winch: x64 atomic stores (#9987)

* add MemOpKind to wasm_store * add fence to x64 asm * implement atomic store for x64 masm * emit error for 128bits atomic store * propagate MemOpKind from Visitor * implement atomic store visitor methods * add atomic store tests * fmt * whitelist spec tests * atomic heap address compute align check * fmt * edit comment * cleanup check align function * add doc for check_align * review edits * review edits - use scratch register for tmp in emit_align_check - pop-push value from stack rather than peeking in emit_align_check * final edits
bytecodealliance · Jan 15, 2025 · 3a4cf0a · 3a4cf0a
1 parent 7f9049b
commit 3a4cf0a
Show file tree

Hide file tree

Showing 19 changed files with 442 additions and 51 deletions.
diff --git a/crates/wast-util/src/lib.rs b/crates/wast-util/src/lib.rs
@@ -488,15 +488,9 @@ impl WastTest {
                 "spec_testsuite/simd_store8_lane.wast",
                 // thread related failures
                 "proposals/threads/atomic.wast",
-                "misc_testsuite/threads/MP_wait.wast",
-                "misc_testsuite/threads/load-store-alignment.wast",
-                "misc_testsuite/threads/MP_atomic.wast",
-                "misc_testsuite/threads/SB_atomic.wast",
                 "misc_testsuite/threads/wait_notify.wast",
-                "misc_testsuite/threads/LB_atomic.wast",
                 "misc_testsuite/threads/atomics_wait_address.wast",
                 "misc_testsuite/threads/atomics_notify.wast",
-                "misc_testsuite/threads/load-store-alignment.wast",
             ];
 
             if unsupported.iter().any(|part| self.path.ends_with(part)) {

diff --git a/tests/disas/winch/x64/atomic/load/i32_atomic_load.wat b/tests/disas/winch/x64/atomic/load/i32_atomic_load.wat
@@ -13,18 +13,23 @@
 ;;       movq    0x10(%r11), %r11
 ;;       addq    $0x20, %r11
 ;;       cmpq    %rsp, %r11
-;;       ja      0x47
+;;       ja      0x57
 ;;   1c: movq    %rdi, %r14
 ;;       subq    $0x20, %rsp
 ;;       movq    %rdi, 0x18(%rsp)
 ;;       movq    %rsi, 0x10(%rsp)
 ;;       movl    %edx, 0xc(%rsp)
 ;;       movl    0xc(%rsp), %eax
+;;       andl    $3, %eax
+;;       cmpl    $0, %eax
+;;       jne     0x59
+;;   41: movl    0xc(%rsp), %eax
 ;;       movq    0x58(%r14), %r11
 ;;       movq    (%r11), %rcx
 ;;       addq    %rax, %rcx
 ;;       movl    (%rcx), %eax
 ;;       addq    $0x20, %rsp
 ;;       popq    %rbp
 ;;       retq
-;;   47: ud2
+;;   57: ud2
+;;   59: ud2
diff --git a/tests/disas/winch/x64/atomic/load/i32_atomic_load16_u.wat b/tests/disas/winch/x64/atomic/load/i32_atomic_load16_u.wat
@@ -13,18 +13,23 @@
 ;;       movq    0x10(%r11), %r11
 ;;       addq    $0x20, %r11
 ;;       cmpq    %rsp, %r11
-;;       ja      0x49
+;;       ja      0x5b
 ;;   1c: movq    %rdi, %r14
 ;;       subq    $0x20, %rsp
 ;;       movq    %rdi, 0x18(%rsp)
 ;;       movq    %rsi, 0x10(%rsp)
 ;;       movl    %edx, 0xc(%rsp)
 ;;       movl    0xc(%rsp), %eax
+;;       andw    $1, %ax
+;;       cmpw    $0, %ax
+;;       jne     0x5d
+;;   43: movl    0xc(%rsp), %eax
 ;;       movq    0x58(%r14), %r11
 ;;       movq    (%r11), %rcx
 ;;       addq    %rax, %rcx
