[ARM] Lower sadd_sat to qadd8 and qadd16

Lower the target independent signed saturating intrinsics to qadd8 and qadd16.
This custom lowers them from a sadd_sat, catching the node early before it is
promoted. It also adds a QADD8b and QADD16b node to mean the bottom "lane" of a
qadd8/qadd16, so that we can call demand bits on it to show that it does not
use the upper bits.

Also handles QSUB8 and QSUB16.

Differential Revision: https://reviews.llvm.org/D68974


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@375402 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 243 insertions, 164 deletions

diff --git a/test/CodeGen/ARM/sadd_sat.ll b/test/CodeGen/ARM/sadd_sat.ll
index 539f3af76d1..1fcc460670e 100644
--- a/test/CodeGen/ARM/sadd_sat.ll
+++ b/test/CodeGen/ARM/sadd_sat.ll
@@ -233,35 +233,63 @@ define signext i16 @func16(i16 signext %x, i16 signext %y) nounwind {
 ; CHECK-T1-NEXT:  .LCPI2_1:
 ; CHECK-T1-NEXT:    .long 4294934528 @ 0xffff8000
 ;
-; CHECK-T2-LABEL: func16:
-; CHECK-T2:       @ %bb.0:
-; CHECK-T2-NEXT:    add r0, r1
-; CHECK-T2-NEXT:    movw r1, #32767
-; CHECK-T2-NEXT:    cmp r0, r1
-; CHECK-T2-NEXT:    it lt
-; CHECK-T2-NEXT:    movlt r1, r0
-; CHECK-T2-NEXT:    movw r0, #32768
-; CHECK-T2-NEXT:    cmn.w r1, #32768
-; CHECK-T2-NEXT:    movt r0, #65535
-; CHECK-T2-NEXT:    it gt
-; CHECK-T2-NEXT:    movgt r0, r1
-; CHECK-T2-NEXT:    bx lr
+; CHECK-T2NODSP-LABEL: func16:
+; CHECK-T2NODSP:       @ %bb.0:
+; CHECK-T2NODSP-NEXT:    add r0, r1
+; CHECK-T2NODSP-NEXT:    movw r1, #32767
+; CHECK-T2NODSP-NEXT:    cmp r0, r1
+; CHECK-T2NODSP-NEXT:    it lt
+; CHECK-T2NODSP-NEXT:    movlt r1, r0
+; CHECK-T2NODSP-NEXT:    movw r0, #32768
+; CHECK-T2NODSP-NEXT:    cmn.w r1, #32768
+; CHECK-T2NODSP-NEXT:    movt r0, #65535
+; CHECK-T2NODSP-NEXT:    it gt
+; CHECK-T2NODSP-NEXT:    movgt r0, r1
+; CHECK-T2NODSP-NEXT:    bx lr
 ;
-; CHECK-ARM-LABEL: func16:
-; CHECK-ARM:       @ %bb.0:
-; CHECK-ARM-NEXT:    add r0, r0, r1
-; CHECK-ARM-NEXT:    mov r1, #255
-; CHECK-ARM-NEXT:    orr r1, r1, #32512
-; CHECK-ARM-NEXT:    cmp r0, r1
-; CHECK-ARM-NEXT:    movlt r1, r0
-; CHECK-ARM-NEXT:    ldr r0, .LCPI2_0
-; CHECK-ARM-NEXT:    cmn r1, #32768
-; CHECK-ARM-NEXT:    movgt r0, r1
-; CHECK-ARM-NEXT:    bx lr
-; CHECK-ARM-NEXT:    .p2align 2
