x64: Remove unnecessary register use when comparing against constants (#4645)

https://github.com/bytecodealliance/wasmtime/pull/4645

cranelift/codegen/src/isa/x64/inst.isle

@@ -1080,6 +1080,9 @@
 (decl cc_invert (CC) CC)
 (extern constructor cc_invert cc_invert)
 
+(decl intcc_reverse (IntCC) IntCC)
+(extern constructor intcc_reverse intcc_reverse)
+
 (decl floatcc_inverse (FloatCC) FloatCC)
 (extern constructor floatcc_inverse floatcc_inverse)
 
@@ -3178,6 +3181,13 @@
       (let ((size OperandSize (raw_operand_size_of_type ty)))
         (icmp_cond_result (x64_cmp size b a) cc)))
 
+;; As a special case, reverse the arguments to the comparison when the LHS is a
+;; constant. This ensures that we avoid moving the constant into a register when
+;; performing the comparison.
+(rule (emit_cmp cc (and (simm32_from_value a) (value_type ty)) b)
+      (let ((size OperandSize (raw_operand_size_of_type ty)))
+        (icmp_cond_result (x64_cmp size a b) (intcc_reverse cc))))
+
 ;; For I128 values (held in two GPRs), the instruction sequences depend on what
 ;; kind of condition is tested.
 (rule (emit_cmp (IntCC.Equal) a @ (value_type $I128) b)

cranelift/codegen/src/isa/x64/lower/isle.rs

@@ -610,6 +610,11 @@ where
         }
     }
 
+    #[inline]
+    fn intcc_reverse(&mut self, cc: &IntCC) -> IntCC {
+        cc.reverse()
+    }
+
     #[inline]
     fn floatcc_inverse(&mut self, cc: &FloatCC) -> FloatCC {
         cc.inverse()

cranelift/filetests/filetests/isa/x64/b1.clif

@@ -73,6 +73,70 @@ block2:
 ;   popq    %rbp
 ;   ret
 
+function %f3(i64) -> i32 {
+block0(v0: i64):
+  v1 = iconst.i32 1
+  v2 = load.i32 v0
+  v3 = icmp eq v1, v2
+  brnz v3, block1
+  jump block2
+block1:
+  v4 = iconst.i32 1
+  return v4
+block2:
+  v5 = iconst.i32 1
+  return v5
+}
+
+;   pushq   %rbp
+;   movq    %rsp, %rbp
+; block0:
+;   movl    0(%rdi), %r8d
+;   cmpl    $1, %r8d
+;   jz      label1; j label2
+; block1:
+;   movl    $1, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; block2:
+;   movl    $1, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+
+function %f4(i64) -> i32 {
+block0(v0: i64):
+  v1 = iconst.i32 1
+  v2 = load.i32 v0
+  v3 = icmp eq v2, v1
+  brnz v3, block1
+  jump block2
+block1:
+  v4 = iconst.i32 1
+  return v4
+block2:
+  v5 = iconst.i32 1
+  return v5
+}
+
+;   pushq   %rbp
+;   movq    %rsp, %rbp
+; block0:
+;   movl    0(%rdi), %r8d
+;   cmpl    $1, %r8d
+;   jz      label1; j label2
+; block1:
+;   movl    $1, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; block2:
+;   movl    $1, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+
 function %test_x_slt_0_i64(i64) -> b1 {
 block0(v0: i64):
     v1 = iconst.i64 0