Merge branch 'main' into upstream-08-28

	modified:   cranelift/filetests/filetests/isa/aarch64/user_stack_maps.`lif

Cargo.toml

@@ -50,7 +50,7 @@ wasmtime-cranelift = { workspace = true, optional = true }
 wasmtime-environ = { workspace = true }
 wasmtime-explorer = { workspace = true, optional = true }
 wasmtime-wast = { workspace = true, optional = true }
-wasi-common = { workspace = true, default-features = true, features = ["exit"], optional = true }
+wasi-common = { workspace = true, default-features = true, features = ["exit", "tokio"], optional = true }
 wasmtime-wasi = { workspace = true, default-features = true, optional = true }
 wasmtime-wasi-nn = { workspace = true, optional = true }
 wasmtime-wasi-runtime-config = { workspace = true, optional = true }
@@ -111,7 +111,7 @@ test-programs-artifacts = { workspace = true }
 bytesize = "1.3.0"
 wit-component = { workspace = true }
 cranelift-filetests = { workspace = true }
-cranelift-codegen = { workspace = true }
+cranelift-codegen = { workspace = true, features = ["disas", "trace-log", "timing"] }
 cranelift-reader = { workspace = true }
 toml = { workspace = true }
 similar = { workspace = true }
@@ -248,7 +248,7 @@ byte-array-literals = { path = "crates/wasi-preview1-component-adapter/byte-arra
 
 # Bytecode Alliance maintained dependencies:
 # ---------------------------
-regalloc2 = "0.9.3"
+regalloc2 = "0.9.4"
 
 # cap-std family:
 target-lexicon = "0.12.16"
@@ -390,6 +390,7 @@ default = [
   # cost, so allow disabling this through disabling of our own `default`
   # feature.
   "clap/default",
+  "clap/wrap_help",
 ]
 
 # ========================================
@@ -441,7 +442,7 @@ explore = ["dep:wasmtime-explorer", "dep:tempfile"]
 wast = ["dep:wasmtime-wast"]
 config = ["cache"]
 compile = ["cranelift"]
-run = ["dep:wasmtime-wasi", "wasmtime/runtime", "dep:listenfd", "dep:wasi-common"]
+run = ["dep:wasmtime-wasi", "wasmtime/runtime", "dep:listenfd", "dep:wasi-common", "dep:tokio"]
 
 [[test]]
 name = "host_segfault"

cranelift/codegen/src/machinst/vcode.rs

@@ -1297,7 +1297,7 @@ impl<I: VCodeInst> RegallocFunction for VCode<I> {
 
     fn block_insns(&self, block: BlockIndex) -> InstRange {
         let range = self.block_ranges.get(block.index());
-        InstRange::forward(InsnIndex::new(range.start), InsnIndex::new(range.end))
+        InstRange::new(InsnIndex::new(range.start), InsnIndex::new(range.end))
     }
 
     fn block_succs(&self, block: BlockIndex) -> &[BlockIndex] {

cranelift/codegen/src/opts/icmp.isle

@@ -15,14 +15,14 @@
 
 ;; Optimize icmp-of-icmp.
 ;; ne(icmp(ty, cc, x, y), 0) == icmp(ty, cc, x, y)
-;; e.g. neq(ugt(x, y), 0) == ugt(x, y) 
+;; e.g. neq(ugt(x, y), 0) == ugt(x, y)
 (rule (simplify (ne ty
                       (uextend_maybe _ inner @ (icmp ty _ _ _))
                       (iconst_u _ 0)))
       (subsume inner))
 
