A lightweight WebAssembly runtime that is fast, secure, and standards-compliant
[Cranelift] `(X ^ Y) ^ Y --> X`, `(X & Y) | X --> X` (#11680)
Bongjun Jang76cd1acacd90
diff --git a/cranelift/codegen/src/opts/bitops.isle b/cranelift/codegen/src/opts/bitops.isleindex cf51c4da8b..82b84547f3 100644--- a/cranelift/codegen/src/opts/bitops.isle+++ b/cranelift/codegen/src/opts/bitops.isle@@ -213,3 +213,11 @@(rule (simplify (iadd ty (bor ty x y) (band ty x y))) (iadd ty x y))(rule (simplify (iadd ty (band ty x y) (bor ty x y))) (iadd ty x y))+; (x & y) | x --> x+(rule (simplify (bor ty (band ty x y) x)) x)+(rule (simplify (bor ty x (band ty x y))) x)++; (x ^ y) ^ y --> x+(rule (simplify (bxor ty (bxor ty x y) y)) x)+(rule (simplify (bxor ty y (bxor ty x y))) x)+diff --git a/cranelift/filetests/filetests/egraph/fold-or-and.clif b/cranelift/filetests/filetests/egraph/fold-or-and.clifnew file mode 100644index 0000000000..d6acdefd40--- /dev/null+++ b/cranelift/filetests/filetests/egraph/fold-or-and.clif@@ -0,0 +1,71 @@+test optimize precise-output+set opt_level=speed+target x86_64++function %fold_or_and11(i8, i8) -> i8 {+block0(v0: i8, v1: i8):+ v2 = band v0, v1+ v3 = bor v2, v0+ return v3+}++; function %fold_or_and11(i8, i8) -> i8 fast {+; block0(v0: i8, v1: i8):+; return v0+; }++++function %fold_or_and12(i16, i16) -> i16 {+block0(v0: i16, v1: i16):+ v2 = band v0, v1+ v3 = bor v2, v0+ return v3+}++; function %fold_or_and12(i16, i16) -> i16 fast {+; block0(v0: i16, v1: i16):+; return v0+; }++++function %fold_or_and13(i32, i32) -> i32 {+block0(v0: i32, v1: i32):+ v2 = band v0, v1+ v3 = bor v2, v0+ return v3+}++; function %fold_or_and13(i32, i32) -> i32 fast {+; block0(v0: i32, v1: i32):+; return v0+; }++++function %fold_or_and14(i64, i64) -> i64 {+block0(v0: i64, v1: i64):+ v2 = band v0, v1+ v3 = bor v2, v0+ return v3+}++; function %fold_or_and14(i64, i64) -> i64 fast {+; block0(v0: i64, v1: i64):+; return v0+; }+++function %fold_or_and14(i64, i64) -> i64 {+block0(v0: i64, v1: i64):+ v2 = band v0, v1+ v3 = bor v0, v2+ return v3+}++; function %fold_or_and14(i64, i64) -> i64 fast {+; block0(v0: i64, v1: i64):+; return v0+; }+diff --git a/cranelift/filetests/filetests/egraph/fold-xor-xor.clif b/cranelift/filetests/filetests/egraph/fold-xor-xor.clifnew file mode 100644index 0000000000..1eb82d4c5d--- /dev/null+++ b/cranelift/filetests/filetests/egraph/fold-xor-xor.clif@@ -0,0 +1,71 @@+test optimize precise-output+set opt_level=speed+target x86_64++function %fold_xor_xor11(i8, i8) -> i8 {+block0(v0: i8, v1: i8):+ v2 = bxor v0, v1+ v3 = bxor v2, v1+ return v3+}++; function %fold_xor_xor11(i8, i8) -> i8 fast {+; block0(v0: i8, v1: i8):+; return v0+; }++++function %fold_xor_xor12(i16, i16) -> i16 {+block0(v0: i16, v1: i16):+ v2 = bxor v0, v1+ v3 = bxor v2, v1+ return v3+}++; function %fold_xor_xor12(i16, i16) -> i16 fast {+; block0(v0: i16, v1: i16):+; return v0+; }++++function %fold_xor_xor13(i32, i32) -> i32 {+block0(v0: i32, v1: i32):+ v2 = bxor v0, v1+ v3 = bxor v2, v1+ return v3+}++; function %fold_xor_xor13(i32, i32) -> i32 fast {+; block0(v0: i32, v1: i32):+; return v0+; }++++function %fold_xor_xor14(i64, i64) -> i64 {+block0(v0: i64, v1: i64):+ v2 = bxor v0, v1+ v3 = bxor v2, v1+ return v3+}++; function %fold_xor_xor14(i64, i64) -> i64 fast {+; block0(v0: i64, v1: i64):+; return v0+; }+++function %fold_xor_xor15(i64, i64) -> i64 {+block0(v0: i64, v1: i64):+ v2 = bxor v0, v1+ v3 = bxor v1, v2+ return v3+}++; function %fold_xor_xor15(i64, i64) -> i64 fast {+; block0(v0: i64, v1: i64):+; return v0+; }+diff --git a/cranelift/filetests/filetests/runtests/bitops.clif b/cranelift/filetests/filetests/runtests/bitops.clifindex fd89e9efc1..e15a268d0c 100644--- a/cranelift/filetests/filetests/runtests/bitops.clif+++ b/cranelift/filetests/filetests/runtests/bitops.clif@@ -194,3 +194,21 @@ block0(v0: i64, v1: i64):; run: %fold_add_and_xor4(1, 2) == 3+function %fold_or_and14(i64, i64) -> i64 {+block0(v0: i64, v1: i64):+ v2 = band v0, v1+ v3 = bor v2, v0+ return v3+}++; run: %fold_or_and14(0xffff0000, 0x0000ffff) == 0xffff0000++function %fold_xor_xor15(i64, i64) -> i64 {+block0(v0: i64, v1: i64):+ v2 = bxor v0, v1+ v3 = bxor v1, v2+ return v3+}++; run: %fold_xor_xor15(0xdead0000, 0x0000beef) == 0xdead0000+