v[[CTLZ:[0-9]+]], 64, vcc
-; SI: {{buffer|flat}}_store_dwordx2 {{.*}}v{{\[}}[[CLTZ_LO]]:[[CTLZ_HI]]{{\]}}
+; GCN-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
+; GCN-DAG: {{buffer|flat}}_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
+; GCN-DAG: v_cmp_eq_i32_e64 [[CMPHI:s\[[0-9]+:[0-9]+\]]], 0, v[[HI]]
+; GCN-DAG: v_ffbh_u32_e32 [[FFBH_LO:v[0-9]+]], v[[LO]]
+; GCN-DAG: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, 32, [[FFBH_LO]]
+; GCN-DAG: v_ffbh_u32_e32 [[FFBH_HI:v[0-9]+]], v[[HI]]
+; GCN-DAG: v_cndmask_b32_e64 v[[CTLZ:[0-9]+]], [[FFBH_HI]], [[ADD]], [[CMPHI]]
+; GCN-DAG: v_or_b32_e32 [[OR:v[0-9]+]], v[[HI]], v[[LO]]
+; GCN-DAG: v_cmp_eq_i32_e32 vcc, 0, [[OR]]
+; GCN-DAG: v_cndmask_b32_e64 v[[CLTZ_LO:[0-9]+]], v[[CTLZ:[0-9]+]], 64, vcc
+; GCN: {{buffer|flat}}_store_dwordx2 {{.*}}v{{\[}}[[CLTZ_LO]]:[[CTLZ_HI]]{{\]}}
 define void @v_ctlz_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
   %tid = call i32 @llvm.r600.read.tidig.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
@@ -170,10 +170,10 @@ define void @v_ctlz_i64_trunc(i32 addrspace(1)* noalias %out, i64 addrspace(1)*
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_i32_sel_eq_neg1:
-; SI: buffer_load_dword [[VAL:v[0-9]+]],
-; SI: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; SI: buffer_store_dword [[RESULT]],
-; SI: s_endpgm
+; GCN: buffer_load_dword [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GCN: buffer_store_dword [[RESULT]],
+; GCN: s_endpgm
  define void @v_ctlz_i32_sel_eq_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
@@ -184,10 +184,10 @@ define void @v_ctlz_i64_trunc(i32 addrspace(1)* noalias %out, i64 addrspace(1)*
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_i32_sel_ne_neg1:
-; SI: buffer_load_dword [[VAL:v[0-9]+]],
-; SI: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; SI: buffer_store_dword [[RESULT]],
-; SI: s_endpgm
+; GCN: buffer_load_dword [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GCN: buffer_store_dword [[RESULT]],
+; GCN: s_endpgm
 define void @v_ctlz_i32_sel_ne_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
@@ -199,11 +199,11 @@ define void @v_ctlz_i32_sel_ne_neg1(i32 addrspace(1)* noalias %out, i32 addrspac
 
 ; TODO: Should be able to eliminate select here as well.
 ; FUNC-LABEL: {{^}}v_ctlz_i32_sel_eq_bitwidth:
-; SI: buffer_load_dword
-; SI: v_ffbh_u32_e32
-; SI: v_cmp
-; SI: v_cndmask
-; SI: s_endpgm
+; GCN: buffer_load_dword
+; GCN: v_ffbh_u32_e32
+; GCN: v_cmp
+; GCN: v_cndmask
+; GCN: s_endpgm
 define void @v_ctlz_i32_sel_eq_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
@@ -214,11 +214,11 @@ define void @v_ctlz_i32_sel_eq_bitwidth(i32 addrspace(1)* noalias %out, i32 addr
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_i32_sel_ne_bitwidth:
-; SI: buffer_load_dword
-; SI: v_ffbh_u32_e32
-; SI: v_cmp
-; SI: v_cndmask
-; SI: s_endpgm
+; GCN: buffer_load_dword
+; GCN: v_ffbh_u32_e32
+; GCN: v_cmp
+; GCN: v_cndmask
+; GCN: s_endpgm
 define void @v_ctlz_i32_sel_ne_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
@@ -229,9 +229,9 @@ define void @v_ctlz_i32_sel_ne_bitwidth(i32 addrspace(1)* noalias %out, i32 addr
 }
 
 ; 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]],
+; GCN: buffer_load_ubyte [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; GCN: 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
@@ -255,10 +255,10 @@ define void @v_ctlz_i32_sel_ne_bitwidth(i32 addrspace(1)* noalias %out, i32 addr
 }
 
