Change usage of naclcall and nacljmp pseudo-instructions to match x86 gas

test/CodeGen/X86/fast-isel-x86-64.ll

 ; RUN: llc < %s -mattr=-avx -fast-isel -mcpu=core2 -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s
 ; RUN: llc < %s -mattr=+avx -fast-isel -mcpu=core2 -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s --check-prefix=AVX
 ; RUN: llc < %s -fast-isel -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort -mtriple=x86_64-none-nacl | FileCheck %s --check-prefix=NACL64
 ; RUN: llc < %s -fast-isel -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort -mtriple=x86_64-none-nacl -relocation-model=pic | FileCheck %s --check-prefix=NACL64_PIC
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
 target triple = "x86_64-apple-darwin10.0.0"
 
 ; Make sure that fast-isel folds the immediate into the binop even though it
 ; is non-canonical.
@@ skipping 38 matching lines @@
 }
 
 ; NACL64 version uses i32 for 32-bit pointers.
 define i32 @test3_nacl64() nounwind {
   %A = ptrtoint i32* @G to i32
   ret i32 %A
 
 ; NACL64_PIC: test3_nacl64:
 ; NACL64_PIC: movl G@GOTPCREL(%rip), %eax
 ; NACL64_PIC-NEXT: popq    %r11
-; NACL64_PIC-NEXT: nacljmp %r11, %r15
+; NACL64_PIC-NEXT: nacljmp %r11d, %r15
 }
 
 
 ; rdar://9289558
 @rtx_length = external global [153 x i8]
 
 define i32 @test4(i64 %idxprom9) nounwind {
   %arrayidx10 = getelementptr inbounds [153 x i8]* @rtx_length, i32 0, i64 %idxprom9
   %tmp11 = load i8* %arrayidx10, align 1
   %conv = zext i8 %tmp11 to i32
@@ skipping 240 matching lines @@
   %b = call i8* @foo23()
   ret void
 ; CHECK-LABEL: test23:
 ; CHECK: call
 ; CHECK: movq  %rdi, %rax
 ; CHECK: ret
 ; NACL64: test23:
 ; NACL64: call
 ; NACL64: movl  %edi, %eax
 ; NACL64: popq %r11
-; NACL64: nacljmp %r11, %r15
+; NACL64: nacljmp %r11d, %r15
 }
 
 declare i8* @foo23()