-; CHECK-ARM-NEXT:  @ %bb.1:
-; CHECK-ARM-NEXT:  .LCPI2_0:
-; CHECK-ARM-NEXT:    .long 4294934528 @ 0xffff8000
+; CHECK-T2DSP-LABEL: func16:
+; CHECK-T2DSP:       @ %bb.0:
+; CHECK-T2DSP-NEXT:    qadd16 r0, r0, r1
+; CHECK-T2DSP-NEXT:    sxth r0, r0
+; CHECK-T2DSP-NEXT:    bx lr
+;
+; CHECK-ARMNODPS-LABEL: func16:
+; CHECK-ARMNODPS:       @ %bb.0:
+; CHECK-ARMNODPS-NEXT:    add r0, r0, r1
+; CHECK-ARMNODPS-NEXT:    mov r1, #255
+; CHECK-ARMNODPS-NEXT:    orr r1, r1, #32512
+; CHECK-ARMNODPS-NEXT:    cmp r0, r1
+; CHECK-ARMNODPS-NEXT:    movlt r1, r0
+; CHECK-ARMNODPS-NEXT:    ldr r0, .LCPI2_0
+; CHECK-ARMNODPS-NEXT:    cmn r1, #32768
+; CHECK-ARMNODPS-NEXT:    movgt r0, r1
+; CHECK-ARMNODPS-NEXT:    bx lr
+; CHECK-ARMNODPS-NEXT:    .p2align 2
+; CHECK-ARMNODPS-NEXT:  @ %bb.1:
+; CHECK-ARMNODPS-NEXT:  .LCPI2_0:
+; CHECK-ARMNODPS-NEXT:    .long 4294934528 @ 0xffff8000
+;
+; CHECK-ARMBASEDSP-LABEL: func16:
+; CHECK-ARMBASEDSP:       @ %bb.0:
+; CHECK-ARMBASEDSP-NEXT:    add r0, r0, r1
+; CHECK-ARMBASEDSP-NEXT:    mov r1, #255
+; CHECK-ARMBASEDSP-NEXT:    orr r1, r1, #32512
+; CHECK-ARMBASEDSP-NEXT:    cmp r0, r1
+; CHECK-ARMBASEDSP-NEXT:    movlt r1, r0
+; CHECK-ARMBASEDSP-NEXT:    ldr r0, .LCPI2_0
+; CHECK-ARMBASEDSP-NEXT:    cmn r1, #32768
+; CHECK-ARMBASEDSP-NEXT:    movgt r0, r1
+; CHECK-ARMBASEDSP-NEXT:    bx lr
+; CHECK-ARMBASEDSP-NEXT:    .p2align 2
+; CHECK-ARMBASEDSP-NEXT:  @ %bb.1:
+; CHECK-ARMBASEDSP-NEXT:  .LCPI2_0:
+; CHECK-ARMBASEDSP-NEXT:    .long 4294934528 @ 0xffff8000
+;
+; CHECK-ARMDSP-LABEL: func16:
+; CHECK-ARMDSP:       @ %bb.0:
+; CHECK-ARMDSP-NEXT:    qadd16 r0, r0, r1
+; CHECK-ARMDSP-NEXT:    sxth r0, r0
+; CHECK-ARMDSP-NEXT:    bx lr
   %tmp = call i16 @llvm.sadd.sat.i16(i16 %x, i16 %y)
   ret i16 %tmp
 }
@@ -284,25 +312,46 @@ define signext i8 @func8(i8 signext %x, i8 signext %y) nounwind {
 ; CHECK-T1-NEXT:  .LBB3_4:
 ; CHECK-T1-NEXT:    bx lr
 ;
-; CHECK-T2-LABEL: func8:
-; CHECK-T2:       @ %bb.0:
-; CHECK-T2-NEXT:    add r0, r1
-; CHECK-T2-NEXT:    cmp r0, #127
-; CHECK-T2-NEXT:    it ge
-; CHECK-T2-NEXT:    movge r0, #127
-; CHECK-T2-NEXT:    cmn.w r0, #128
-; CHECK-T2-NEXT:    it le
-; CHECK-T2-NEXT:    mvnle r0, #127
-; CHECK-T2-NEXT:    bx lr
+; CHECK-T2NODSP-LABEL: func8:
