[Codegen] Alter the default promotion for saturating adds and subs

The default promotion for the add_sat/sub_sat nodes currently does:
    ANY_EXTEND iN to iM
    SHL by M-N
    [US][ADD|SUB]SAT
    L/ASHR by M-N

If the promoted add_sat or sub_sat node is not legal, this can produce code
that effectively does a lot of shifting (and requiring large constants to be
materialised) just to use the overflow flag. It is simpler to just do the
saturation manually, using the higher bitwidth addition and a min/max against
the saturating bounds. That is what this patch attempts to do.

Differential Revision: https://reviews.llvm.org/D68926

llvm-svn: 375211

8 files changed, 124 insertions, 143 deletions

diff --git a/llvm/test/CodeGen/X86/sadd_sat.ll b/llvm/test/CodeGen/X86/sadd_sat.ll
index 485e547f594..ffb4e731f9e 100644
--- a/llvm/test/CodeGen/X86/sadd_sat.ll
+++ b/llvm/test/CodeGen/X86/sadd_sat.ll
@@ -158,36 +158,29 @@ define signext i8 @func8(i8 signext %x, i8 signext %y) nounwind {
 define signext i4 @func3(i4 signext %x, i4 signext %y) nounwind {
 ; X86-LABEL: func3:
 ; X86:       # %bb.0:
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %dl
-; X86-NEXT:    shlb $4, %dl
-; X86-NEXT:    shlb $4, %cl
-; X86-NEXT:    xorl %eax, %eax
-; X86-NEXT:    movb %cl, %ch
-; X86-NEXT:    addb %dl, %ch
-; X86-NEXT:    setns %al
-; X86-NEXT:    addl $127, %eax
-; X86-NEXT:    addb %dl, %cl
-; X86-NEXT:    movzbl %cl, %ecx
-; X86-NEXT:    cmovol %eax, %ecx
-; X86-NEXT:    sarb $4, %cl
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    addb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    movzbl %al, %ecx
+; X86-NEXT:    cmpb $7, %al
+; X86-NEXT:    movl $7, %eax
+; X86-NEXT:    cmovll %ecx, %eax
+; X86-NEXT:    cmpb $-8, %al
+; X86-NEXT:    movl $248, %ecx
+; X86-NEXT:    cmovgl %eax, %ecx
 ; X86-NEXT:    movsbl %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: func3:
 ; X64:       # %bb.0:
-; X64-NEXT:    shlb $4, %sil
-; X64-NEXT:    shlb $4, %dil
-; X64-NEXT:    xorl %eax, %eax
-; X64-NEXT:    movl %edi, %ecx
-; X64-NEXT:    addb %sil, %cl
-; X64-NEXT:    setns %al
-; X64-NEXT:    addl $127, %eax
 ; X64-NEXT:    addb %sil, %dil
-; X64-NEXT:    movzbl %dil, %ecx
-; X64-NEXT:    cmovol %eax, %ecx
-; X64-NEXT:    sarb $4, %cl
-; X64-NEXT:    movsbl %cl, %eax
+; X64-NEXT:    movzbl %dil, %eax
+; X64-NEXT:    cmpb $7, %al
+; X64-NEXT:    movl $7, %ecx
+; X64-NEXT:    cmovll %eax, %ecx
+; X64-NEXT:    cmpb $-8, %cl
+; X64-NEXT:    movl $248, %eax
+; X64-NEXT:    cmovgl %ecx, %eax
+; X64-NEXT:    movsbl %al, %eax
 ; X64-NEXT:    retq
   %tmp = call i4 @llvm.sadd.sat.i4(i4 %x, i4 %y);
   ret i4 %tmp;
diff --git a/llvm/test/CodeGen/X86/sadd_sat_plus.ll b/llvm/test/CodeGen/X86/sadd_sat_plus.ll
index 76cc1613b47..2c902be8097 100644
--- a/llvm/test/CodeGen/X86/sadd_sat_plus.ll
+++ b/llvm/test/CodeGen/X86/sadd_sat_plus.ll
@@ -169,21 +169,19 @@ define signext i8 @func8(i8 signext %x, i8 signext %y, i8 signext %z) nounwind {
 define signext i4 @func4(i4 signext %x, i4 signext %y, i4 signext %z) nounwind {