 ;; eq(icmp(ty, cc, x, y), 0) == icmp(ty, cc_complement, x, y)
-;; e.g. eq(ugt(x, y), 0) == ule(x, y) 
+;; e.g. eq(ugt(x, y), 0) == ule(x, y)
 (rule (simplify (eq ty
                       (uextend_maybe _ (icmp ty cc x y))
                       (iconst_u _ 0)))
@@ -213,3 +213,46 @@
 (rule (intcc_class (IntCC.SignedGreaterThanOrEqual)) 2)
 (rule (intcc_class (IntCC.Equal)) 3)
 (rule (intcc_class (IntCC.NotEqual)) 3)
+
+;; Pattern-match what LLVM emits today for 128-bit comparisons into actual
+;; 128-bit comparisons. Platforms like x64 and aarch64 have more optimal
+;; lowerings for 128-bit arithmetic than the default structure.
+(rule (simplify (select ty (eq _   a_hi @ (value_type $I64) b_hi @ (value_type $I64))
+                           (uge ty a_lo @ (value_type $I64) b_lo @ (value_type $I64))
+                           (uge ty a_hi b_hi)))
+      (uge ty (iconcat $I64 a_lo a_hi) (iconcat $I64 b_lo b_hi)))
+
+(rule (simplify (select ty (eq _   a_hi @ (value_type $I64) b_hi @ (value_type $I64))
+                           (uge ty a_lo @ (value_type $I64) b_lo @ (value_type $I64))
+                           (sge ty a_hi b_hi)))
+      (sge ty (iconcat $I64 a_lo a_hi) (iconcat $I64 b_lo b_hi)))
+
+(rule (simplify (select ty (eq _   a_hi @ (value_type $I64) b_hi @ (value_type $I64))
+                           (ugt ty a_lo @ (value_type $I64) b_lo @ (value_type $I64))
+                           (ugt ty a_hi b_hi)))
+      (ugt ty (iconcat $I64 a_lo a_hi) (iconcat $I64 b_lo b_hi)))
+
+(rule (simplify (select ty (eq _   a_hi @ (value_type $I64) b_hi @ (value_type $I64))
+                           (ugt ty a_lo @ (value_type $I64) b_lo @ (value_type $I64))
+                           (sgt ty a_hi b_hi)))
+      (sgt ty (iconcat $I64 a_lo a_hi) (iconcat $I64 b_lo b_hi)))
+
+(rule (simplify (select ty (eq _   a_hi @ (value_type $I64) b_hi @ (value_type $I64))
+                           (ule ty a_lo @ (value_type $I64) b_lo @ (value_type $I64))
+                           (ule ty a_hi b_hi)))
+      (ule ty (iconcat $I64 a_lo a_hi) (iconcat $I64 b_lo b_hi)))
+
+(rule (simplify (select ty (eq _   a_hi @ (value_type $I64) b_hi @ (value_type $I64))
+                           (ule ty a_lo @ (value_type $I64) b_lo @ (value_type $I64))
+                           (sle ty a_hi b_hi)))
+      (sle ty (iconcat $I64 a_lo a_hi) (iconcat $I64 b_lo b_hi)))
+
+(rule (simplify (select ty (eq _   a_hi @ (value_type $I64) b_hi @ (value_type $I64))
+                           (ult ty a_lo @ (value_type $I64) b_lo @ (value_type $I64))
+                           (ult ty a_hi b_hi)))
+      (ult ty (iconcat $I64 a_lo a_hi) (iconcat $I64 b_lo b_hi)))
+
+(rule (simplify (select ty (eq _   a_hi @ (value_type $I64) b_hi @ (value_type $I64))
+                           (ult ty a_lo @ (value_type $I64) b_lo @ (value_type $I64))
+                           (slt ty a_hi b_hi)))
+      (slt ty (iconcat $I64 a_lo a_hi) (iconcat $I64 b_lo b_hi)))