 ; 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]],
+; GCN: buffer_load_ubyte [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; GCN: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0x7f, [[FFBH]]
+; GCN: 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
diff --git a/llvm/test/CodeGen/AMDGPU/mul_uint24.ll b/llvm/test/CodeGen/AMDGPU/mul_uint24-amdgcn.ll
index b882a4dd634..05c4dfa5acf 100644
--- a/llvm/test/CodeGen/AMDGPU/mul_uint24.ll
+++ b/llvm/test/CodeGen/AMDGPU/mul_uint24-amdgcn.ll
@@ -1,11 +1,11 @@
-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
-; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
-; RUN: llc -march=r600 -mcpu=cayman < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=FUNC %s
+
+declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
+declare i32 @llvm.amdgcn.workitem.id.y() nounwind readnone
 
 ; FUNC-LABEL: {{^}}test_umul24_i32:
-; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]\.[XYZW]}}, KC0[2].Z, KC0[2].W
-; SI: v_mul_u32_u24
+; GCN: v_mul_u32_u24
 define void @test_umul24_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) {
 entry:
   %0 = shl i32 %a, 8
@@ -18,13 +18,10 @@ entry:
 }
 
 ; FUNC-LABEL: {{^}}test_umul24_i16_sext:
-; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]}}.[[MUL_CHAN:[XYZW]]]
-; The result must be sign-extended
-; EG: BFE_INT {{[* ]*}}T{{[0-9]}}.{{[XYZW]}}, PV.[[MUL_CHAN]], 0.0, literal.x
-; EG: 16
-
-; SI: v_mul_u32_u24_e{{(32|64)}} [[MUL:v[0-9]]], {{[sv][0-9], [sv][0-9]}}
-; SI: v_bfe_i32 v{{[0-9]}}, [[MUL]], 0, 16
+; SI: v_mul_u32_u24_e{{(32|64)}} [[VI_MUL:v[0-9]]], {{[sv][0-9], [sv][0-9]}}
+; SI: v_bfe_i32 v{{[0-9]}}, [[VI_MUL]], 0, 16
+; VI: s_mul_i32 [[SI_MUL:s[0-9]]], s{{[0-9]}}, s{{[0-9]}}
+; VI: s_sext_i32_i16 s{{[0-9]}}, [[SI_MUL]]
 define void @test_umul24_i16_sext(i32 addrspace(1)* %out, i16 %a, i16 %b) {
 entry:
   %mul = mul i16 %a, %b
@@ -33,10 +30,29 @@ entry:
   ret void
 }
 
+; FUNC-LABEL: {{^}}test_umul24_i16_vgpr_sext:
+; GCN: v_mul_u32_u24_e{{(32|64)}} [[MUL:v[0-9]]], {{[sv][0-9], [sv][0-9]}}
+; GCN: v_bfe_i32 v{{[0-9]}}, [[MUL]], 0, 16
+define void @test_umul24_i16_vgpr_sext(i32 addrspace(1)* %out, i16 addrspace(1)* %in) {
+  %tid.x = call i32 @llvm.amdgcn.workitem.id.x()
+  %tid.y = call i32 @llvm.amdgcn.workitem.id.y()
+  %ptr_a = getelementptr i16, i16 addrspace(1)* %in, i32 %tid.x
+  %ptr_b = getelementptr i16, i16 addrspace(1)* %in, i32 %tid.y
+  %a = load i16, i16 addrspace(1)* %ptr_a
+  %b = load i16, i16 addrspace(1)* %ptr_b
+  %mul = mul i16 %a, %b
+  %val = sext i16 %mul to i32
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}test_umul24_i16:
 ; SI: s_and_b32
 ; SI: v_mul_u32_u24_e32
 ; SI: v_and_b32_e32
+; VI: s_mul_i32
+; VI: s_and_b32
+; VI: v_mov_b32_e32
 define void @test_umul24_i16(i32 addrspace(1)* %out, i16 %a, i16 %b) {
 entry:
   %mul = mul i16 %a, %b
@@ -45,13 +61,25 @@ entry:
   ret void
 }
 