+; CHECK-T2NODSP:       @ %bb.0:
+; CHECK-T2NODSP-NEXT:    add r0, r1
+; CHECK-T2NODSP-NEXT:    cmp r0, #127
+; CHECK-T2NODSP-NEXT:    it ge
+; CHECK-T2NODSP-NEXT:    movge r0, #127
+; CHECK-T2NODSP-NEXT:    cmn.w r0, #128
+; CHECK-T2NODSP-NEXT:    it le
+; CHECK-T2NODSP-NEXT:    mvnle r0, #127
+; CHECK-T2NODSP-NEXT:    bx lr
 ;
-; CHECK-ARM-LABEL: func8:
-; CHECK-ARM:       @ %bb.0:
-; CHECK-ARM-NEXT:    add r0, r0, r1
-; CHECK-ARM-NEXT:    cmp r0, #127
-; CHECK-ARM-NEXT:    movge r0, #127
-; CHECK-ARM-NEXT:    cmn r0, #128
-; CHECK-ARM-NEXT:    mvnle r0, #127
-; CHECK-ARM-NEXT:    bx lr
+; CHECK-T2DSP-LABEL: func8:
+; CHECK-T2DSP:       @ %bb.0:
+; CHECK-T2DSP-NEXT:    qadd8 r0, r0, r1
+; CHECK-T2DSP-NEXT:    sxtb r0, r0
+; CHECK-T2DSP-NEXT:    bx lr
+;
+; CHECK-ARMNODPS-LABEL: func8:
+; CHECK-ARMNODPS:       @ %bb.0:
+; CHECK-ARMNODPS-NEXT:    add r0, r0, r1
+; CHECK-ARMNODPS-NEXT:    cmp r0, #127
+; CHECK-ARMNODPS-NEXT:    movge r0, #127
+; CHECK-ARMNODPS-NEXT:    cmn r0, #128
+; CHECK-ARMNODPS-NEXT:    mvnle r0, #127
+; CHECK-ARMNODPS-NEXT:    bx lr
+;
+; CHECK-ARMBASEDSP-LABEL: func8:
+; CHECK-ARMBASEDSP:       @ %bb.0:
+; CHECK-ARMBASEDSP-NEXT:    add r0, r0, r1
+; CHECK-ARMBASEDSP-NEXT:    cmp r0, #127
+; CHECK-ARMBASEDSP-NEXT:    movge r0, #127
+; CHECK-ARMBASEDSP-NEXT:    cmn r0, #128
+; CHECK-ARMBASEDSP-NEXT:    mvnle r0, #127
+; CHECK-ARMBASEDSP-NEXT:    bx lr
+;
+; CHECK-ARMDSP-LABEL: func8:
+; CHECK-ARMDSP:       @ %bb.0:
+; CHECK-ARMDSP-NEXT:    qadd8 r0, r0, r1
+; CHECK-ARMDSP-NEXT:    sxtb r0, r0
+; CHECK-ARMDSP-NEXT:    bx lr
   %tmp = call i8 @llvm.sadd.sat.i8(i8 %x, i8 %y)
   ret i8 %tmp
 }
diff --git a/test/CodeGen/ARM/sadd_sat_plus.ll b/test/CodeGen/ARM/sadd_sat_plus.ll
index 94aca12a78b..370d5d989a8 100644
--- a/test/CodeGen/ARM/sadd_sat_plus.ll
+++ b/test/CodeGen/ARM/sadd_sat_plus.ll
@@ -258,29 +258,15 @@ define signext i16 @func16(i16 signext %x, i16 signext %y, i16 signext %z) nounw
 ; CHECK-T2DSP-LABEL: func16:
 ; CHECK-T2DSP:       @ %bb.0:
 ; CHECK-T2DSP-NEXT:    muls r1, r2, r1
-; CHECK-T2DSP-NEXT:    sxtah r0, r0, r1
-; CHECK-T2DSP-NEXT:    movw r1, #32767
-; CHECK-T2DSP-NEXT:    cmp r0, r1
-; CHECK-T2DSP-NEXT:    it lt
-; CHECK-T2DSP-NEXT:    movlt r1, r0
-; CHECK-T2DSP-NEXT:    movw r0, #32768