 ;;       movzwq  (%rcx), %rax
 ;;       addq    $0x20, %rsp
 ;;       popq    %rbp
 ;;       retq
-;;   49: ud2
+;;   5b: ud2
+;;   5d: ud2
diff --git a/tests/disas/winch/x64/atomic/load/i64_atomic_load.wat b/tests/disas/winch/x64/atomic/load/i64_atomic_load.wat
@@ -14,16 +14,21 @@
 ;;       movq    0x10(%r11), %r11
 ;;       addq    $0x10, %r11
 ;;       cmpq    %rsp, %r11
-;;       ja      0x41
+;;       ja      0x54
 ;;   1c: movq    %rdi, %r14
 ;;       subq    $0x10, %rsp
 ;;       movq    %rdi, 8(%rsp)
 ;;       movq    %rsi, (%rsp)
 ;;       movl    $0, %eax
+;;       andq    $7, %rax
+;;       cmpq    $0, %rax
+;;       jne     0x56
+;;   3f: movl    $0, %eax
 ;;       movq    0x60(%r14), %rcx
 ;;       addq    %rax, %rcx
 ;;       movq    (%rcx), %rax
 ;;       addq    $0x10, %rsp
 ;;       popq    %rbp
 ;;       retq
-;;   41: ud2
+;;   54: ud2
+;;   56: ud2
diff --git a/tests/disas/winch/x64/atomic/load/i64_atomic_load16_u.wat b/tests/disas/winch/x64/atomic/load/i64_atomic_load16_u.wat
@@ -14,16 +14,21 @@
 ;;       movq    0x10(%r11), %r11
 ;;       addq    $0x10, %r11
 ;;       cmpq    %rsp, %r11
-;;       ja      0x42
+;;       ja      0x55
 ;;   1c: movq    %rdi, %r14
 ;;       subq    $0x10, %rsp
 ;;       movq    %rdi, 8(%rsp)
 ;;       movq    %rsi, (%rsp)
 ;;       movl    $0, %eax
+;;       andw    $1, %ax
+;;       cmpw    $0, %ax
+;;       jne     0x57
+;;   3f: movl    $0, %eax
 ;;       movq    0x60(%r14), %rcx
 ;;       addq    %rax, %rcx
 ;;       movzwq  (%rcx), %rax
 ;;       addq    $0x10, %rsp
 ;;       popq    %rbp
 ;;       retq
-;;   42: ud2
+;;   55: ud2
+;;   57: ud2
diff --git a/tests/disas/winch/x64/atomic/load/i64_atomic_load32_u.wat b/tests/disas/winch/x64/atomic/load/i64_atomic_load32_u.wat
@@ -14,16 +14,21 @@
 ;;       movq    0x10(%r11), %r11
 ;;       addq    $0x10, %r11
 ;;       cmpq    %rsp, %r11
-;;       ja      0x40
+;;       ja      0x51
 ;;   1c: movq    %rdi, %r14
 ;;       subq    $0x10, %rsp
 ;;       movq    %rdi, 8(%rsp)
 ;;       movq    %rsi, (%rsp)
 ;;       movl    $0, %eax
+;;       andl    $3, %eax
+;;       cmpl    $0, %eax
+;;       jne     0x53
+;;   3d: movl    $0, %eax
 ;;       movq    0x60(%r14), %rcx
 ;;       addq    %rax, %rcx
 ;;       movl    (%rcx), %eax
 ;;       addq    $0x10, %rsp
 ;;       popq    %rbp
 ;;       retq
-;;   40: ud2
+;;   51: ud2
+;;   53: ud2
diff --git a/tests/disas/winch/x64/atomic/store/i32_atomic_store.wat b/tests/disas/winch/x64/atomic/store/i32_atomic_store.wat
@@ -0,0 +1,34 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module 
+  (import "env" "memory" (memory 1 1 shared))
+  (func (i32.atomic.store (i32.const 0) (i32.const 42))))
+;; wasm[0]::function[0]:
+;;       pushq   %rbp
+;;       movq    %rsp, %rbp
+;;       movq    8(%rdi), %r11
+;;       movq    0x10(%r11), %r11
+;;       addq    $0x10, %r11
+;;       cmpq    %rsp, %r11
+;;       ja      0x5c
+;;   1c: movq    %rdi, %r14
+;;       subq    $0x10, %rsp
+;;       movq    %rdi, 8(%rsp)
+;;       movq    %rsi, (%rsp)
+;;       movl    $0x2a, %eax
+;;       movl    $0, %ecx
+;;       andl    $3, %ecx
+;;       cmpl    $0, %ecx
+;;       jne     0x5e
+;;   42: movl    $0, %ecx
