Subzero: Fix srem.i8/urem.i8 lowering for x86-64.

tests_lit/llvm2ice_tests/8bit.pnacl.ll

 ; This tries to be a comprehensive test of i8 operations.
 
 ; RUN: %p2i --filetype=obj --disassemble -i %s --args -O2 \
 ; RUN:   -allow-externally-defined-symbols | FileCheck %s
 ; RUN: %p2i --filetype=obj --disassemble -i %s --args -Om1 \
 ; RUN:   -allow-externally-defined-symbols | FileCheck %s
 
+; The following tests i8 srem/urem lowering on x86-64, specifically that the %ah
+; result gets copied into %al/%bl/%cl/%dl before moved into its final register.
+; This extra copy is forced by excluding al/bl/cl/dl by default (-reg-exclude),
+; but allowing them to be used if absolutely necessary (-reg-reserve).
+
+; RUN: %p2i --target=x8664 --filetype=obj --disassemble -i %s --args -O2 \
+; RUN:   -reg-exclude=al,bl,cl,dl -reg-reserve \
+; RUN:   -allow-externally-defined-symbols | FileCheck %s --check-prefix=REM
+
 declare void @useInt(i32 %x)
 
 define internal i32 @add8Bit(i32 %a, i32 %b) {
 entry:
   %a_8 = trunc i32 %a to i8
   %b_8 = trunc i32 %b to i8
   %add = add i8 %b_8, %a_8
   %ret = zext i8 %add to i32
   ret i32 %ret
 }
@@ skipping 78 matching lines @@
 define internal i32 @urem8Bit(i32 %a, i32 %b) {
 entry:
   %a_8 = trunc i32 %a to i8
   %b_8 = trunc i32 %b to i8
   %urem = urem i8 %b_8, %a_8
   %ret = zext i8 %urem to i32
   ret i32 %ret
 }
 ; CHECK-LABEL: urem8Bit
 ; CHECK: div {{[abcd]l|BYTE PTR}}
+; REM-LABEL: urem8Bit
+; REM: div
+; REM-NEXT: mov {{[abcd]}}l,ah
 
 define internal i32 @urem8BitConst(i32 %a) {
 entry:
   %a_8 = trunc i32 %a to i8
   %urem = urem i8 %a_8, 123
   %ret = zext i8 %urem to i32
   ret i32 %ret
 }
 ; CHECK-LABEL: urem8BitConst
 ; CHECK: div {{[abcd]l|BYTE PTR}}
+; REM-LABEL: urem8BitConst
 
 
 define internal i32 @sdiv8Bit(i32 %a, i32 %b) {
 entry:
   %a_8 = trunc i32 %a to i8
   %b_8 = trunc i32 %b to i8
   %sdiv = sdiv i8 %b_8, %a_8
   %ret = zext i8 %sdiv to i32
   ret i32 %ret
 }
@@ skipping 13 matching lines @@
 define internal i32 @srem8Bit(i32 %a, i32 %b) {
 entry:
   %a_8 = trunc i32 %a to i8
   %b_8 = trunc i32 %b to i8
   %srem = srem i8 %b_8, %a_8
   %ret = zext i8 %srem to i32
   ret i32 %ret
 }
 ; CHECK-LABEL: srem8Bit
 ; CHECK: idiv {{[abcd]l|BYTE PTR}}
+; REM-LABEL: srem8Bit
+; REM: idiv
+; REM-NEXT: mov {{[abcd]}}l,ah
 
 define internal i32 @srem8BitConst(i32 %a) {
 entry:
   %a_8 = trunc i32 %a to i8
   %srem = srem i8 %a_8, 123
   %ret = zext i8 %srem to i32
   ret i32 %ret
 }
 ; CHECK-LABEL: srem8BitConst
 ; CHECK: idiv {{[abcd]l|BYTE PTR}}
+; REM-LABEL: srem8BitConst
 
 define internal i32 @shl8Bit(i32 %a, i32 %b) {
 entry:
   %a_8 = trunc i32 %a to i8
   %b_8 = trunc i32 %b to i8
   %shl = shl i8 %b_8, %a_8
   %ret = zext i8 %shl to i32
   ret i32 %ret
 }
 ; CHECK-LABEL: shl8Bit
@@ skipping 197 matching lines @@
 ; CHECK: mov BYTE PTR {{.*}},{{[a-d]l}}
 
 define void @store_i8_const(i32 %addr_arg) {
 entry:
   %addr = inttoptr i32 %addr_arg to i8*
   store i8 123, i8* %addr, align 1
   ret void
 }
 ; CHECK-LABEL: store_i8_const
 ; CHECK: mov BYTE PTR {{.*}},0x7b