-; FUNC-LABEL: {{^}}test_umul24_i8:
-; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]}}.[[MUL_CHAN:[XYZW]]]
-; The result must be sign-extended
-; EG: BFE_INT {{[* ]*}}T{{[0-9]}}.{{[XYZW]}}, PV.[[MUL_CHAN]], 0.0, literal.x
-; SI: v_mul_u32_u24_e{{(32|64)}} [[MUL:v[0-9]]], {{[sv][0-9], [sv][0-9]}}
-; SI: v_bfe_i32 v{{[0-9]}}, [[MUL]], 0, 8
+; FUNC-LABEL: {{^}}test_umul24_i16_vgpr:
+; GCN: v_mul_u32_u24_e32
+; GCN: v_and_b32_e32
+define void @test_umul24_i16_vgpr(i32 addrspace(1)* %out, i16 addrspace(1)* %in) {
+  %tid.x = call i32 @llvm.amdgcn.workitem.id.x()
+  %tid.y = call i32 @llvm.amdgcn.workitem.id.y()
+  %ptr_a = getelementptr i16, i16 addrspace(1)* %in, i32 %tid.x
+  %ptr_b = getelementptr i16, i16 addrspace(1)* %in, i32 %tid.y
+  %a = load i16, i16 addrspace(1)* %ptr_a
+  %b = load i16, i16 addrspace(1)* %ptr_b
+  %mul = mul i16 %a, %b
+  %val = zext i16 %mul to i32
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
 
+; FUNC-LABEL: {{^}}test_umul24_i8:
+; GCN: v_mul_u32_u24_e{{(32|64)}} [[MUL:v[0-9]]], {{[sv][0-9], [sv][0-9]}}
+; GCN: v_bfe_i32 v{{[0-9]}}, [[MUL]], 0, 8
 define void @test_umul24_i8(i32 addrspace(1)* %out, i8 %a, i8 %b) {
 entry:
   %mul = mul i8 %a, %b
@@ -61,11 +89,9 @@ entry:
 }
 
 ; FUNC-LABEL: {{^}}test_umulhi24_i32_i64:
-; SI-NOT: and
-; SI: v_mul_hi_u32_u24_e32 [[RESULT:v[0-9]+]],
-; SI-NEXT: buffer_store_dword [[RESULT]]
-
-; EG: MULHI_UINT24 {{[* ]*}}T{{[0-9]\.[XYZW]}}, KC0[2].Z, KC0[2].W
+; GCN-NOT: and
+; GCN: v_mul_hi_u32_u24_e32 [[RESULT:v[0-9]+]],
+; GCN-NEXT: buffer_store_dword [[RESULT]]
 define void @test_umulhi24_i32_i64(i32 addrspace(1)* %out, i32 %a, i32 %b) {
 entry:
   %a.24 = and i32 %a, 16777215
@@ -80,11 +106,9 @@ entry:
 }
 
 ; FUNC-LABEL: {{^}}test_umulhi24:
-; SI-NOT: and
-; SI: v_mul_hi_u32_u24_e32 [[RESULT:v[0-9]+]],
-; SI-NEXT: buffer_store_dword [[RESULT]]
-
-; EG: MULHI_UINT24 {{[* ]*}}T{{[0-9]\.[XYZW]}}, KC0[2].W, KC0[3].Y
+; GCN-NOT: and
+; GCN: v_mul_hi_u32_u24_e32 [[RESULT:v[0-9]+]],
+; GCN-NEXT: buffer_store_dword [[RESULT]]
 define void @test_umulhi24(i32 addrspace(1)* %out, i64 %a, i64 %b) {
 entry:
   %a.24 = and i64 %a, 16777215
@@ -96,18 +120,13 @@ entry:
   ret void
 }
 
-; Multiply with 24-bit inputs and 64-bit output
+; Multiply with 24-bit inputs and 64-bit output.
 ; FUNC-LABEL: {{^}}test_umul24_i64:
-; EG; MUL_UINT24
-; EG: MULHI
-
-; SI-NOT: and
-; SI-NOT: lshr
-
-; SI-DAG: v_mul_u32_u24_e32
-; SI-DAG: v_mul_hi_u32_u24_e32
-
-; SI: buffer_store_dwordx2
+; GCN-NOT: and
+; GCN-NOT: lshr
+; GCN-DAG: v_mul_u32_u24_e32
+; GCN-DAG: v_mul_hi_u32_u24_e32
+; GCN: buffer_store_dwordx2
 define void @test_umul24_i64(i64 addrspace(1)* %out, i64 %a, i64 %b) {
 entry:
   %tmp0 = shl i64 %a, 40
@@ -119,12 +138,12 @@ entry:
   ret void
 }
 