 ; X86-LABEL: func4:
 ; X86:       # %bb.0:
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %dl
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    mulb {{[0-9]+}}(%esp)
 ; X86-NEXT:    shlb $4, %al
-; X86-NEXT:    shlb $4, %dl
-; X86-NEXT:    xorl %ecx, %ecx
-; X86-NEXT:    movb %dl, %ah
-; X86-NEXT:    addb %al, %ah
-; X86-NEXT:    setns %cl
-; X86-NEXT:    addl $127, %ecx
-; X86-NEXT:    addb %al, %dl
-; X86-NEXT:    movzbl %dl, %eax
-; X86-NEXT:    cmovol %ecx, %eax
 ; X86-NEXT:    sarb $4, %al
-; X86-NEXT:    movsbl %al, %eax
+; X86-NEXT:    addb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    movzbl %al, %ecx
+; X86-NEXT:    cmpb $7, %al
+; X86-NEXT:    movl $7, %eax
+; X86-NEXT:    cmovll %ecx, %eax
+; X86-NEXT:    cmpb $-8, %al
+; X86-NEXT:    movl $248, %ecx
+; X86-NEXT:    cmovgl %eax, %ecx
+; X86-NEXT:    movsbl %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: func4:
@@ -192,16 +190,15 @@ define signext i4 @func4(i4 signext %x, i4 signext %y, i4 signext %z) nounwind {
 ; X64-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-NEXT:    mulb %dl
 ; X64-NEXT:    shlb $4, %al
-; X64-NEXT:    shlb $4, %dil
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    movl %edi, %edx
-; X64-NEXT:    addb %al, %dl
-; X64-NEXT:    setns %cl
-; X64-NEXT:    addl $127, %ecx
-; X64-NEXT:    addb %al, %dil
-; X64-NEXT:    movzbl %dil, %eax
-; X64-NEXT:    cmovol %ecx, %eax
 ; X64-NEXT:    sarb $4, %al
+; X64-NEXT:    addb %dil, %al
+; X64-NEXT:    movzbl %al, %eax
+; X64-NEXT:    cmpb $7, %al
+; X64-NEXT:    movl $7, %ecx
+; X64-NEXT:    cmovll %eax, %ecx
+; X64-NEXT:    cmpb $-8, %cl
+; X64-NEXT:    movl $248, %eax
+; X64-NEXT:    cmovgl %ecx, %eax
 ; X64-NEXT:    movsbl %al, %eax
 ; X64-NEXT:    retq
   %a = mul i4 %y, %z
diff --git a/llvm/test/CodeGen/X86/ssub_sat.ll b/llvm/test/CodeGen/X86/ssub_sat.ll
index a92905a975b..5e7a3f93506 100644
--- a/llvm/test/CodeGen/X86/ssub_sat.ll
+++ b/llvm/test/CodeGen/X86/ssub_sat.ll
@@ -147,34 +147,29 @@ define signext i8 @func8(i8 signext %x, i8 signext %y) nounwind {
 define signext i4 @func3(i4 signext %x, i4 signext %y) nounwind {
 ; X86-LABEL: func3:
 ; X86:       # %bb.0:
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %dl
-; X86-NEXT:    shlb $4, %dl
-; X86-NEXT:    shlb $4, %cl
-; X86-NEXT:    xorl %eax, %eax
-; X86-NEXT:    cmpb %dl, %cl
-; X86-NEXT:    setns %al
-; X86-NEXT:    addl $127, %eax
-; X86-NEXT:    subb %dl, %cl
-; X86-NEXT:    movzbl %cl, %ecx
-; X86-NEXT:    cmovol %eax, %ecx
-; X86-NEXT:    sarb $4, %cl
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    subb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    movzbl %al, %ecx
+; X86-NEXT:    cmpb $7, %al
+; X86-NEXT:    movl $7, %eax
+; X86-NEXT:    cmovll %ecx, %eax
+; X86-NEXT:    cmpb $-8, %al
+; X86-NEXT:    movl $248, %ecx
+; X86-NEXT:    cmovgl %eax, %ecx
 ; X86-NEXT:    movsbl %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: func3:
 ; X64:       # %bb.0:
-; X64-NEXT:    shlb $4, %sil