-; CHECK-T2DSP-NEXT:    cmn.w r1, #32768
-; CHECK-T2DSP-NEXT:    movt r0, #65535
-; CHECK-T2DSP-NEXT:    it gt
-; CHECK-T2DSP-NEXT:    movgt r0, r1
+; CHECK-T2DSP-NEXT:    qadd16 r0, r0, r1
+; CHECK-T2DSP-NEXT:    sxth r0, r0
 ; CHECK-T2DSP-NEXT:    bx lr
 ;
 ; CHECK-ARM-LABEL: func16:
 ; CHECK-ARM:       @ %bb.0:
 ; CHECK-ARM-NEXT:    smulbb r1, r1, r2
-; CHECK-ARM-NEXT:    sxtah r0, r0, r1
-; CHECK-ARM-NEXT:    movw r1, #32767
-; CHECK-ARM-NEXT:    cmp r0, r1
-; CHECK-ARM-NEXT:    movlt r1, r0
-; CHECK-ARM-NEXT:    movw r0, #32768
-; CHECK-ARM-NEXT:    movt r0, #65535
-; CHECK-ARM-NEXT:    cmn r1, #32768
-; CHECK-ARM-NEXT:    movgt r0, r1
+; CHECK-ARM-NEXT:    qadd16 r0, r0, r1
+; CHECK-ARM-NEXT:    sxth r0, r0
 ; CHECK-ARM-NEXT:    bx lr
   %a = mul i16 %y, %z
   %tmp = call i16 @llvm.sadd.sat.i16(i16 %x, i16 %a)
@@ -323,23 +309,15 @@ define signext i8 @func8(i8 signext %x, i8 signext %y, i8 signext %z) nounwind {
 ; CHECK-T2DSP-LABEL: func8:
 ; CHECK-T2DSP:       @ %bb.0:
 ; CHECK-T2DSP-NEXT:    muls r1, r2, r1
-; CHECK-T2DSP-NEXT:    sxtab r0, r0, r1
-; CHECK-T2DSP-NEXT:    cmp r0, #127
-; CHECK-T2DSP-NEXT:    it ge
-; CHECK-T2DSP-NEXT:    movge r0, #127
-; CHECK-T2DSP-NEXT:    cmn.w r0, #128
-; CHECK-T2DSP-NEXT:    it le
-; CHECK-T2DSP-NEXT:    mvnle r0, #127
+; CHECK-T2DSP-NEXT:    qadd8 r0, r0, r1
+; CHECK-T2DSP-NEXT:    sxtb r0, r0
 ; CHECK-T2DSP-NEXT:    bx lr
 ;
 ; CHECK-ARM-LABEL: func8:
 ; CHECK-ARM:       @ %bb.0:
 ; CHECK-ARM-NEXT:    smulbb r1, r1, r2
-; CHECK-ARM-NEXT:    sxtab r0, r0, r1
-; CHECK-ARM-NEXT:    cmp r0, #127
-; CHECK-ARM-NEXT:    movge r0, #127
-; CHECK-ARM-NEXT:    cmn r0, #128
-; CHECK-ARM-NEXT:    mvnle r0, #127
+; CHECK-ARM-NEXT:    qadd8 r0, r0, r1
+; CHECK-ARM-NEXT:    sxtb r0, r0
 ; CHECK-ARM-NEXT:    bx lr
   %a = mul i8 %y, %z
   %tmp = call i8 @llvm.sadd.sat.i8(i8 %x, i8 %a)
diff --git a/test/CodeGen/ARM/ssub_sat.ll b/test/CodeGen/ARM/ssub_sat.ll
index 6efb34df7b0..0c9e125316c 100644
--- a/test/CodeGen/ARM/ssub_sat.ll
+++ b/test/CodeGen/ARM/ssub_sat.ll
@@ -235,35 +235,63 @@ define signext i16 @func16(i16 signext %x, i16 signext %y) nounwind {
 ; CHECK-T1-NEXT:  .LCPI2_1:
 ; CHECK-T1-NEXT:    .long 4294934528 @ 0xffff8000
 ;
-; CHECK-T2-LABEL: func16:
-; CHECK-T2:       @ %bb.0:
-; CHECK-T2-NEXT:    subs r0, r0, r1
