Handle GPR and vector shift ops. Handle pmull also.

tests_lit/llvm2ice_tests/nacl-other-intrinsics.ll

 ; This tests the NaCl intrinsics not related to atomic operations.
 
 ; TODO(jvoung): fix extra "CALLTARGETS" run. The llvm-objdump symbolizer
 ; doesn't know how to symbolize non-section-local functions.
 ; The newer LLVM 3.6 one does work, but watch out for other bugs.
 
 ; RUN: %p2i -i %s --args -O2 --verbose none \
 ; RUN:   | FileCheck --check-prefix=CALLTARGETS %s
 ; RUN: %p2i -i %s --args -O2 --verbose none -sandbox \
 ; RUN:   | llvm-mc -triple=i686-none-nacl -x86-asm-syntax=intel -filetype=obj \
@@ skipping 308 matching lines @@
 ; CHECK: ud2
 
 define i32 @test_bswap_16(i32 %x) {
 entry:
   %x_trunc = trunc i32 %x to i16
   %r = call i16 @llvm.bswap.i16(i16 %x_trunc)
   %r_zext = zext i16 %r to i32
   ret i32 %r_zext
 }
 ; CHECK-LABEL: test_bswap_16
-; CHECK: rol {{.*}}, 8
+; Make sure this is the right operand size so that the most significant bit
+; to least significant bit rotation happens at the right boundary.
+; CHECK: rol {{[abcd]x|si|di|bp|word ptr}}, 8
 
 define i32 @test_bswap_32(i32 %x) {
 entry:
   %r = call i32 @llvm.bswap.i32(i32 %x)
   ret i32 %r
 }
 ; CHECK-LABEL: test_bswap_32
 ; CHECK: bswap e{{.*}}
 
 define i64 @test_bswap_64(i64 %x) {
@@ skipping 176 matching lines @@
   ret void
 }
 ; CHECK-LABEL: test_stacksave_multiple
 ; At least 3 copies of esp, but probably more from having to do the allocas.
 ; CHECK: mov {{.*}}, esp
 ; CHECK: mov {{.*}}, esp
 ; CHECK: mov {{.*}}, esp
 ; CHECK: mov esp, {{.*}}
 
 ; ERRORS-NOT: ICE translation error