Skip to content

A lightweight WebAssembly runtime that is fast, secure, and standards-compliant

winch(x64): Prefer operands in the value stack for atomic operations (#14074)

Saúl Cabrera45af25f4984d
diff --git a/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw16_andu.wat b/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw16_andu.wat
index 2216fe92bd..d65aa73d42 100644
--- a/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw16_andu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw16_andu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x8f
+;; ja 0x91
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,24 +21,26 @@
;; movl $0, %ecx
;; andl $1, %ecx
;; cmpl $0, %ecx
-;; jne 0x91
+;; jne 0x93
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movzwq (%rdx), %rax
;; movq %rax, %r11
;; andq %rcx, %r11
;; lock cmpxchgw %r11w, (%rdx)
-;; jne 0x71
-;; 83: movzwl %ax, %eax
+;; jne 0x73
+;; 85: movzwl %ax, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 8f: ud2
;; 91: ud2
+;; 93: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw8_andu.wat b/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw8_andu.wat
index 805b4be84c..2dd2dc71f9 100644
--- a/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw8_andu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw8_andu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x7a
+;; ja 0x7c
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -23,17 +23,19 @@
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movzbq (%rdx), %rax
;; movq %rax, %r11
;; andq %rcx, %r11
;; lock cmpxchgb %r11b, (%rdx)
-;; jne 0x5d
-;; 6e: movzbl %al, %eax
+;; jne 0x5f
+;; 70: movzbl %al, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 7a: ud2
+;; 7c: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw_and.wat b/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw_and.wat
index 9f7d6cac30..e9f4fc5345 100644
--- a/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw_and.wat
+++ b/tests/disas/winch/x64/atomic/rmw/and/i32_atomic_rmw_and.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x89
+;; ja 0x8b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,23 +21,25 @@
;; movl $0, %ecx
;; andl $3, %ecx
;; cmpl $0, %ecx
-;; jne 0x8b
+;; jne 0x8d
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movl (%rdx), %eax
;; movq %rax, %r11
;; andq %rcx, %r11
;; lock cmpxchgl %r11d, (%rdx)
-;; jne 0x6f
-;; 80: addq $0x10, %rsp
+;; jne 0x71
+;; 82: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 89: ud2
;; 8b: ud2
+;; 8d: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw16_andu.wat b/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw16_andu.wat
index 2dd9023fba..75fbaebf1d 100644
--- a/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw16_andu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw16_andu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x7e
+;; ja 0x80
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,22 +21,24 @@
;; movl $0, %ecx
;; andl $1, %ecx
;; cmpl $0, %ecx
-;; jne 0x80
+;; jne 0x82
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movzwq (%rdx), %rax
;; movq %rax, %r11
;; andq %rcx, %r11
;; lock cmpxchgw %r11w, (%rdx)
-;; jne 0x5f
-;; 71: movzwq %ax, %rax
+;; jne 0x61
+;; 73: movzwq %ax, %rax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 7e: ud2
;; 80: ud2
+;; 82: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw32_andu.wat b/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw32_andu.wat
index ebeac1c6e0..fd055c58f7 100644
--- a/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw32_andu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw32_andu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x77
+;; ja 0x79
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,21 +21,23 @@
;; movl $0, %ecx
;; andl $3, %ecx
;; cmpl $0, %ecx
-;; jne 0x79
+;; jne 0x7b
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movl (%rdx), %eax
;; movq %rax, %r11
;; andq %rcx, %r11
;; lock cmpxchgl %r11d, (%rdx)
-;; jne 0x5d
-;; 6e: addq $0x10, %rsp
+;; jne 0x5f
+;; 70: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 77: ud2
;; 79: ud2
+;; 7b: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw8_andu.wat b/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw8_andu.wat
index 0a6d9e215c..d691dcc19f 100644
--- a/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw8_andu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw8_andu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x69
+;; ja 0x6b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -23,15 +23,17 @@
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movzbq (%rdx), %rax
;; movq %rax, %r11
;; andq %rcx, %r11
;; lock cmpxchgb %r11b, (%rdx)
-;; jne 0x4b
-;; 5c: movzbq %al, %rax
+;; jne 0x4d
