implement BYTE

Signed-off-by: xermicus <cyrill@parity.io>

crates/integration/contracts/Bitwise.sol

@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+
+pragma solidity ^0.8;
+
+contract Bitwise {
+    function opByte(uint i, uint x) public payable returns (uint ret) {
+        assembly {
+            ret := byte(i, x)
+        }
+    }
+}

crates/integration/src/cases.rs

@@ -154,6 +154,12 @@ sol!(
     }
 );
 
+sol!(
+    contract Bitwise {
+        function opByte(uint i, uint x) public payable returns (uint ret);
+    }
+);
+
 impl Contract {
     /// Execute the contract.
     ///
@@ -508,6 +514,18 @@ impl Contract {
             calldata: Value::balance_ofCall::new((address,)).abi_encode(),
         }
     }
+
+    pub fn bitwise_byte(index: U256, value: U256) -> Self {
+        let code = include_str!("../contracts/Bitwise.sol");
+        let name = "Bitwise";
+
+        Self {
+            name,
+            evm_runtime: crate::compile_evm_bin_runtime(name, code),
+            pvm_runtime: crate::compile_blob(name, code),
+            calldata: Bitwise::opByteCall::new((index, value)).abi_encode(),
+        }
+    }
 }
 
 #[cfg(test)]

crates/integration/src/tests.rs

@@ -1,3 +1,5 @@
+use std::str::FromStr;
+
 use alloy_primitives::{keccak256, Address, FixedBytes, B256, I256, U256};
 use alloy_sol_types::{sol, SolCall, SolValue};
 use rayon::iter::{IntoParallelRefIterator, ParallelIterator};
@@ -590,3 +592,25 @@ fn balance() {
     let received = U256::from_be_slice(&output.data);
     assert_eq!(expected, received)
 }
+
+#[test]
+fn bitwise_byte() {
+    assert_success(&Contract::bitwise_byte(U256::ZERO, U256::ZERO), true);
+    assert_success(&Contract::bitwise_byte(U256::ZERO, U256::MAX), true);
+    assert_success(&Contract::bitwise_byte(U256::MAX, U256::ZERO), true);
+    assert_success(
+        &Contract::bitwise_byte(U256::from_str("18446744073709551619").unwrap(), U256::MAX),
+        true,
+    );
+
+    let de_bruijn_sequence =
+        hex::decode("4060503824160d0784426150b864361d0f88c4a27148ac5a2f198d46e391d8f4").unwrap();
+    let value = U256::from_be_bytes::<32>(de_bruijn_sequence.clone().try_into().unwrap());
+
+    for (index, byte) in de_bruijn_sequence.iter().enumerate() {
+        let (_, output) = assert_success(&Contract::bitwise_byte(U256::from(index), value), true);
+        let expected = U256::from(*byte as i32);
+        let received = U256::abi_decode(&output.data, true).unwrap();
+        assert_eq!(expected, received)
+    }
+}

crates/llvm-context/src/polkavm/evm/bitwise.rs

@@ -195,7 +195,14 @@ where
     context.build_load(result_pointer, "shift_right_arithmetic_result")
 }
 
-/// Translates the `byte` instruction.
+/// Translates the `byte` instruction, extracting the byte of `operand_2`
+/// found at index `operand_1`, starting from the most significant bit.
+///
+/// Builds a logical `and` with a corresponding bit mask.
+///
+/// Because this opcode returns zero on overflows, the index `operand_1`
+/// is checked for overflow. On overflow, the mask will be all zeros,
+/// resulting in a branchless implementation.
 pub fn byte<'ctx, D>(
     context: &mut Context<'ctx, D>,
     operand_1: inkwell::values::IntValue<'ctx>,
@@ -204,14 +211,61 @@ pub fn byte<'ctx, D>(
 where
     D: Dependency + Clone,
 {
-    Ok(context
+    const MAX_INDEX_BYTES: u64 = 31;
-        .build_call(
+
-            context.llvm_runtime().byte,
+    let is_overflow_bit = context.builder().build_int_compare(
-            &[
+        inkwell::IntPredicate::ULE,
-                operand_1.as_basic_value_enum(),
+        operand_1,
-                operand_2.as_basic_value_enum(),
+        context.word_const(MAX_INDEX_BYTES),
-            ],
+        "is_overflow_bit",
-            "byte_call",
+    )?;
-        )
+    let is_overflow_byte = context.builder().build_int_z_extend(
-        .expect("Always exists"))
+        is_overflow_bit,
+        context.byte_type(),
+        "is_overflow_byte",
+    )?;
+    let mask_byte = context.builder().build_int_mul(
+        context.byte_type().const_all_ones(),
+        is_overflow_byte,
+        "mask_byte",
+    )?;
+    let mask_byte_word =
+        context
+            .builder()
+            .build_int_z_extend(mask_byte, context.word_type(), "mask_byte_word")?;
+
+    let index_truncated =
+        context
+            .builder()
+            .build_int_truncate(operand_1, context.byte_type(), "index_truncated")?;
+    let index_in_bits = context.builder().build_int_mul(
+        index_truncated,
+        context
+            .byte_type()
+            .const_int(revive_common::BIT_LENGTH_BYTE as u64, false),
+        "index_in_bits",
+    )?;
+    let index_from_most_significant_bit = context.builder().build_int_sub(
+        context.byte_type().const_int(
+            MAX_INDEX_BYTES * revive_common::BIT_LENGTH_BYTE as u64,
+            false,
+        ),
+        index_in_bits,
+        "index_from_msb",
+    )?;
+    let index_extended = context.builder().build_int_z_extend(
+        index_from_most_significant_bit,
+        context.word_type(),
+        "index",
+    )?;
+
+    let mask = context
+        .builder()
+        .build_left_shift(mask_byte_word, index_extended, "mask")?;
+    let masked_value = context.builder().build_and(operand_2, mask, "masked")?;
+    let byte = context
+        .builder()
+        .build_right_shift(masked_value, index_extended, false, "byte")?;
+
+    Ok(byte.as_basic_value_enum())
 }