Merge pull request #11046 from ethereum/moveAndInsideOr

Move AND with constant inside OR.

Changelog.md

@@ -4,6 +4,7 @@ Language Features:
 
 
 Compiler Features:
+ * Optimizer: Try to move ``and`` with constant inside ``or`` to improve storage writes of small types.
 
 
 Bugfixes:

libevmasm/RuleList.h

@@ -507,6 +507,23 @@ std::vector<SimplificationRule<Pattern>> simplificationRuleListPart7(
 		});
 	}
 
+	// Combine alternating AND/OR/AND with constant,
+	// AND(OR(AND(X, A), Y), B) -> OR(AND(X, A & B), AND(Y, B))
+	// Many versions due to commutativity.
+	for (auto const& inner: {Builtins::AND(X, A), Builtins::AND(A, X)})
+		for (auto const& second: {Builtins::OR(inner, Y), Builtins::OR(Y, inner)})
+		{
+			// We might swap X and Y but this is not an issue anymore.
+			rules.push_back({
+				Builtins::AND(second, B),
+				[=]() -> Pattern { return Builtins::OR(Builtins::AND(X, A.d() & B.d()), Builtins::AND(Y, B)); }
+			});
+			rules.push_back({
+				Builtins::AND(B, second),
+				[=]() -> Pattern { return Builtins::OR(Builtins::AND(X, A.d() & B.d()), Builtins::AND(Y, B)); }
+			});
+		}
+
 	rules.push_back({
 		// MUL(X, SHL(Y, 1)) -> SHL(Y, X)
 		Builtins::MUL(X, Builtins::SHL(Y, Word(1))),

test/cmdlineTests/optimize_full_storage_write/args

@@ -0,0 +1 @@
+--optimize --asm --metadata-hash none

test/cmdlineTests/optimize_full_storage_write/input.sol

@@ -0,0 +1,12 @@
+// SPDX-License-Identifier: GPL-3.0
+pragma solidity >=0.0;
+
+contract OptimizeFullSlotWrite {
+	uint64[4] nums;
+	function f() public {
+		nums[0] = 11111;
+		nums[1] = 22222;
+		nums[2] = 33333;
+		nums[3] = 44444;
+	}
+}

test/cmdlineTests/optimize_full_storage_write/output

@@ -0,0 +1,64 @@
+
+======= optimize_full_storage_write/input.sol:OptimizeFullSlotWrite =======
+EVM assembly:
+    /* "optimize_full_storage_write/input.sol":60:213  contract OptimizeFullSlotWrite {... */
+  mstore(0x40, 0x80)
+  callvalue
+  dup1
+  iszero
+  tag_1
+  jumpi
+  0x00
+  dup1
+  revert
+tag_1:
+  pop
+  dataSize(sub_0)
+  dup1
+  dataOffset(sub_0)
+  0x00
+  codecopy
+  0x00
+  return
+stop
+
+sub_0: assembly {
+        /* "optimize_full_storage_write/input.sol":60:213  contract OptimizeFullSlotWrite {... */
+      mstore(0x40, 0x80)
+      callvalue
+      dup1
+      iszero
+      tag_1
+      jumpi
+      0x00
+      dup1
+      revert
+    tag_1:
+      pop
+      jumpi(tag_2, lt(calldatasize, 0x04))
+      shr(0xe0, calldataload(0x00))
+      dup1
+      0x26121ff0
+      eq
+      tag_3
+      jumpi
+    tag_2:
+      0x00
+      dup1
+      revert
+        /* "optimize_full_storage_write/input.sol":111:211  function f() public {... */
+    tag_3:
+      tag_4
+        /* "optimize_full_storage_write/input.sol":192:207  nums[3] = 44444 */
+      0xad9c000000000000823500000000000056ce0000000000002b67
+        /* "optimize_full_storage_write/input.sol":135:139  nums */
+      0x00
+        /* "optimize_full_storage_write/input.sol":192:207  nums[3] = 44444 */
+      sstore
+        /* "optimize_full_storage_write/input.sol":111:211  function f() public {... */
+      jump
+    tag_4:
+      stop
+
+    auxdata: <AUXDATA REMOVED>
+}

test/formal/move_and_inside_or.py

@@ -0,0 +1,32 @@
+from rule import Rule
+from opcodes import *
+
+"""
+Rule:
+AND(OR(AND(X, A), Y), B) -> OR(AND(X, A & B), AND(Y, B))
+"""
+
+rule = Rule()
+
+# bit width is irrelevant
+n_bits = 128
+
+# Input vars
+X = BitVec('X', n_bits)
+Y = BitVec('Y', n_bits)
+A = BitVec('A', n_bits)
+B = BitVec('B', n_bits)
+
+# Non optimized result, explicit form
+nonopt = AND(OR(AND(X, A), Y), B)
+
+# Optimized result
+opt = OR(AND(X, A & B), AND(Y, B))
+
+rule.check(nonopt, opt)
+
+# Now the forms as they are constructod in the code.
+for inner in [AND(X, A), AND(A, X)]:
+    for second in [OR(inner, Y), OR(Y, inner)]:
+        rule.check(AND(second, B), opt)
+        rule.check(AND(B, second), opt)

test/libyul/yulOptimizerTests/fullSuite/and_or_combination.yul

@@ -0,0 +1,25 @@
+{
+    // Tests that masks that "add" up to
+    // the full bit width are removed.
+    let a := sload(0)
+    let x := sload(a)
+    let mask := 0xffffffffffffffff
+    x := or(and(x, not(mask)), 0x2b67)
+    mask := shl(64, mask)
+    x := or(and(x, not(mask)), 0x56ce0000000000000000)
+    mask := shl(64, mask)
+    x := or(and(x, not(mask)), shl(0x80, 0x8235))
+    mask := shl(64, mask)
+    x := or(and(x, not(mask)), shl(0xc2, 0x2b67))
+    sstore(a, x)
+}
+// ====
+// EVMVersion: >byzantium
+// ----
+// step: fullSuite
+//
+// {
+//     {
+//         sstore(sload(0), 0xad9c000000000000823500000000000056ce0000000000002b67)
+//     }
+// }