+;; 5e: movzbq %al, %rax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 69: ud2
+;; 6b: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw_and.wat b/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw_and.wat
index cbb085743c..9e161c86fa 100644
--- a/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw_and.wat
+++ b/tests/disas/winch/x64/atomic/rmw/and/i64_atomic_rmw_and.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x78
+;; ja 0x7a
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,21 +21,23 @@
;; movl $0, %ecx
;; andl $7, %ecx
;; cmpl $0, %ecx
-;; jne 0x7a
+;; jne 0x7c
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movq (%rdx), %rax
;; movq %rax, %r11
;; andq %rcx, %r11
;; lock cmpxchgq %r11, (%rdx)
-;; jne 0x5e
-;; 6f: addq $0x10, %rsp
+;; jne 0x60
+;; 71: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 78: ud2
;; 7a: ud2
+;; 7c: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw16_cmpxchgu.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw16_cmpxchgu.wat
index bc8c92da94..cbedca161e 100644
--- a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw16_cmpxchgu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw16_cmpxchgu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x97
+;; ja 0x99
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -22,12 +22,13 @@
;; movl $0, %edx
;; andl $1, %edx
;; cmpl $0, %edx
-;; jne 0x99
+;; jne 0x9b
;; 4d: movl $0, %edx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rbx
;; movl %edx, %edx
;; addq %rdx, %rbx
+;; pushq %rbx
;; subq $4, %rsp
;; movl %ecx, (%rsp)
;; subq $4, %rsp
@@ -36,10 +37,11 @@
;; addq $4, %rsp
;; movl (%rsp), %eax
;; addq $4, %rsp
-;; lock cmpxchgw %cx, (%rbx)
+;; popq %rdx
+;; lock cmpxchgw %cx, (%rdx)
;; movzwl %ax, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 97: ud2
;; 99: ud2
+;; 9b: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw8_cmpxchgu.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw8_cmpxchgu.wat
index 4bd5ff094b..50bf28abe0 100644
--- a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw8_cmpxchgu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw8_cmpxchgu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x82
+;; ja 0x84
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -24,6 +24,7 @@
;; movq (%r11), %rbx
;; movl %edx, %edx
;; addq %rdx, %rbx
+;; pushq %rbx
;; subq $4, %rsp
;; movl %ecx, (%rsp)
;; subq $4, %rsp
@@ -32,9 +33,10 @@
;; addq $4, %rsp
;; movl (%rsp), %eax
;; addq $4, %rsp
-;; lock cmpxchgb %cl, (%rbx)
+;; popq %rdx
+;; lock cmpxchgb %cl, (%rdx)
;; movzbl %al, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 82: ud2
+;; 84: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw_cmpxchg.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw_cmpxchg.wat
index d2e923d3fb..4e73ac0c23 100644
--- a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw_cmpxchg.wat
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw_cmpxchg.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x93
+;; ja 0x95
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -22,12 +22,13 @@
;; movl $0, %edx
;; andl $3, %edx
;; cmpl $0, %edx
-;; jne 0x95
+;; jne 0x97
;; 4d: movl $0, %edx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rbx
;; movl %edx, %edx
;; addq %rdx, %rbx
+;; pushq %rbx
;; subq $4, %rsp
;; movl %ecx, (%rsp)
;; subq $4, %rsp
@@ -36,9 +37,10 @@
;; addq $4, %rsp
;; movl (%rsp), %eax
;; addq $4, %rsp
-;; lock cmpxchgl %ecx, (%rbx)
+;; popq %rdx
+;; lock cmpxchgl %ecx, (%rdx)
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 93: ud2
;; 95: ud2
+;; 97: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw_cmpxchg_computed_address.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw_cmpxchg_computed_address.wat
new file mode 100644
index 0000000000..f723736959
--- /dev/null
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i32_atomic_rmw_cmpxchg_computed_address.wat
@@ -0,0 +1,55 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module
+ (memory 1)
+ (func (export "_start") (param i32) (result i32)
+ (i32.atomic.rmw.cmpxchg
+ (i32.add (i32.const 1) (local.get 0))
+ (i32.xor (i32.const 1) (local.get 0))
+ (i32.xor (i32.const 2) (local.get 0)))))
+;; wasm[0]::function[0]:
+;; pushq %rbp
+;; movq %rsp, %rbp
+;; movq 8(%rdi), %r11
+;; movq 0x18(%r11), %r11
+;; addq $0x30, %r11
+;; cmpq %rsp, %r11
+;; ja 0xa5
+;; 1c: movq %rdi, %r14
+;; subq $0x20, %rsp
+;; movq %rdi, 0x18(%rsp)
+;; movq %rsi, 0x10(%rsp)
+;; movl %edx, 0xc(%rsp)
+;; movl 0xc(%rsp), %eax
+;; movl $1, %ecx
+;; addl %eax, %ecx
+;; movl 0xc(%rsp), %eax
+;; movl $1, %edx
+;; xorl %eax, %edx
+;; movl 0xc(%rsp), %eax
+;; movl $2, %ebx
+;; xorl %eax, %ebx
+;; movl %ecx, %eax
+;; andl $3, %eax
+;; cmpl $0, %eax
+;; jne 0xa7
+;; 65: movq 0x38(%r14), %rax
+;; movl %ecx, %ecx
+;; addq %rcx, %rax