-; FIXME: Should be able to eliminate the and
+; FIXME: Should be able to eliminate the and.
 ; FUNC-LABEL: {{^}}test_umul24_i64_square:
-; SI: s_load_dword [[A:s[0-9]+]]
-; SI: s_and_b32 [[TRUNC:s[0-9]+]], [[A]], 0xffffff{{$}}
-; SI-DAG: v_mul_hi_u32_u24_e64 v{{[0-9]+}}, [[TRUNC]], [[TRUNC]]
-; SI-DAG: v_mul_u32_u24_e64 v{{[0-9]+}}, [[TRUNC]], [[TRUNC]]
+; GCN: s_load_dword [[A:s[0-9]+]]
+; GCN: s_and_b32 [[TRUNC:s[0-9]+]], [[A]], 0xffffff{{$}}
+; GCN-DAG: v_mul_hi_u32_u24_e64 v{{[0-9]+}}, [[TRUNC]], [[TRUNC]]
+; GCN-DAG: v_mul_u32_u24_e64 v{{[0-9]+}}, [[TRUNC]], [[TRUNC]]
 define void @test_umul24_i64_square(i64 addrspace(1)* %out, i64 %a) {
 entry:
   %tmp0 = shl i64 %a, 40
@@ -135,10 +154,10 @@ entry:
 }
 
 ; FUNC-LABEL: {{^}}test_umulhi16_i32:
-; SI: s_and_b32
-; SI: s_and_b32
-; SI: v_mul_u32_u24_e32 [[MUL24:v[0-9]+]]
-; SI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, [[MUL24]]
+; GCN: s_and_b32
+; GCN: s_and_b32
+; GCN: v_mul_u32_u24_e32 [[MUL24:v[0-9]+]]
+; GCN: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, [[MUL24]]
 define void @test_umulhi16_i32(i16 addrspace(1)* %out, i32 %a, i32 %b) {
 entry:
   %a.16 = and i32 %a, 65535
@@ -151,16 +170,14 @@ entry:
 }
 
 ; FUNC-LABEL: {{^}}test_umul24_i33:
-; SI: s_load_dword s
-; SI: s_load_dword s
-
-; SI-NOT: and
-; SI-NOT: lshr
-
-; SI-DAG: v_mul_u32_u24_e32 v[[MUL_LO:[0-9]+]],
-; SI-DAG: v_mul_hi_u32_u24_e32 v[[MUL_HI:[0-9]+]],
-; SI-DAG: v_and_b32_e32 v[[HI:[0-9]+]], 1, v[[MUL_HI]]
-; SI: buffer_store_dwordx2 v{{\[}}[[MUL_LO]]:[[HI]]{{\]}}
+; GCN: s_load_dword s
+; GCN: s_load_dword s
+; GCN-NOT: and
+; GCN-NOT: lshr
+; GCN-DAG: v_mul_u32_u24_e32 v[[MUL_LO:[0-9]+]],
+; GCN-DAG: v_mul_hi_u32_u24_e32 v[[MUL_HI:[0-9]+]],
+; GCN-DAG: v_and_b32_e32 v[[HI:[0-9]+]], 1, v[[MUL_HI]]
+; GCN: buffer_store_dwordx2 v{{\[}}[[MUL_LO]]:[[HI]]{{\]}}
 define void @test_umul24_i33(i64 addrspace(1)* %out, i33 %a, i33 %b) {
 entry:
   %tmp0 = shl i33 %a, 9
@@ -174,15 +191,13 @@ entry:
 }
 