-; CHECK-T2-NEXT:    movw r1, #32767
-; CHECK-T2-NEXT:    cmp r0, r1
-; CHECK-T2-NEXT:    it lt
-; CHECK-T2-NEXT:    movlt r1, r0
-; CHECK-T2-NEXT:    movw r0, #32768
-; CHECK-T2-NEXT:    cmn.w r1, #32768
-; CHECK-T2-NEXT:    movt r0, #65535
-; CHECK-T2-NEXT:    it gt
-; CHECK-T2-NEXT:    movgt r0, r1
-; CHECK-T2-NEXT:    bx lr
+; CHECK-T2NODSP-LABEL: func16:
+; CHECK-T2NODSP:       @ %bb.0:
+; CHECK-T2NODSP-NEXT:    subs r0, r0, r1
+; CHECK-T2NODSP-NEXT:    movw r1, #32767
+; CHECK-T2NODSP-NEXT:    cmp r0, r1
+; CHECK-T2NODSP-NEXT:    it lt
+; CHECK-T2NODSP-NEXT:    movlt r1, r0
+; CHECK-T2NODSP-NEXT:    movw r0, #32768
+; CHECK-T2NODSP-NEXT:    cmn.w r1, #32768
+; CHECK-T2NODSP-NEXT:    movt r0, #65535
+; CHECK-T2NODSP-NEXT:    it gt
+; CHECK-T2NODSP-NEXT:    movgt r0, r1
+; CHECK-T2NODSP-NEXT:    bx lr
 ;
-; CHECK-ARM-LABEL: func16:
-; CHECK-ARM:       @ %bb.0:
-; CHECK-ARM-NEXT:    sub r0, r0, r1
-; CHECK-ARM-NEXT:    mov r1, #255
-; CHECK-ARM-NEXT:    orr r1, r1, #32512
-; CHECK-ARM-NEXT:    cmp r0, r1
-; CHECK-ARM-NEXT:    movlt r1, r0
-; CHECK-ARM-NEXT:    ldr r0, .LCPI2_0
-; CHECK-ARM-NEXT:    cmn r1, #32768
-; CHECK-ARM-NEXT:    movgt r0, r1
-; CHECK-ARM-NEXT:    bx lr
-; CHECK-ARM-NEXT:    .p2align 2
-; CHECK-ARM-NEXT:  @ %bb.1:
-; CHECK-ARM-NEXT:  .LCPI2_0:
-; CHECK-ARM-NEXT:    .long 4294934528 @ 0xffff8000
+; CHECK-T2DSP-LABEL: func16:
+; CHECK-T2DSP:       @ %bb.0:
+; CHECK-T2DSP-NEXT:    qsub16 r0, r0, r1
+; CHECK-T2DSP-NEXT:    sxth r0, r0
+; CHECK-T2DSP-NEXT:    bx lr
+;
+; CHECK-ARMNODPS-LABEL: func16:
+; CHECK-ARMNODPS:       @ %bb.0:
+; CHECK-ARMNODPS-NEXT:    sub r0, r0, r1
+; CHECK-ARMNODPS-NEXT:    mov r1, #255
+; CHECK-ARMNODPS-NEXT:    orr r1, r1, #32512
+; CHECK-ARMNODPS-NEXT:    cmp r0, r1
+; CHECK-ARMNODPS-NEXT:    movlt r1, r0
+; CHECK-ARMNODPS-NEXT:    ldr r0, .LCPI2_0
+; CHECK-ARMNODPS-NEXT:    cmn r1, #32768
+; CHECK-ARMNODPS-NEXT:    movgt r0, r1
+; CHECK-ARMNODPS-NEXT:    bx lr
+; CHECK-ARMNODPS-NEXT:    .p2align 2
+; CHECK-ARMNODPS-NEXT:  @ %bb.1:
+; CHECK-ARMNODPS-NEXT:  .LCPI2_0:
+; CHECK-ARMNODPS-NEXT:    .long 4294934528 @ 0xffff8000
+;
+; CHECK-ARMBASEDSP-LABEL: func16:
+; CHECK-ARMBASEDSP:       @ %bb.0:
+; CHECK-ARMBASEDSP-NEXT:    sub r0, r0, r1
+; CHECK-ARMBASEDSP-NEXT:    mov r1, #255