+;;       movq    0x58(%r14), %r11
+;;       movq    (%r11), %rdx
+;;       addq    %rcx, %rdx
+;;       movl    %eax, (%rdx)
+;;       mfence
+;;       addq    $0x10, %rsp
+;;       popq    %rbp
+;;       retq
+;;   5c: ud2
+;;   5e: ud2
diff --git a/tests/disas/winch/x64/atomic/store/i32_atomic_store16.wat b/tests/disas/winch/x64/atomic/store/i32_atomic_store16.wat
@@ -0,0 +1,34 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module 
+  (import "env" "memory" (memory 1 1 shared))
+  (func (i32.atomic.store16 (i32.const 0) (i32.const 42))))
+;; wasm[0]::function[0]:
+;;       pushq   %rbp
+;;       movq    %rsp, %rbp
+;;       movq    8(%rdi), %r11
+;;       movq    0x10(%r11), %r11
+;;       addq    $0x10, %r11
+;;       cmpq    %rsp, %r11
+;;       ja      0x5f
+;;   1c: movq    %rdi, %r14
+;;       subq    $0x10, %rsp
+;;       movq    %rdi, 8(%rsp)
+;;       movq    %rsi, (%rsp)
+;;       movl    $0x2a, %eax
+;;       movl    $0, %ecx
+;;       andw    $1, %cx
+;;       cmpw    $0, %cx
+;;       jne     0x61
+;;   44: movl    $0, %ecx
+;;       movq    0x58(%r14), %r11
+;;       movq    (%r11), %rdx
+;;       addq    %rcx, %rdx
+;;       movw    %ax, (%rdx)
+;;       mfence
+;;       addq    $0x10, %rsp
+;;       popq    %rbp
+;;       retq
+;;   5f: ud2
+;;   61: ud2
diff --git a/tests/disas/winch/x64/atomic/store/i32_atomic_store8.wat b/tests/disas/winch/x64/atomic/store/i32_atomic_store8.wat
@@ -0,0 +1,29 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module 
+  (import "env" "memory" (memory 1 1 shared))
+  (func (i32.atomic.store8 (i32.const 0) (i32.const 42))))
+;; wasm[0]::function[0]:
+;;       pushq   %rbp
+;;       movq    %rsp, %rbp
+;;       movq    8(%rdi), %r11
+;;       movq    0x10(%r11), %r11
+;;       addq    $0x10, %r11
+;;       cmpq    %rsp, %r11
+;;       ja      0x4b
+;;   1c: movq    %rdi, %r14
+;;       subq    $0x10, %rsp
+;;       movq    %rdi, 8(%rsp)
+;;       movq    %rsi, (%rsp)
+;;       movl    $0x2a, %eax
+;;       movl    $0, %ecx
+;;       movq    0x58(%r14), %r11
+;;       movq    (%r11), %rdx
+;;       addq    %rcx, %rdx
+;;       movb    %al, (%rdx)
+;;       mfence
+;;       addq    $0x10, %rsp
+;;       popq    %rbp
+;;       retq
+;;   4b: ud2
diff --git a/tests/disas/winch/x64/atomic/store/i64_atomic_store.wat b/tests/disas/winch/x64/atomic/store/i64_atomic_store.wat
@@ -0,0 +1,34 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module 
+  (import "env" "memory" (memory 1 1 shared))
+  (func (i64.atomic.store (i32.const 0) (i64.const 42))))
+;; wasm[0]::function[0]:
+;;       pushq   %rbp
+;;       movq    %rsp, %rbp
+;;       movq    8(%rdi), %r11
+;;       movq    0x10(%r11), %r11
+;;       addq    $0x10, %r11
+;;       cmpq    %rsp, %r11
+;;       ja      0x61
+;;   1c: movq    %rdi, %r14
+;;       subq    $0x10, %rsp
+;;       movq    %rdi, 8(%rsp)
+;;       movq    %rsi, (%rsp)
+;;       movq    $0x2a, %rax
+;;       movl    $0, %ecx
+;;       andq    $7, %rcx
+;;       cmpq    $0, %rcx
+;;       jne     0x63
+;;   46: movl    $0, %ecx
+;;       movq    0x58(%r14), %r11
+;;       movq    (%r11), %rdx
+;;       addq    %rcx, %rdx
+;;       movq    %rax, (%rdx)