+;; pushq %rax
+;; subq $4, %rsp
+;; movl %edx, (%rsp)
+;; subq $4, %rsp
+;; movl %ebx, (%rsp)
+;; movl (%rsp), %ecx
+;; addq $4, %rsp
+;; movl (%rsp), %eax
+;; addq $4, %rsp
+;; popq %rdx
+;; lock cmpxchgl %ecx, (%rdx)
+;; addq $0x20, %rsp
+;; popq %rbp
+;; retq
+;; a5: ud2
+;; a7: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw16_cmpxchgu.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw16_cmpxchgu.wat
index c0cb871d21..42143d7422 100644
--- a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw16_cmpxchgu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw16_cmpxchgu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x20, %r11
+;; addq $0x28, %r11
;; cmpq %rsp, %r11
-;; ja 0x74
+;; ja 0x76
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -22,20 +22,22 @@
;; movl $0, %edx
;; andl $1, %edx
;; cmpl $0, %edx
-;; jne 0x76
+;; jne 0x78
;; 4d: movl $0, %edx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rbx
;; movl %edx, %edx
;; addq %rdx, %rbx
+;; pushq %rbx
;; pushq %rcx
;; pushq %rax
;; popq %rcx
;; popq %rax
-;; lock cmpxchgw %cx, (%rbx)
+;; popq %rdx
+;; lock cmpxchgw %cx, (%rdx)
;; movzwq %ax, %rax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 74: ud2
;; 76: ud2
+;; 78: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw32_cmpxchgu.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw32_cmpxchgu.wat
index a6c3a954e7..2987acc19c 100644
--- a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw32_cmpxchgu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw32_cmpxchgu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x20, %r11
+;; addq $0x28, %r11
;; cmpq %rsp, %r11
-;; ja 0x71
+;; ja 0x73
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -22,20 +22,22 @@
;; movl $0, %edx
;; andl $3, %edx
;; cmpl $0, %edx
-;; jne 0x73
+;; jne 0x75
;; 4d: movl $0, %edx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rbx
;; movl %edx, %edx
;; addq %rdx, %rbx
+;; pushq %rbx
;; pushq %rcx
;; pushq %rax
;; popq %rcx
;; popq %rax
-;; lock cmpxchgl %ecx, (%rbx)
+;; popq %rdx
+;; lock cmpxchgl %ecx, (%rdx)
;; movl %eax, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 71: ud2
;; 73: ud2
+;; 75: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw32_cmpxchgu_extend.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw32_cmpxchgu_extend.wat
index 1221f51cda..67fd57c725 100644
--- a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw32_cmpxchgu_extend.wat
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw32_cmpxchgu_extend.wat
@@ -13,9 +13,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x20, %r11
+;; addq $0x28, %r11
;; cmpq %rsp, %r11
-;; ja 0x76
+;; ja 0x78
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -25,20 +25,22 @@
;; movl $0, %edx
;; andl $3, %edx
;; cmpl $0, %edx
-;; jne 0x78
+;; jne 0x7a
;; 52: movl $0, %edx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rbx
;; movl %edx, %edx
;; addq %rdx, %rbx
+;; pushq %rbx
;; pushq %rcx
;; pushq %rax
;; popq %rcx
;; popq %rax
-;; lock cmpxchgl %ecx, (%rbx)
+;; popq %rdx
+;; lock cmpxchgl %ecx, (%rdx)
;; movl %eax, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 76: ud2
;; 78: ud2
+;; 7a: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw8_cmpxchgu.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw8_cmpxchgu.wat
index 96898ac61b..0f54d818d0 100644
--- a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw8_cmpxchgu.wat
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw8_cmpxchgu.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x20, %r11
+;; addq $0x28, %r11
;; cmpq %rsp, %r11
-;; ja 0x5f
+;; ja 0x61
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -24,13 +24,15 @@
;; movq (%r11), %rbx
;; movl %edx, %edx
;; addq %rdx, %rbx
+;; pushq %rbx
;; pushq %rcx
;; pushq %rax
;; popq %rcx
;; popq %rax
-;; lock cmpxchgb %cl, (%rbx)
+;; popq %rdx
+;; lock cmpxchgb %cl, (%rdx)
;; movzbq %al, %rax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 5f: ud2
+;; 61: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw_cmpxchg.wat b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw_cmpxchg.wat
index a1a39590ba..c40af6d312 100644
--- a/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw_cmpxchg.wat
+++ b/tests/disas/winch/x64/atomic/rmw/cmpxchg/i64_atomic_rmw_cmpxchg.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x20, %r11
+;; addq $0x28, %r11
;; cmpq %rsp, %r11
-;; ja 0x70
+;; ja 0x72
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -22,19 +22,21 @@
;; movl $0, %edx
;; andl $7, %edx
;; cmpl $0, %edx
-;; jne 0x72
+;; jne 0x74
;; 4d: movl $0, %edx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rbx
;; movl %edx, %edx
;; addq %rdx, %rbx
+;; pushq %rbx
;; pushq %rcx
;; pushq %rax
;; popq %rcx
;; popq %rax
-;; lock cmpxchgq %rcx, (%rbx)
+;; popq %rdx