-; X64-NEXT:    shlb $4, %dil
-; X64-NEXT:    xorl %eax, %eax
-; X64-NEXT:    cmpb %sil, %dil
-; X64-NEXT:    setns %al
-; X64-NEXT:    addl $127, %eax
 ; X64-NEXT:    subb %sil, %dil
-; X64-NEXT:    movzbl %dil, %ecx
-; X64-NEXT:    cmovol %eax, %ecx
-; X64-NEXT:    sarb $4, %cl
-; X64-NEXT:    movsbl %cl, %eax
+; X64-NEXT:    movzbl %dil, %eax
+; X64-NEXT:    cmpb $7, %al
+; X64-NEXT:    movl $7, %ecx
+; X64-NEXT:    cmovll %eax, %ecx
+; X64-NEXT:    cmpb $-8, %cl
+; X64-NEXT:    movl $248, %eax
+; X64-NEXT:    cmovgl %ecx, %eax
+; X64-NEXT:    movsbl %al, %eax
 ; X64-NEXT:    retq
   %tmp = call i4 @llvm.ssub.sat.i4(i4 %x, i4 %y)
   ret i4 %tmp
diff --git a/llvm/test/CodeGen/X86/ssub_sat_plus.ll b/llvm/test/CodeGen/X86/ssub_sat_plus.ll
index ed0718ea6a8..12c587f1a19 100644
--- a/llvm/test/CodeGen/X86/ssub_sat_plus.ll
+++ b/llvm/test/CodeGen/X86/ssub_sat_plus.ll
@@ -158,19 +158,19 @@ define signext i8 @func8(i8 signext %x, i8 signext %y, i8 signext %z) nounwind {
 define signext i4 @func4(i4 signext %x, i4 signext %y, i4 signext %z) nounwind {
 ; X86-LABEL: func4:
 ; X86:       # %bb.0:
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    mulb {{[0-9]+}}(%esp)
 ; X86-NEXT:    shlb $4, %al
-; X86-NEXT:    shlb $4, %dl
-; X86-NEXT:    xorl %ecx, %ecx
-; X86-NEXT:    cmpb %al, %dl
-; X86-NEXT:    setns %cl
-; X86-NEXT:    addl $127, %ecx
-; X86-NEXT:    subb %al, %dl
-; X86-NEXT:    movzbl %dl, %eax
-; X86-NEXT:    cmovol %ecx, %eax
 ; X86-NEXT:    sarb $4, %al
+; X86-NEXT:    subb %al, %cl
+; X86-NEXT:    movzbl %cl, %eax
+; X86-NEXT:    cmpb $7, %cl
+; X86-NEXT:    movl $7, %ecx
+; X86-NEXT:    cmovll %eax, %ecx
+; X86-NEXT:    cmpb $-8, %cl
+; X86-NEXT:    movl $248, %eax
+; X86-NEXT:    cmovgl %ecx, %eax
 ; X86-NEXT:    movsbl %al, %eax
 ; X86-NEXT:    retl
 ;
@@ -180,15 +180,15 @@ define signext i4 @func4(i4 signext %x, i4 signext %y, i4 signext %z) nounwind {
 ; X64-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-NEXT:    mulb %dl
 ; X64-NEXT:    shlb $4, %al
-; X64-NEXT:    shlb $4, %dil
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    cmpb %al, %dil
-; X64-NEXT:    setns %cl
-; X64-NEXT:    addl $127, %ecx
+; X64-NEXT:    sarb $4, %al
 ; X64-NEXT:    subb %al, %dil
 ; X64-NEXT:    movzbl %dil, %eax
-; X64-NEXT:    cmovol %ecx, %eax
-; X64-NEXT:    sarb $4, %al
+; X64-NEXT:    cmpb $7, %al
+; X64-NEXT:    movl $7, %ecx
+; X64-NEXT:    cmovll %eax, %ecx
+; X64-NEXT:    cmpb $-8, %cl
+; X64-NEXT:    movl $248, %eax
+; X64-NEXT:    cmovgl %ecx, %eax
 ; X64-NEXT:    movsbl %al, %eax
 ; X64-NEXT:    retq
   %a = mul i4 %y, %z
diff --git a/llvm/test/CodeGen/X86/uadd_sat.ll b/llvm/test/CodeGen/X86/uadd_sat.ll
index 7124e453523..3d3a1f534b0 100644
--- a/llvm/test/CodeGen/X86/uadd_sat.ll
+++ b/llvm/test/CodeGen/X86/uadd_sat.ll
@@ -98,26 +98,21 @@ define zeroext i4 @func3(i4 zeroext %x, i4 zeroext %y) nounwind {
 ; X86-LABEL: func3:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT:    shlb $4, %cl
-; X86-NEXT:    shlb $4, %al
-; X86-NEXT:    addb %cl, %al