+; CHECK-ARMBASEDSP-NEXT:    orr r1, r1, #32512
+; CHECK-ARMBASEDSP-NEXT:    cmp r0, r1
+; CHECK-ARMBASEDSP-NEXT:    movlt r1, r0
+; CHECK-ARMBASEDSP-NEXT:    ldr r0, .LCPI2_0
+; CHECK-ARMBASEDSP-NEXT:    cmn r1, #32768
+; CHECK-ARMBASEDSP-NEXT:    movgt r0, r1
+; CHECK-ARMBASEDSP-NEXT:    bx lr
+; CHECK-ARMBASEDSP-NEXT:    .p2align 2
+; CHECK-ARMBASEDSP-NEXT:  @ %bb.1:
+; CHECK-ARMBASEDSP-NEXT:  .LCPI2_0:
+; CHECK-ARMBASEDSP-NEXT:    .long 4294934528 @ 0xffff8000
+;
+; CHECK-ARMDSP-LABEL: func16:
+; CHECK-ARMDSP:       @ %bb.0:
+; CHECK-ARMDSP-NEXT:    qsub16 r0, r0, r1
+; CHECK-ARMDSP-NEXT:    sxth r0, r0
+; CHECK-ARMDSP-NEXT:    bx lr
   %tmp = call i16 @llvm.ssub.sat.i16(i16 %x, i16 %y)
   ret i16 %tmp
 }
@@ -286,25 +314,46 @@ define signext i8 @func8(i8 signext %x, i8 signext %y) nounwind {
 ; CHECK-T1-NEXT:  .LBB3_4:
 ; CHECK-T1-NEXT:    bx lr
 ;
-; CHECK-T2-LABEL: func8:
-; CHECK-T2:       @ %bb.0:
-; CHECK-T2-NEXT:    subs r0, r0, r1
-; CHECK-T2-NEXT:    cmp r0, #127
-; CHECK-T2-NEXT:    it ge
-; CHECK-T2-NEXT:    movge r0, #127
-; CHECK-T2-NEXT:    cmn.w r0, #128
-; CHECK-T2-NEXT:    it le
-; CHECK-T2-NEXT:    mvnle r0, #127
-; CHECK-T2-NEXT:    bx lr
+; CHECK-T2NODSP-LABEL: func8:
+; CHECK-T2NODSP:       @ %bb.0:
+; CHECK-T2NODSP-NEXT:    subs r0, r0, r1
+; CHECK-T2NODSP-NEXT:    cmp r0, #127
+; CHECK-T2NODSP-NEXT:    it ge
+; CHECK-T2NODSP-NEXT:    movge r0, #127
+; CHECK-T2NODSP-NEXT:    cmn.w r0, #128
+; CHECK-T2NODSP-NEXT:    it le
+; CHECK-T2NODSP-NEXT:    mvnle r0, #127
+; CHECK-T2NODSP-NEXT:    bx lr
 ;
-; CHECK-ARM-LABEL: func8:
-; CHECK-ARM:       @ %bb.0:
-; CHECK-ARM-NEXT:    sub r0, r0, r1
-; CHECK-ARM-NEXT:    cmp r0, #127
-; CHECK-ARM-NEXT:    movge r0, #127
-; CHECK-ARM-NEXT:    cmn r0, #128
-; CHECK-ARM-NEXT:    mvnle r0, #127
-; CHECK-ARM-NEXT:    bx lr
+; CHECK-T2DSP-LABEL: func8:
+; CHECK-T2DSP:       @ %bb.0:
+; CHECK-T2DSP-NEXT:    qsub8 r0, r0, r1
+; CHECK-T2DSP-NEXT:    sxtb r0, r0
+; CHECK-T2DSP-NEXT:    bx lr
+;
+; CHECK-ARMNODPS-LABEL: func8:
+; CHECK-ARMNODPS:       @ %bb.0:
+; CHECK-ARMNODPS-NEXT:    sub r0, r0, r1
+; CHECK-ARMNODPS-NEXT:    cmp r0, #127
+; CHECK-ARMNODPS-NEXT:    movge r0, #127
+; CHECK-ARMNODPS-NEXT:    cmn r0, #128
+; CHECK-ARMNODPS-NEXT:    mvnle r0, #127