+;; lock cmpxchgq %rcx, (%rdx)
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 70: ud2
;; 72: ud2
+;; 74: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw16_oru.wat b/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw16_oru.wat
index 329eb5169e..3b0759edf8 100644
--- a/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw16_oru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw16_oru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x8f
+;; ja 0x91
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,24 +21,26 @@
;; movl $0, %ecx
;; andl $1, %ecx
;; cmpl $0, %ecx
-;; jne 0x91
+;; jne 0x93
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movzwq (%rdx), %rax
;; movq %rax, %r11
;; orq %rcx, %r11
;; lock cmpxchgw %r11w, (%rdx)
-;; jne 0x71
-;; 83: movzwl %ax, %eax
+;; jne 0x73
+;; 85: movzwl %ax, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 8f: ud2
;; 91: ud2
+;; 93: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw8_oru.wat b/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw8_oru.wat
index 59b2b5d1bd..60be2a022f 100644
--- a/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw8_oru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw8_oru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x7a
+;; ja 0x7c
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -23,17 +23,19 @@
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movzbq (%rdx), %rax
;; movq %rax, %r11
;; orq %rcx, %r11
;; lock cmpxchgb %r11b, (%rdx)
-;; jne 0x5d
-;; 6e: movzbl %al, %eax
+;; jne 0x5f
+;; 70: movzbl %al, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 7a: ud2
+;; 7c: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw_or.wat b/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw_or.wat
index 1472e2d27f..6818994f45 100644
--- a/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw_or.wat
+++ b/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw_or.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x89
+;; ja 0x8b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,23 +21,25 @@
;; movl $0, %ecx
;; andl $3, %ecx
;; cmpl $0, %ecx
-;; jne 0x8b
+;; jne 0x8d
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movl (%rdx), %eax
;; movq %rax, %r11
;; orq %rcx, %r11
;; lock cmpxchgl %r11d, (%rdx)
-;; jne 0x6f
-;; 80: addq $0x10, %rsp
+;; jne 0x71
+;; 82: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 89: ud2
;; 8b: ud2
+;; 8d: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw_or_computed_address.wat b/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw_or_computed_address.wat
new file mode 100644
index 0000000000..61111d8968
--- /dev/null
+++ b/tests/disas/winch/x64/atomic/rmw/or/i32_atomic_rmw_or_computed_address.wat
@@ -0,0 +1,51 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module
+ (memory 1)
+ (func (export "_start") (param i32) (result i32)
+ (i32.atomic.rmw.or
+ (i32.add (i32.const 1) (local.get 0))
+ (i32.xor (i32.const 1) (local.get 0)))))
+;; wasm[0]::function[0]:
+;; pushq %rbp
+;; movq %rsp, %rbp
+;; movq 8(%rdi), %r11
+;; movq 0x18(%r11), %r11
+;; addq $0x2c, %r11
+;; cmpq %rsp, %r11
+;; ja 0x95
+;; 1c: movq %rdi, %r14
+;; subq $0x20, %rsp
+;; movq %rdi, 0x18(%rsp)
+;; movq %rsi, 0x10(%rsp)
+;; movl %edx, 0xc(%rsp)
+;; movl 0xc(%rsp), %eax
+;; movl $1, %ecx
+;; addl %eax, %ecx
+;; movl 0xc(%rsp), %eax
+;; movl $1, %edx
+;; xorl %eax, %edx
+;; movl %ecx, %eax
+;; andl $3, %eax
+;; cmpl $0, %eax
+;; jne 0x97
+;; 5a: movq 0x38(%r14), %rax
+;; movl %ecx, %ecx
+;; addq %rcx, %rax
+;; pushq %rax
+;; subq $4, %rsp
+;; movl %edx, (%rsp)
+;; movl (%rsp), %ecx
+;; addq $4, %rsp
+;; popq %rdx
+;; movl (%rdx), %eax
+;; movq %rax, %r11
+;; orq %rcx, %r11
+;; lock cmpxchgl %r11d, (%rdx)
+;; jne 0x7b
+;; 8c: addq $0x20, %rsp
+;; popq %rbp
+;; retq
+;; 95: ud2
+;; 97: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw16_oru.wat b/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw16_oru.wat
index cb5c1ee3d6..29acce6c47 100644
--- a/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw16_oru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw16_oru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x7e
+;; ja 0x80
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,22 +21,24 @@
;; movl $0, %ecx
;; andl $1, %ecx
;; cmpl $0, %ecx
-;; jne 0x80
+;; jne 0x82
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movzwq (%rdx), %rax
;; movq %rax, %r11
;; orq %rcx, %r11
;; lock cmpxchgw %r11w, (%rdx)
-;; jne 0x5f
-;; 71: movzwq %ax, %rax
+;; jne 0x61
+;; 73: movzwq %ax, %rax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 7e: ud2
;; 80: ud2
+;; 82: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw32_oru.wat b/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw32_oru.wat
index 56f38739df..101c5ddc27 100644