+;;       mfence
+;;       addq    $0x10, %rsp
+;;       popq    %rbp
+;;       retq
+;;   61: ud2
+;;   63: ud2
diff --git a/tests/disas/winch/x64/atomic/store/i64_atomic_store16.wat b/tests/disas/winch/x64/atomic/store/i64_atomic_store16.wat
@@ -0,0 +1,34 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module 
+  (import "env" "memory" (memory 1 1 shared))
+  (func (i64.atomic.store16 (i32.const 0) (i64.const 42))))
+;; wasm[0]::function[0]:
+;;       pushq   %rbp
+;;       movq    %rsp, %rbp
+;;       movq    8(%rdi), %r11
+;;       movq    0x10(%r11), %r11
+;;       addq    $0x10, %r11
+;;       cmpq    %rsp, %r11
+;;       ja      0x61
+;;   1c: movq    %rdi, %r14
+;;       subq    $0x10, %rsp
+;;       movq    %rdi, 8(%rsp)
+;;       movq    %rsi, (%rsp)
+;;       movq    $0x2a, %rax
+;;       movl    $0, %ecx
+;;       andw    $1, %cx
+;;       cmpw    $0, %cx
+;;       jne     0x63
+;;   46: movl    $0, %ecx
+;;       movq    0x58(%r14), %r11
+;;       movq    (%r11), %rdx
+;;       addq    %rcx, %rdx
+;;       movw    %ax, (%rdx)
+;;       mfence
+;;       addq    $0x10, %rsp
+;;       popq    %rbp
+;;       retq
+;;   61: ud2
+;;   63: ud2
diff --git a/tests/disas/winch/x64/atomic/store/i64_atomic_store32.wat b/tests/disas/winch/x64/atomic/store/i64_atomic_store32.wat
@@ -0,0 +1,34 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module 
+  (import "env" "memory" (memory 1 1 shared))
+  (func (i64.atomic.store32 (i32.const 0) (i64.const 42))))
+;; wasm[0]::function[0]:
+;;       pushq   %rbp
+;;       movq    %rsp, %rbp
+;;       movq    8(%rdi), %r11
+;;       movq    0x10(%r11), %r11
+;;       addq    $0x10, %r11
+;;       cmpq    %rsp, %r11
+;;       ja      0x5e
+;;   1c: movq    %rdi, %r14
+;;       subq    $0x10, %rsp
+;;       movq    %rdi, 8(%rsp)
+;;       movq    %rsi, (%rsp)
+;;       movq    $0x2a, %rax
+;;       movl    $0, %ecx
+;;       andl    $3, %ecx
+;;       cmpl    $0, %ecx
+;;       jne     0x60
+;;   44: movl    $0, %ecx
+;;       movq    0x58(%r14), %r11
+;;       movq    (%r11), %rdx
+;;       addq    %rcx, %rdx
+;;       movl    %eax, (%rdx)
+;;       mfence
+;;       addq    $0x10, %rsp
+;;       popq    %rbp
+;;       retq
+;;   5e: ud2
+;;   60: ud2
diff --git a/tests/disas/winch/x64/atomic/store/i64_atomic_store8.wat b/tests/disas/winch/x64/atomic/store/i64_atomic_store8.wat
@@ -0,0 +1,29 @@
+;;! target = "x86_64"
+;;! test = "winch"
+
+(module 
+  (import "env" "memory" (memory 1 1 shared))
+  (func (i64.atomic.store8 (i32.const 0) (i64.const 42))))
+;; wasm[0]::function[0]:
+;;       pushq   %rbp
+;;       movq    %rsp, %rbp
+;;       movq    8(%rdi), %r11
+;;       movq    0x10(%r11), %r11
+;;       addq    $0x10, %r11
+;;       cmpq    %rsp, %r11
+;;       ja      0x4d
+;;   1c: movq    %rdi, %r14
+;;       subq    $0x10, %rsp
+;;       movq    %rdi, 8(%rsp)
+;;       movq    %rsi, (%rsp)
+;;       movq    $0x2a, %rax
+;;       movl    $0, %ecx
+;;       movq    0x58(%r14), %r11
+;;       movq    (%r11), %rdx
+;;       addq    %rcx, %rdx
+;;       movb    %al, (%rdx)
+;;       mfence
+;;       addq    $0x10, %rsp
+;;       popq    %rbp
+;;       retq
+;;   4d: ud2