cranelift/codegen/src/opts/selects.isle

@@ -75,3 +75,14 @@
 (rule (simplify (bor (ty_vec128 ty) (band ty (bnot ty c) y) (band ty x c))) (bitselect ty c x y))
 (rule (simplify (bor (ty_vec128 ty) (band ty y (bnot ty c)) (band ty c x))) (bitselect ty c x y))
 (rule (simplify (bor (ty_vec128 ty) (band ty y (bnot ty c)) (band ty x c))) (bitselect ty c x y))
+
+;; Lift an extend operation outside of a `select` if the extend is happening
+;; on both the consequent and the alternative.
+(rule (simplify (select ty cond
+                           (uextend ty a @ (value_type small))
+                           (uextend ty b @ (value_type small))))
+      (uextend ty (select small cond a b)))
+(rule (simplify (select ty cond
+                           (sextend ty a @ (value_type small))
+                           (sextend ty b @ (value_type small))))
+      (sextend ty (select small cond a b)))

cranelift/filetests/filetests/egraph/select.clif

@@ -208,3 +208,25 @@ block0(v0: i32, v1: i32):
 ; check: v6 = icmp sgt v0, v1
 ; check: v8 = bmask.i64 v6
 ; check: return v8
+
+function %lift_uextend_out_of_select(i8, i32, i32) -> i64 {
+block0(v0: i8, v1: i32, v2: i32):
+    v3 = uextend.i64 v1
+    v4 = uextend.i64 v2
+    v5 = select v0, v3, v4
+    return v5
+}
+; check: v6 = select v0, v1, v2
+; check: v7 = uextend.i64 v6
+; check: return v7
+
+function %lift_sextend_out_of_select(i8, i32, i32) -> i64 {
+block0(v0: i8, v1: i32, v2: i32):
+    v3 = sextend.i64 v1
+    v4 = sextend.i64 v2
+    v5 = select v0, v3, v4
+    return v5
+}
+; check: v6 = select v0, v1, v2
+; check: v7 = sextend.i64 v6
+; check: return v7

cranelift/filetests/filetests/isa/aarch64/call-pauth-bkey.clif

@@ -63,8 +63,8 @@ block0(v0: i64):
 ;   stp fp, lr, [sp, #-16]!
 ;   mov fp, sp
 ; block0:
-;   mov x0, x2
 ;   load_ext_name x3, TestCase(%g)+0
+;   mov x0, x2
 ;   blr x3
 ;   mov x2, x0
 ;   ldp fp, lr, [sp], #16
@@ -76,11 +76,11 @@ block0(v0: i64):
 ;   stp x29, x30, [sp, #-0x10]!
 ;   mov x29, sp
 ; block1: ; offset 0xc
-;   mov x0, x2
-;   ldr x3, #0x18
-;   b #0x20
+;   ldr x3, #0x14
+;   b #0x1c
 ;   .byte 0x00, 0x00, 0x00, 0x00 ; reloc_external Abs8 %g 0
 ;   .byte 0x00, 0x00, 0x00, 0x00
+;   mov x0, x2
 ;   blr x3
 ;   mov x2, x0
 ;   ldp x29, x30, [sp], #0x10

cranelift/filetests/filetests/isa/aarch64/call-pauth.clif

@@ -64,8 +64,8 @@ block0(v0: i64):
 ;   stp fp, lr, [sp, #-16]!
 ;   mov fp, sp
 ; block0:
-;   mov x0, x2
 ;   load_ext_name x3, TestCase(%g)+0
+;   mov x0, x2
 ;   blr x3
 ;   mov x2, x0
 ;   ldp fp, lr, [sp], #16
@@ -77,11 +77,11 @@ block0(v0: i64):
 ;   stp x29, x30, [sp, #-0x10]!
 ;   mov x29, sp
 ; block1: ; offset 0xc
-;   mov x0, x2
-;   ldr x3, #0x18
-;   b #0x20
+;   ldr x3, #0x14
+;   b #0x1c
 ;   .byte 0x00, 0x00, 0x00, 0x00 ; reloc_external Abs8 %g 0
 ;   .byte 0x00, 0x00, 0x00, 0x00
+;   mov x0, x2
 ;   blr x3
 ;   mov x2, x0
 ;   ldp x29, x30, [sp], #0x10