--- a/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw32_oru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw32_oru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x77
+;; ja 0x79
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,21 +21,23 @@
;; movl $0, %ecx
;; andl $3, %ecx
;; cmpl $0, %ecx
-;; jne 0x79
+;; jne 0x7b
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movl (%rdx), %eax
;; movq %rax, %r11
;; orq %rcx, %r11
;; lock cmpxchgl %r11d, (%rdx)
-;; jne 0x5d
-;; 6e: addq $0x10, %rsp
+;; jne 0x5f
+;; 70: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 77: ud2
;; 79: ud2
+;; 7b: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw8_oru.wat b/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw8_oru.wat
index 78a1f3acf9..6ba5961acb 100644
--- a/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw8_oru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw8_oru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x69
+;; ja 0x6b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -23,15 +23,17 @@
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movzbq (%rdx), %rax
;; movq %rax, %r11
;; orq %rcx, %r11
;; lock cmpxchgb %r11b, (%rdx)
-;; jne 0x4b
-;; 5c: movzbq %al, %rax
+;; jne 0x4d
+;; 5e: movzbq %al, %rax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 69: ud2
+;; 6b: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw_or.wat b/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw_or.wat
index c68efac930..bb0422157f 100644
--- a/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw_or.wat
+++ b/tests/disas/winch/x64/atomic/rmw/or/i64_atomic_rmw_or.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x78
+;; ja 0x7a
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,21 +21,23 @@
;; movl $0, %ecx
;; andl $7, %ecx
;; cmpl $0, %ecx
-;; jne 0x7a
+;; jne 0x7c
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movq (%rdx), %rax
;; movq %rax, %r11
;; orq %rcx, %r11
;; lock cmpxchgq %r11, (%rdx)
-;; jne 0x5e
-;; 6f: addq $0x10, %rsp
+;; jne 0x60
+;; 71: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 78: ud2
;; 7a: ud2
+;; 7c: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw16_xoru.wat b/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw16_xoru.wat
index 2dd77dbff6..5da0033754 100644
--- a/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw16_xoru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw16_xoru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x8f
+;; ja 0x91
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,24 +21,26 @@
;; movl $0, %ecx
;; andl $1, %ecx
;; cmpl $0, %ecx
-;; jne 0x91
+;; jne 0x93
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movzwq (%rdx), %rax
;; movq %rax, %r11
;; xorq %rcx, %r11
;; lock cmpxchgw %r11w, (%rdx)
-;; jne 0x71
-;; 83: movzwl %ax, %eax
+;; jne 0x73
+;; 85: movzwl %ax, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 8f: ud2
;; 91: ud2
+;; 93: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw8_xoru.wat b/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw8_xoru.wat
index 5c9253842b..95428ae0d8 100644
--- a/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw8_xoru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw8_xoru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x7a
+;; ja 0x7c
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -23,17 +23,19 @@
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movzbq (%rdx), %rax
;; movq %rax, %r11
;; xorq %rcx, %r11
;; lock cmpxchgb %r11b, (%rdx)
-;; jne 0x5d
-;; 6e: movzbl %al, %eax
+;; jne 0x5f
+;; 70: movzbl %al, %eax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 7a: ud2
+;; 7c: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw_xor.wat b/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw_xor.wat
index cf0e18bd09..a7cf0a6fd4 100644
--- a/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw_xor.wat
+++ b/tests/disas/winch/x64/atomic/rmw/xor/i32_atomic_rmw_xor.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x14, %r11
+;; addq $0x1c, %r11
;; cmpq %rsp, %r11
-;; ja 0x89
+;; ja 0x8b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,23 +21,25 @@
;; movl $0, %ecx
;; andl $3, %ecx
;; cmpl $0, %ecx
-;; jne 0x8b
+;; jne 0x8d
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; subq $4, %rsp
;; movl %eax, (%rsp)
;; movl (%rsp), %ecx
;; addq $4, %rsp
+;; popq %rdx
;; movl (%rdx), %eax
;; movq %rax, %r11
;; xorq %rcx, %r11
;; lock cmpxchgl %r11d, (%rdx)
-;; jne 0x6f
-;; 80: addq $0x10, %rsp
+;; jne 0x71
+;; 82: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 89: ud2
;; 8b: ud2
+;; 8d: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw16_xoru.wat b/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw16_xoru.wat
index d557a144e3..88bda7282d 100644