+; CHECK-ARMNODPS-NEXT:    bx lr
+;
+; CHECK-ARMBASEDSP-LABEL: func8:
+; CHECK-ARMBASEDSP:       @ %bb.0:
+; CHECK-ARMBASEDSP-NEXT:    sub r0, r0, r1
+; CHECK-ARMBASEDSP-NEXT:    cmp r0, #127
+; CHECK-ARMBASEDSP-NEXT:    movge r0, #127
+; CHECK-ARMBASEDSP-NEXT:    cmn r0, #128
+; CHECK-ARMBASEDSP-NEXT:    mvnle r0, #127
+; CHECK-ARMBASEDSP-NEXT:    bx lr
+;
+; CHECK-ARMDSP-LABEL: func8:
+; CHECK-ARMDSP:       @ %bb.0:
+; CHECK-ARMDSP-NEXT:    qsub8 r0, r0, r1
+; CHECK-ARMDSP-NEXT:    sxtb r0, r0
+; CHECK-ARMDSP-NEXT:    bx lr
   %tmp = call i8 @llvm.ssub.sat.i8(i8 %x, i8 %y)
   ret i8 %tmp
 }
diff --git a/test/CodeGen/ARM/ssub_sat_plus.ll b/test/CodeGen/ARM/ssub_sat_plus.ll
index 9a598408d9e..02a0de24f1e 100644
--- a/test/CodeGen/ARM/ssub_sat_plus.ll
+++ b/test/CodeGen/ARM/ssub_sat_plus.ll
@@ -245,34 +245,34 @@ define signext i16 @func16(i16 signext %x, i16 signext %y, i16 signext %z) nounw
 ; CHECK-T1-NEXT:  .LCPI2_1:
 ; CHECK-T1-NEXT:    .long 4294934528 @ 0xffff8000
 ;
-; CHECK-T2-LABEL: func16:
-; CHECK-T2:       @ %bb.0:
-; CHECK-T2-NEXT:    muls r1, r2, r1
-; CHECK-T2-NEXT:    sxth r1, r1
-; CHECK-T2-NEXT:    subs r0, r0, r1
-; CHECK-T2-NEXT:    movw r1, #32767
-; CHECK-T2-NEXT:    cmp r0, r1
-; CHECK-T2-NEXT:    it lt
-; CHECK-T2-NEXT:    movlt r1, r0
-; CHECK-T2-NEXT:    movw r0, #32768
-; CHECK-T2-NEXT:    movt r0, #65535
-; CHECK-T2-NEXT:    cmn.w r1, #32768
-; CHECK-T2-NEXT:    it gt
-; CHECK-T2-NEXT:    movgt r0, r1
-; CHECK-T2-NEXT:    bx lr
+; CHECK-T2NODSP-LABEL: func16:
+; CHECK-T2NODSP:       @ %bb.0:
+; CHECK-T2NODSP-NEXT:    muls r1, r2, r1
+; CHECK-T2NODSP-NEXT:    sxth r1, r1
+; CHECK-T2NODSP-NEXT:    subs r0, r0, r1
+; CHECK-T2NODSP-NEXT:    movw r1, #32767
+; CHECK-T2NODSP-NEXT:    cmp r0, r1
+; CHECK-T2NODSP-NEXT:    it lt
+; CHECK-T2NODSP-NEXT:    movlt r1, r0
+; CHECK-T2NODSP-NEXT:    movw r0, #32768
+; CHECK-T2NODSP-NEXT:    movt r0, #65535
+; CHECK-T2NODSP-NEXT:    cmn.w r1, #32768
+; CHECK-T2NODSP-NEXT:    it gt
+; CHECK-T2NODSP-NEXT:    movgt r0, r1
+; CHECK-T2NODSP-NEXT:    bx lr
+;
+; CHECK-T2DSP-LABEL: func16:
+; CHECK-T2DSP:       @ %bb.0:
+; CHECK-T2DSP-NEXT:    muls r1, r2, r1
+; CHECK-T2DSP-NEXT:    qsub16 r0, r0, r1
+; CHECK-T2DSP-NEXT:    sxth r0, r0
+; CHECK-T2DSP-NEXT:    bx lr
 ;
 ; CHECK-ARM-LABEL: func16:
 ; CHECK-ARM:       @ %bb.0:
 ; CHECK-ARM-NEXT:    smulbb r1, r1, r2
-; CHECK-ARM-NEXT:    sxth r1, r1
-; CHECK-ARM-NEXT:    sub r0, r0, r1
-; CHECK-ARM-NEXT:    movw r1, #32767
-; CHECK-ARM-NEXT:    cmp r0, r1
-; CHECK-ARM-NEXT:    movlt r1, r0
-; CHECK-ARM-NEXT:    movw r0, #32768
-; CHECK-ARM-NEXT:    movt r0, #65535
-; CHECK-ARM-NEXT:    cmn r1, #32768
-; CHECK-ARM-NEXT:    movgt r0, r1
+; CHECK-ARM-NEXT:    qsub16 r0, r0, r1
+; CHECK-ARM-NEXT:    sxth r0, r0
 ; CHECK-ARM-NEXT:    bx lr
   %a = mul i16 %y, %z
   %tmp = call i16 @llvm.ssub.sat.i16(i16 %x, i16 %a)
@@ -299,28 +299,31 @@ define signext i8 @func8(i8 signext %x, i8 signext %y, i8 signext %z) nounwind {
 ; CHECK-T1-NEXT:  .LBB3_4:
 ; CHECK-T1-NEXT:    bx lr
 ;
-; CHECK-T2-LABEL: func8:
-; CHECK-T2:       @ %bb.0:
-; CHECK-T2-NEXT:    muls r1, r2, r1
-; CHECK-T2-NEXT:    sxtb r1, r1
-; CHECK-T2-NEXT:    subs r0, r0, r1
-; CHECK-T2-NEXT:    cmp r0, #127
-; CHECK-T2-NEXT:    it ge
-; CHECK-T2-NEXT:    movge r0, #127
-; CHECK-T2-NEXT:    cmn.w r0, #128
-; CHECK-T2-NEXT:    it le
-; CHECK-T2-NEXT:    mvnle r0, #127
-; CHECK-T2-NEXT:    bx lr
+; CHECK-T2NODSP-LABEL: func8:
+; CHECK-T2NODSP:       @ %bb.0:
+; CHECK-T2NODSP-NEXT:    muls r1, r2, r1
+; CHECK-T2NODSP-NEXT:    sxtb r1, r1
+; CHECK-T2NODSP-NEXT:    subs r0, r0, r1
+; CHECK-T2NODSP-NEXT:    cmp r0, #127
+; CHECK-T2NODSP-NEXT:    it ge
+; CHECK-T2NODSP-NEXT:    movge r0, #127
+; CHECK-T2NODSP-NEXT:    cmn.w r0, #128
+; CHECK-T2NODSP-NEXT:    it le
+; CHECK-T2NODSP-NEXT:    mvnle r0, #127
+; CHECK-T2NODSP-NEXT:    bx lr
+;
+; CHECK-T2DSP-LABEL: func8:
+; CHECK-T2DSP:       @ %bb.0:
+; CHECK-T2DSP-NEXT:    muls r1, r2, r1
+; CHECK-T2DSP-NEXT:    qsub8 r0, r0, r1
+; CHECK-T2DSP-NEXT:    sxtb r0, r0
+; CHECK-T2DSP-NEXT:    bx lr
 ;
 ; CHECK-ARM-LABEL: func8:
 ; CHECK-ARM:       @ %bb.0:
 ; CHECK-ARM-NEXT:    smulbb r1, r1, r2
-; CHECK-ARM-NEXT:    sxtb r1, r1
-; CHECK-ARM-NEXT:    sub r0, r0, r1
-; CHECK-ARM-NEXT:    cmp r0, #127
-; CHECK-ARM-NEXT:    movge r0, #127
-; CHECK-ARM-NEXT:    cmn r0, #128
-; CHECK-ARM-NEXT:    mvnle r0, #127
+; CHECK-ARM-NEXT:    qsub8 r0, r0, r1
+; CHECK-ARM-NEXT:    sxtb r0, r0
 ; CHECK-ARM-NEXT:    bx lr
   %a = mul i8 %y, %z
   %tmp = call i8 @llvm.ssub.sat.i8(i8 %x, i8 %a)