 ; FUNC-LABEL: {{^}}test_umulhi24_i33:
-; SI: s_load_dword s
-; SI: s_load_dword s
-
-; SI-NOT: and
-; SI-NOT: lshr
-
-; SI: v_mul_hi_u32_u24_e32 v[[MUL_HI:[0-9]+]],
-; SI-NEXT: v_and_b32_e32 v[[HI:[0-9]+]], 1, v[[MUL_HI]]
-; SI-NEXT: buffer_store_dword v[[HI]]
+; GCN: s_load_dword s
+; GCN: s_load_dword s
+; GCN-NOT: and
+; GCN-NOT: lshr
+; GCN: v_mul_hi_u32_u24_e32 v[[MUL_HI:[0-9]+]],
+; GCN-NEXT: v_and_b32_e32 v[[HI:[0-9]+]], 1, v[[MUL_HI]]
+; GCN-NEXT: buffer_store_dword v[[HI]]
 define void @test_umulhi24_i33(i32 addrspace(1)* %out, i33 %a, i33 %b) {
 entry:
   %tmp0 = shl i33 %a, 9
diff --git a/llvm/test/CodeGen/AMDGPU/mul_uint24-r600.ll b/llvm/test/CodeGen/AMDGPU/mul_uint24-r600.ll
new file mode 100644
index 00000000000..da1c111fa5c
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/mul_uint24-r600.ll
@@ -0,0 +1,83 @@
+; RUN: llc -march=r600 -mcpu=cayman < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
+
+; FUNC-LABEL: {{^}}test_umul24_i32:
+; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]\.[XYZW]}}, KC0[2].Z, KC0[2].W
+define void @test_umul24_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) {
+entry:
+  %0 = shl i32 %a, 8
+  %a_24 = lshr i32 %0, 8
+  %1 = shl i32 %b, 8
+  %b_24 = lshr i32 %1, 8
+  %2 = mul i32 %a_24, %b_24
+  store i32 %2, i32 addrspace(1)* %out
+  ret void
+}
+
+; The result must be sign-extended.
+; FUNC-LABEL: {{^}}test_umul24_i16_sext:
+; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]}}.[[MUL_CHAN:[XYZW]]]
+; EG: BFE_INT {{[* ]*}}T{{[0-9]}}.{{[XYZW]}}, PV.[[MUL_CHAN]], 0.0, literal.x
+; EG: 16
+define void @test_umul24_i16_sext(i32 addrspace(1)* %out, i16 %a, i16 %b) {
+entry:
+  %mul = mul i16 %a, %b
+  %ext = sext i16 %mul to i32
+  store i32 %ext, i32 addrspace(1)* %out
+  ret void
+}
+
+; The result must be sign-extended.
+; FUNC-LABEL: {{^}}test_umul24_i8:
+; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]}}.[[MUL_CHAN:[XYZW]]]
+; EG: BFE_INT {{[* ]*}}T{{[0-9]}}.{{[XYZW]}}, PV.[[MUL_CHAN]], 0.0, literal.x
+define void @test_umul24_i8(i32 addrspace(1)* %out, i8 %a, i8 %b) {
+entry:
+  %mul = mul i8 %a, %b
+  %ext = sext i8 %mul to i32
+  store i32 %ext, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}test_umulhi24_i32_i64:
+; EG: MULHI_UINT24 {{[* ]*}}T{{[0-9]\.[XYZW]}}, KC0[2].Z, KC0[2].W
+define void @test_umulhi24_i32_i64(i32 addrspace(1)* %out, i32 %a, i32 %b) {
+entry:
+  %a.24 = and i32 %a, 16777215
+  %b.24 = and i32 %b, 16777215
+  %a.24.i64 = zext i32 %a.24 to i64
+  %b.24.i64 = zext i32 %b.24 to i64
+  %mul48 = mul i64 %a.24.i64, %b.24.i64
+  %mul48.hi = lshr i64 %mul48, 32
+  %mul24hi = trunc i64 %mul48.hi to i32
+  store i32 %mul24hi, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}test_umulhi24:
+; EG: MULHI_UINT24 {{[* ]*}}T{{[0-9]\.[XYZW]}}, KC0[2].W, KC0[3].Y
+define void @test_umulhi24(i32 addrspace(1)* %out, i64 %a, i64 %b) {
+entry:
+  %a.24 = and i64 %a, 16777215
+  %b.24 = and i64 %b, 16777215
+  %mul48 = mul i64 %a.24, %b.24
+  %mul48.hi = lshr i64 %mul48, 32
+  %mul24.hi = trunc i64 %mul48.hi to i32
+  store i32 %mul24.hi, i32 addrspace(1)* %out
+  ret void
+}
+
+; Multiply with 24-bit inputs and 64-bit output.
+; FUNC-LABEL: {{^}}test_umul24_i64:
+; EG; MUL_UINT24
+; EG: MULHI
+define void @test_umul24_i64(i64 addrspace(1)* %out, i64 %a, i64 %b) {
+entry:
+  %tmp0 = shl i64 %a, 40
+  %a_24 = lshr i64 %tmp0, 40
+  %tmp1 = shl i64 %b, 40
+  %b_24 = lshr i64 %tmp1, 40
+  %tmp2 = mul i64 %a_24, %b_24
+  store i64 %tmp2, i64 addrspace(1)* %out
+  ret void
+}