--- a/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw16_xoru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw16_xoru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x7e
+;; ja 0x80
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,22 +21,24 @@
;; movl $0, %ecx
;; andl $1, %ecx
;; cmpl $0, %ecx
-;; jne 0x80
+;; jne 0x82
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movzwq (%rdx), %rax
;; movq %rax, %r11
;; xorq %rcx, %r11
;; lock cmpxchgw %r11w, (%rdx)
-;; jne 0x5f
-;; 71: movzwq %ax, %rax
+;; jne 0x61
+;; 73: movzwq %ax, %rax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 7e: ud2
;; 80: ud2
+;; 82: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw32_xoru.wat b/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw32_xoru.wat
index e86fb4dac9..bfb6ff6cc6 100644
--- a/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw32_xoru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw32_xoru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x77
+;; ja 0x79
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,21 +21,23 @@
;; movl $0, %ecx
;; andl $3, %ecx
;; cmpl $0, %ecx
-;; jne 0x79
+;; jne 0x7b
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movl (%rdx), %eax
;; movq %rax, %r11
;; xorq %rcx, %r11
;; lock cmpxchgl %r11d, (%rdx)
-;; jne 0x5d
-;; 6e: addq $0x10, %rsp
+;; jne 0x5f
+;; 70: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 77: ud2
;; 79: ud2
+;; 7b: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw8_xoru.wat b/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw8_xoru.wat
index fc8fdb115c..0656839849 100644
--- a/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw8_xoru.wat
+++ b/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw8_xoru.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x69
+;; ja 0x6b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -23,15 +23,17 @@
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movzbq (%rdx), %rax
;; movq %rax, %r11
;; xorq %rcx, %r11
;; lock cmpxchgb %r11b, (%rdx)
-;; jne 0x4b
-;; 5c: movzbq %al, %rax
+;; jne 0x4d
+;; 5e: movzbq %al, %rax
;; addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 69: ud2
+;; 6b: ud2
diff --git a/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw_xor.wat b/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw_xor.wat
index 5125257018..197130fc5e 100644
--- a/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw_xor.wat
+++ b/tests/disas/winch/x64/atomic/rmw/xor/i64_atomic_rmw_xor.wat
@@ -10,9 +10,9 @@
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x18(%r11), %r11
-;; addq $0x18, %r11
+;; addq $0x20, %r11
;; cmpq %rsp, %r11
-;; ja 0x78
+;; ja 0x7a
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
@@ -21,21 +21,23 @@
;; movl $0, %ecx
;; andl $7, %ecx
;; cmpl $0, %ecx
-;; jne 0x7a
+;; jne 0x7c
;; 48: movl $0, %ecx
;; movq 0x30(%r14), %r11
;; movq (%r11), %rdx
;; movl %ecx, %ecx
;; addq %rcx, %rdx
+;; pushq %rdx
;; pushq %rax
;; popq %rcx
+;; popq %rdx
;; movq (%rdx), %rax
;; movq %rax, %r11
;; xorq %rcx, %r11
;; lock cmpxchgq %r11, (%rdx)
-;; jne 0x5e
-;; 6f: addq $0x10, %rsp
+;; jne 0x60
+;; 71: addq $0x10, %rsp
;; popq %rbp
;; retq
-;; 78: ud2
;; 7a: ud2
+;; 7c: ud2
diff --git a/winch/codegen/src/codegen/mod.rs b/winch/codegen/src/codegen/mod.rs
index 34d7d1e90e..ac0fd0b1c1 100644
--- a/winch/codegen/src/codegen/mod.rs
+++ b/winch/codegen/src/codegen/mod.rs
@@ -2337,14 +2337,24 @@ where
let heap = self.env.resolve_heap(memory_index);
// We need to pop-push the operand to compute the address before passing control over to
// masm, because some architectures may have specific requirements for the registers used
- // in some atomic operations.
+ // in some atomic operations. The computed address is pushed back to the context's stack
+ // too, rather than handed over as a register, since registers that are not tracked by the
+ // value stack can't be spilled, so an untracked address register would make any request
+ // for a fixed register fail if the address happened to be allocated to it. For this
+ // reason, the address is pushed as a register to be dereferenced prior to emission, after
+ // all the ISA-specifc constraints have been solved.
let operand = self.context.pop_to_reg(self.masm, None)?;
if let Some(addr) = self.emit_compute_heap_address_align_checked(&heap, arg, size)? {
- let src = self.masm.address_at_reg(addr, 0)?;
+ self.context
+ .stack
+ .push(TypedReg::new(self.env.ptr_type(), addr).into());
self.context.stack.push(operand.into());
self.masm
- .atomic_rmw(&mut self.context, src, size, op, UNTRUSTED_FLAGS, extend)?;
- self.context.free_reg(addr);
+ .atomic_rmw(&mut self.context, size, op, UNTRUSTED_FLAGS, extend)?;
+ } else {
+ // Ensure that the operand register is not left allocated if the access was proven to
+ // be out of bounds at compile time.