+; X86-NEXT:    addb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    movzbl %al, %ecx
+; X86-NEXT:    cmpb $15, %al
+; X86-NEXT:    movl $15, %eax
+; X86-NEXT:    cmovbl %ecx, %eax
 ; X86-NEXT:    movzbl %al, %eax
-; X86-NEXT:    movl $255, %ecx
-; X86-NEXT:    cmovael %eax, %ecx
-; X86-NEXT:    shrb $4, %cl
-; X86-NEXT:    movzbl %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: func3:
 ; X64:       # %bb.0:
-; X64-NEXT:    shlb $4, %sil
-; X64-NEXT:    shlb $4, %dil
 ; X64-NEXT:    addb %sil, %dil
 ; X64-NEXT:    movzbl %dil, %eax
-; X64-NEXT:    movl $255, %ecx
-; X64-NEXT:    cmovael %eax, %ecx
-; X64-NEXT:    shrb $4, %cl
+; X64-NEXT:    cmpb $15, %al
+; X64-NEXT:    movl $15, %ecx
+; X64-NEXT:    cmovbl %eax, %ecx
 ; X64-NEXT:    movzbl %cl, %eax
 ; X64-NEXT:    retq
   %tmp = call i4 @llvm.uadd.sat.i4(i4 %x, i4 %y)
diff --git a/llvm/test/CodeGen/X86/uadd_sat_plus.ll b/llvm/test/CodeGen/X86/uadd_sat_plus.ll
index 428e2933c46..ac102033717 100644
--- a/llvm/test/CodeGen/X86/uadd_sat_plus.ll
+++ b/llvm/test/CodeGen/X86/uadd_sat_plus.ll
@@ -108,17 +108,15 @@ define zeroext i8 @func8(i8 zeroext %x, i8 zeroext %y, i8 zeroext %z) nounwind {
 define zeroext i4 @func4(i4 zeroext %x, i4 zeroext %y, i4 zeroext %z) nounwind {
 ; X86-LABEL: func4:
 ; X86:       # %bb.0:
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    mulb {{[0-9]+}}(%esp)
-; X86-NEXT:    shlb $4, %al
-; X86-NEXT:    shlb $4, %cl
-; X86-NEXT:    addb %al, %cl
-; X86-NEXT:    movzbl %cl, %eax
-; X86-NEXT:    movl $255, %ecx
-; X86-NEXT:    cmovael %eax, %ecx
-; X86-NEXT:    shrb $4, %cl
-; X86-NEXT:    movzbl %cl, %eax
+; X86-NEXT:    andb $15, %al
+; X86-NEXT:    addb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    movzbl %al, %ecx
+; X86-NEXT:    cmpb $15, %al
+; X86-NEXT:    movl $15, %eax
+; X86-NEXT:    cmovbl %ecx, %eax
+; X86-NEXT:    movzbl %al, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: func4:
@@ -126,13 +124,12 @@ define zeroext i4 @func4(i4 zeroext %x, i4 zeroext %y, i4 zeroext %z) nounwind {
 ; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-NEXT:    mulb %dl
-; X64-NEXT:    shlb $4, %al
-; X64-NEXT:    shlb $4, %dil
-; X64-NEXT:    addb %al, %dil
-; X64-NEXT:    movzbl %dil, %eax
-; X64-NEXT:    movl $255, %ecx
-; X64-NEXT:    cmovael %eax, %ecx
-; X64-NEXT:    shrb $4, %cl
+; X64-NEXT:    andb $15, %al
+; X64-NEXT:    addb %dil, %al
+; X64-NEXT:    movzbl %al, %eax
+; X64-NEXT:    cmpb $15, %al
+; X64-NEXT:    movl $15, %ecx
+; X64-NEXT:    cmovbl %eax, %ecx
 ; X64-NEXT:    movzbl %cl, %eax
 ; X64-NEXT:    retq
   %a = mul i4 %y, %z
diff --git a/llvm/test/CodeGen/X86/usub_sat.ll b/llvm/test/CodeGen/X86/usub_sat.ll
index a7de6b593be..f39aee388fd 100644
--- a/llvm/test/CodeGen/X86/usub_sat.ll
+++ b/llvm/test/CodeGen/X86/usub_sat.ll
@@ -97,28 +97,27 @@ define zeroext i8 @func8(i8 zeroext %x, i8 zeroext %y) nounwind {
 define zeroext i4 @func3(i4 zeroext %x, i4 zeroext %y) nounwind {
 ; X86-LABEL: func3:
 ; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT:    shlb $4, %cl
