x64: emit_cmp: use x64_test for comparisons with 0 (#6086)

* x64: emit_cmp: use x64_test for comparisons with 0

See #5869

* fixup! x64: emit_cmp: use x64_test for comparisons with 0

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

@@ -4375,9 +4375,20 @@
       (let ((size OperandSize (raw_operand_size_of_type ty)))
         (icmp_cond_result (x64_cmp size a b) (intcc_reverse cc))))
 
+;; Special case: use the test instruction for comparisons with 0.
+(rule 2 (emit_cmp cc a @ (value_type ty) (u64_from_iconst 0))
+      (let ((size OperandSize (raw_operand_size_of_type ty))
+            (a Gpr (put_in_reg a)))
+        (icmp_cond_result (x64_test size a a) cc)))
+
+(rule 3 (emit_cmp cc (u64_from_iconst 0) b @ (value_type ty))
+      (let ((size OperandSize (raw_operand_size_of_type ty))
+            (b Gpr (put_in_reg b)))
+        (icmp_cond_result (x64_test size b b) (intcc_reverse cc))))
+
 ;; For I128 values (held in two GPRs), the instruction sequences depend on what
 ;; kind of condition is tested.
-(rule 3 (emit_cmp (IntCC.Equal) a @ (value_type $I128) b)
+(rule 5 (emit_cmp (IntCC.Equal) a @ (value_type $I128) b)
       (let ((a_lo Gpr (value_regs_get_gpr a 0))
             (a_hi Gpr (value_regs_get_gpr a 1))
             (b_lo Gpr (value_regs_get_gpr b 0))
@@ -4398,7 +4409,7 @@
           (x64_test (OperandSize.Size64) (RegMemImm.Imm 1) cmp)
           (CC.NZ))))
 
-(rule 3 (emit_cmp (IntCC.NotEqual) a @ (value_type $I128) b)
+(rule 5 (emit_cmp (IntCC.NotEqual) a @ (value_type $I128) b)
       (let ((a_lo Gpr (value_regs_get_gpr a 0))
             (a_hi Gpr (value_regs_get_gpr a 1))
             (b_lo Gpr (value_regs_get_gpr b 0))
@@ -4413,7 +4424,7 @@
 
 ;; Result = (a_hi <> b_hi) ||
 ;;          (a_hi == b_hi && a_lo <> b_lo)
-(rule 2 (emit_cmp cc a @ (value_type $I128) b)
+(rule 4 (emit_cmp cc a @ (value_type $I128) b)
       (let ((a_lo Gpr (value_regs_get_gpr a 0))
             (a_hi Gpr (value_regs_get_gpr a 1))
             (b_lo Gpr (value_regs_get_gpr b 0))