+ self.context.free_reg(operand);
}
Ok(())
@@ -2368,10 +2378,15 @@ where
// with regard to the registers used for some arguments, so we
// need to pass the context to the masm. To solve this issue,
// we pop the two first arguments from the stack, compute the
- // address, push back the arguments, and hand over the control
- // to masm. The implementer of `atomic_cas` can expect to find
- // `expected` and `replacement` at the top the context's
- // stack.
+ // address, push back the address and the arguments, and hand
+ // over the control to masm. The implementer of `atomic_cas`
+ // can expect to find `address`, `expected` and `replacement`
+ // at the top the context's stack.
+ //
+ // The computed address is pushed back to the stack as a
+ // register, rather than handed over directly, so that the
+ // register allocator is able to spill it if the target
+ // requires a fixed register.
let replacement = self.context.pop_to_reg(self.masm, None)?;
let expected = self.context.pop_to_reg(self.masm, None)?;
@@ -2379,14 +2394,19 @@ where
let memory_index = MemoryIndex::from_u32(arg.memory);
let heap = self.env.resolve_heap(memory_index);
if let Some(addr) = self.emit_compute_heap_address_align_checked(&heap, arg, size)? {
+ self.context
+ .stack
+ .push(TypedReg::new(self.env.ptr_type(), addr).into());
self.context.stack.push(expected.into());
self.context.stack.push(replacement.into());
- let src = self.masm.address_at_reg(addr, 0)?;
self.masm
- .atomic_cas(&mut self.context, src, size, UNTRUSTED_FLAGS, extend)?;
-
- self.context.free_reg(addr);
+ .atomic_cas(&mut self.context, size, UNTRUSTED_FLAGS, extend)?;
+ } else {
+ // Ensure that the argument registers are not left allocated if the access was proven
+ // to be out of bounds at compile time.
+ self.context.free_reg(expected);
+ self.context.free_reg(replacement);
}
Ok(())
}
diff --git a/winch/codegen/src/isa/aarch64/masm.rs b/winch/codegen/src/isa/aarch64/masm.rs
index c5c882216c..9ac7842106 100644
--- a/winch/codegen/src/isa/aarch64/masm.rs
+++ b/winch/codegen/src/isa/aarch64/masm.rs
@@ -1471,7 +1471,6 @@ impl Masm for MacroAssembler {
fn atomic_rmw(
&mut self,
_context: &mut CodeGenContext<Emission>,
- _addr: Self::Address,
_size: OperandSize,
_op: RmwOp,
_flags: MemFlagsData,
@@ -1557,7 +1556,6 @@ impl Masm for MacroAssembler {
fn atomic_cas(
&mut self,
_context: &mut CodeGenContext<Emission>,
- _addr: Self::Address,
_size: OperandSize,
_flags: MemFlagsData,
_extend: Option<Extend<Zero>>,
diff --git a/winch/codegen/src/isa/x64/masm.rs b/winch/codegen/src/isa/x64/masm.rs
index c29ddb9541..0044af0222 100644
--- a/winch/codegen/src/isa/x64/masm.rs
+++ b/winch/codegen/src/isa/x64/masm.rs
@@ -1604,29 +1604,31 @@ impl Masm for MacroAssembler {
fn atomic_rmw(
&mut self,
context: &mut CodeGenContext<Emission>,
- addr: Self::Address,
size: OperandSize,
op: RmwOp,
flags: MemFlagsData,
extend: Option<Extend<Zero>>,
) -> Result<()> {
let res = match op {
- RmwOp::Add => {
+ RmwOp::Add | RmwOp::Sub | RmwOp::Xchg => {
let operand = context.pop_to_reg(self, None)?;
- self.asm
- .lock_xadd(addr, writable!(operand.reg), size, flags);
- operand.reg
- }
- RmwOp::Sub => {
- let operand = context.pop_to_reg(self, None)?;
- self.asm.neg(operand.reg, writable!(operand.reg), size);
- self.asm
- .lock_xadd(addr, writable!(operand.reg), size, flags);
- operand.reg
- }
- RmwOp::Xchg => {
- let operand = context.pop_to_reg(self, None)?;
- self.asm.xchg(addr, writable!(operand.reg), size, flags);
+ let base = context.pop_to_reg(self, None)?;
+ let addr = self.address_at_reg(base.reg, 0)?;
+
+ match op {
+ RmwOp::Add => self
+ .asm
+ .lock_xadd(addr, writable!(operand.reg), size, flags),
+ RmwOp::Sub => {
+ self.asm.neg(operand.reg, writable!(operand.reg), size);
+ self.asm
+ .lock_xadd(addr, writable!(operand.reg), size, flags);
+ }
+ RmwOp::Xchg => self.asm.xchg(addr, writable!(operand.reg), size, flags),
+ _ => unreachable!(),
+ }
+
+ context.free_reg(base.reg);
operand.reg
}
RmwOp::And | RmwOp::Or | RmwOp::Xor => {
@@ -1638,8 +1640,11 @@ impl Masm for MacroAssembler {
"invalid op for atomic_rmw_seq, should be one of `or`, `and` or `xor`"
),
};
+
let dst = context.reg(regs::rax(), self)?;
let operand = context.pop_to_reg(self, None)?;
+ let base = context.pop_to_reg(self, None)?;
+ let addr = self.address_at_reg(base.reg, 0)?;
self.with_scratch::<IntScratch, _>(|masm, scratch| {
masm.asm.atomic_rmw_seq(
@@ -1654,6 +1659,7 @@ impl Masm for MacroAssembler {
});
context.free_reg(operand.reg);
+ context.free_reg(base.reg);
dst
}
};
@@ -1791,20 +1797,18 @@ impl Masm for MacroAssembler {
fn atomic_cas(
&mut self,
context: &mut CodeGenContext<Emission>,
- addr: Self::Address,
size: OperandSize,
flags: MemFlagsData,
extend: Option<Extend<Zero>>,
) -> Result<()> {
- // `cmpxchg` expects `expected` to be in the `*a*` register.