-; X86-NEXT:    shlb $4, %al
-; X86-NEXT:    xorl %edx, %edx
-; X86-NEXT:    subb %cl, %al
-; X86-NEXT:    movzbl %al, %eax
-; X86-NEXT:    cmovbl %edx, %eax
-; X86-NEXT:    shrb $4, %al
-; X86-NEXT:    movzbl %al, %eax
+; X86-NEXT:    movzbl %cl, %edx
+; X86-NEXT:    movzbl %al, %ebx
+; X86-NEXT:    cmpb %al, %cl
+; X86-NEXT:    cmoval %edx, %ebx
+; X86-NEXT:    subb %al, %bl
+; X86-NEXT:    movzbl %bl, %eax
+; X86-NEXT:    andl $15, %eax
+; X86-NEXT:    popl %ebx
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: func3:
 ; X64:       # %bb.0:
-; X64-NEXT:    shlb $4, %sil
-; X64-NEXT:    shlb $4, %dil
-; X64-NEXT:    xorl %eax, %eax
-; X64-NEXT:    subb %sil, %dil
-; X64-NEXT:    movzbl %dil, %ecx
-; X64-NEXT:    cmovbl %eax, %ecx
-; X64-NEXT:    shrb $4, %cl
-; X64-NEXT:    movzbl %cl, %eax
+; X64-NEXT:    cmpb %sil, %dil
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    cmoval %edi, %eax
+; X64-NEXT:    subb %sil, %al
+; X64-NEXT:    movzbl %al, %eax
+; X64-NEXT:    andl $15, %eax
 ; X64-NEXT:    retq
   %tmp = call i4 @llvm.usub.sat.i4(i4 %x, i4 %y)
   ret i4 %tmp
diff --git a/llvm/test/CodeGen/X86/usub_sat_plus.ll b/llvm/test/CodeGen/X86/usub_sat_plus.ll
index 6a9e5a46642..f56e0962eac 100644
--- a/llvm/test/CodeGen/X86/usub_sat_plus.ll
+++ b/llvm/test/CodeGen/X86/usub_sat_plus.ll
@@ -111,17 +111,22 @@ define zeroext i8 @func8(i8 zeroext %x, i8 zeroext %y, i8 zeroext %z) nounwind {
 define zeroext i4 @func4(i4 zeroext %x, i4 zeroext %y, i4 zeroext %z) nounwind {
 ; X86-LABEL: func4:
 ; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    mulb {{[0-9]+}}(%esp)
-; X86-NEXT:    shlb $4, %al
-; X86-NEXT:    shlb $4, %cl
-; X86-NEXT:    xorl %edx, %edx
-; X86-NEXT:    subb %al, %cl
-; X86-NEXT:    movzbl %cl, %eax
-; X86-NEXT:    cmovbl %edx, %eax
-; X86-NEXT:    shrb $4, %al
-; X86-NEXT:    movzbl %al, %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andb $15, %dl
+; X86-NEXT:    movzbl %dl, %esi
+; X86-NEXT:    movzbl %cl, %ebx
+; X86-NEXT:    cmpb %dl, %cl
+; X86-NEXT:    cmovbel %esi, %ebx
+; X86-NEXT:    subb %al, %bl
+; X86-NEXT:    movzbl %bl, %eax
+; X86-NEXT:    andl $15, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %ebx
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: func4:
@@ -129,14 +134,14 @@ define zeroext i4 @func4(i4 zeroext %x, i4 zeroext %y, i4 zeroext %z) nounwind {
 ; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-NEXT:    mulb %dl
-; X64-NEXT:    shlb $4, %al
-; X64-NEXT:    shlb $4, %dil
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    subb %al, %dil
-; X64-NEXT:    movzbl %dil, %eax
-; X64-NEXT:    cmovbl %ecx, %eax
-; X64-NEXT:    shrb $4, %al
-; X64-NEXT:    movzbl %al, %eax
+; X64-NEXT:    movl %eax, %ecx
+; X64-NEXT:    andb $15, %cl
+; X64-NEXT:    movzbl %cl, %ecx
+; X64-NEXT:    cmpb %cl, %dil
+; X64-NEXT:    cmoval %edi, %ecx
+; X64-NEXT:    subb %al, %cl
+; X64-NEXT:    movzbl %cl, %eax
+; X64-NEXT:    andl $15, %eax
 ; X64-NEXT:    retq
   %a = mul i4 %y, %z
   %tmp = call i4 @llvm.usub.sat.i4(i4 %x, i4 %a)