cranelift/filetests/filetests/isa/x64/branches.clif

@@ -383,7 +383,7 @@ block2:
 ;   pushq   %rbp
 ;   movq    %rsp, %rbp
 ; block0:
-;   cmpq    $0, %rdi
+;   testq   %rdi, %rdi
 ;   jl      label2; j label1
 ; block1:
 ;   xorl    %eax, %eax, %eax
@@ -401,14 +401,14 @@ block2:
 ;   pushq %rbp
 ;   movq %rsp, %rbp
 ; block1: ; offset 0x4
-;   cmpq $0, %rdi
+;   testq %rdi, %rdi
-;   jl 0x15
+;   jl 0x14
-; block2: ; offset 0xe
+; block2: ; offset 0xd
 ;   xorl %eax, %eax
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
-; block3: ; offset 0x15
+; block3: ; offset 0x14
 ;   movl $1, %eax
 ;   movq %rbp, %rsp
 ;   popq %rbp
@@ -431,7 +431,7 @@ block2:
 ;   pushq   %rbp
 ;   movq    %rsp, %rbp
 ; block0:
-;   cmpl    $0, %edi
+;   testl   %edi, %edi
 ;   jl      label2; j label1
 ; block1:
 ;   xorl    %eax, %eax, %eax
@@ -449,8 +449,152 @@ block2:
 ;   pushq %rbp
 ;   movq %rsp, %rbp
 ; block1: ; offset 0x4
-;   cmpl $0, %edi
+;   testl %edi, %edi
-;   jl 0x14
+;   jl 0x13
+; block2: ; offset 0xc
+;   xorl %eax, %eax
+;   movq %rbp, %rsp
+;   popq %rbp
+;   retq
+; block3: ; offset 0x13
+;   movl $1, %eax
+;   movq %rbp, %rsp
+;   popq %rbp
+;   retq
+
+function %f8(i64) -> i8 {
+block0(v0: i64):
+  v1 = iconst.i64 0
+  v2 = icmp eq v0, v1
+  brif v2, block1, block2
+block1:
+  v3 = iconst.i8 1
+  return v3
+block2:
+  v4 = iconst.i8 0
+  return v4
+}
+
+; VCode:
+;   pushq   %rbp
+;   movq    %rsp, %rbp
+; block0:
+;   testq   %rdi, %rdi
+;   jz      label2; j label1
+; block1:
+;   xorl    %eax, %eax, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; block2:
+;   movl    $1, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; 
+; Disassembled:
+; block0: ; offset 0x0
+;   pushq %rbp
+;   movq %rsp, %rbp
+; block1: ; offset 0x4
+;   testq %rdi, %rdi
+;   je 0x14
+; block2: ; offset 0xd
+;   xorl %eax, %eax
+;   movq %rbp, %rsp
+;   popq %rbp
+;   retq
+; block3: ; offset 0x14
+;   movl $1, %eax
+;   movq %rbp, %rsp
+;   popq %rbp
+;   retq
+
+function %f9(i32) -> i8 {
+block0(v0: i32):
+  v1 = iconst.i32 0
+  v2 = icmp eq v0, v1
+  brif v2, block1, block2
+block1:
+  v3 = iconst.i8 1
+  return v3
+block2:
+  v4 = iconst.i8 0
+  return v4
+}
+
+; VCode:
+;   pushq   %rbp
+;   movq    %rsp, %rbp
+; block0:
+;   testl   %edi, %edi
+;   jz      label2; j label1
+; block1:
+;   xorl    %eax, %eax, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; block2:
+;   movl    $1, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; 
+; Disassembled:
+; block0: ; offset 0x0
+;   pushq %rbp
+;   movq %rsp, %rbp
+; block1: ; offset 0x4
+;   testl %edi, %edi
+;   je 0x13
+; block2: ; offset 0xc
+;   xorl %eax, %eax
+;   movq %rbp, %rsp
+;   popq %rbp
+;   retq
+; block3: ; offset 0x13
+;   movl $1, %eax
+;   movq %rbp, %rsp
+;   popq %rbp
+;   retq
+
+function %f10(i64) -> i8 {
+block0(v0: i64):
+  v1 = iconst.i64 0
+  v2 = icmp slt v1, v0
+  brif v2, block1, block2
+block1:
+  v3 = iconst.i8 1
+  return v3
+block2:
+  v4 = iconst.i8 0
+  return v4
+}
+
+; VCode:
+;   pushq   %rbp
+;   movq    %rsp, %rbp
+; block0:
+;   testq   %rdi, %rdi
+;   jnle    label2; j label1
+; block1:
+;   xorl    %eax, %eax, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; block2:
+;   movl    $1, %eax
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; 
+; Disassembled:
+; block0: ; offset 0x0
+;   pushq %rbp
+;   movq %rsp, %rbp
+; block1: ; offset 0x4
+;   testq %rdi, %rdi
+;   jg 0x14
 ; block2: ; offset 0xd
 ;   xorl %eax, %eax
 ;   movq %rbp, %rsp

cranelift/filetests/filetests/isa/x64/load-op.clif

@@ -238,3 +238,33 @@ block0(v0: i64):
 ;   popq %rbp
 ;   retq
 
+function %cmp_mem_zero(i64) -> i8 {
+block0(v0: i64):
+  v1 = load.i64 v0
+  v2 = icmp_imm eq v1, 0
+  return v2
+}
+
+; VCode:
+;   pushq   %rbp
+;   movq    %rsp, %rbp
+; block0:
+;   movq    0(%rdi), %rcx
+;   testq   %rcx, %rcx
+;   setz    %al
+;   movq    %rbp, %rsp
+;   popq    %rbp
+;   ret
+; 
+; Disassembled:
+; block0: ; offset 0x0
+;   pushq %rbp
+;   movq %rsp, %rbp
+; block1: ; offset 0x4
+;   movq (%rdi), %rcx ; trap: heap_oob
+;   testq %rcx, %rcx
+;   sete %al
+;   movq %rbp, %rsp
+;   popq %rbp
+;   retq
+