- // reserve rax for the expected argument.
-
let replacement =
context.without::<Result<TypedReg>, _, _>(&[regs::rax()], self, |cx, masm| {
cx.pop_to_reg(masm, None)
})??;
let expected = context.pop_to_reg(self, Some(regs::rax()))?;
+ let base = context.pop_to_reg(self, None)?;
+ let addr = self.address_at_reg(base.reg, 0)?;
self.asm
.cmpxchg(addr, replacement.reg, writable!(expected.reg), size, flags);
@@ -1816,6 +1820,7 @@ impl Masm for MacroAssembler {
context.stack.push(expected.into());
context.free_reg(replacement);
+ context.free_reg(base.reg);
Ok(())
}
diff --git a/winch/codegen/src/masm.rs b/winch/codegen/src/masm.rs
index 9483ccb0f8..62c8318c15 100644
--- a/winch/codegen/src/masm.rs
+++ b/winch/codegen/src/masm.rs
@@ -2074,13 +2074,22 @@ pub(crate) trait MacroAssembler {
/// Performs a swizzle between two 128-bit vectors into a 128-bit result.
fn swizzle(&mut self, dst: WritableReg, lhs: Reg, rhs: Reg) -> Result<()>;
- /// Performs the RMW `op` operation on the passed `addr`.
+ /// Performs the RMW `op` operation on the address at the top of the
+ /// context's stack.
///
- /// The value *before* the operation was performed is written back to the `operand` register.
+ /// This method takes the `CodeGenContext` as an argument to accommodate
+ /// architectures that expect parameters in specific registers. The context
+ /// stack contains the `address` and the `operand` values, in that order,
+ /// and both are owned by this function. The implementer is expected to
+ /// push the value *before* the operation was performed to the context's
+ /// stack before returning.
+ ///
+ /// Note that the address is passed through the context's stack rather than
+ /// as a register to ensure that any spills can be perfomed when solving
+ /// ISA-specific constraints prior to emission.
fn atomic_rmw(
&mut self,
context: &mut CodeGenContext<Emission>,
- addr: Self::Address,
size: OperandSize,
op: RmwOp,
flags: MemFlagsData,
@@ -2105,17 +2114,21 @@ pub(crate) trait MacroAssembler {
kind: ReplaceLaneKind,
) -> Result<()>;
- /// Perform an atomic CAS (compare-and-swap) operation with the value at `addr`, and `expected`
- /// and `replacement` (at the top of the context's stack).
+ /// Perform an atomic CAS (compare-and-swap) operation with the `address`, `expected` and
+ /// `replacement` values at the top of the context's stack.
///
/// This method takes the `CodeGenContext` as an arguments to accommodate architectures that
- /// expect parameters in specific registers. The context stack contains the `replacement`,
- /// and `expected` values in that order. The implementer is expected to push the value at
- /// `addr` before the update to the context's stack before returning.
+ /// expect parameters in specific registers. The context stack contains the `address`,
+ /// `expected` and `replacement` values in that order, and all of them are owned by this
+ /// function. The implementer is expected to push the value at `address` before the update to
+ /// the context's stack before returning.
+ ///
+ /// Like in [`MacroAssembler::atomic_rmw`], the address is passed through the context's stack
+ /// so that it can be spilled; implementations that require fixed registers must request them
+ /// *before* popping any of the values above.
fn atomic_cas(
&mut self,
context: &mut CodeGenContext<Emission>,
- addr: Self::Address,
size: OperandSize,
flags: MemFlagsData,
extend: Option<Extend<Zero>>,