Ohadn/qm31 operations

CHANGELOG.md

@@ -2,6 +2,8 @@
 
 #### Upcoming Changes
 
+* feat: implement an opcode that computes QM31 arithmetics (add, sub, mul, div) in the VM [#1938](https://github.com/lambdaclass/cairo-vm/pull/1938)
+
 * feat: add functions that compute packed reduced qm31 arithmetics to `math_utils` [#1944](https://github.com/lambdaclass/cairo-vm/pull/1944)
 
 * feat: implement `Blake2sLastBlock` opcode in VM [#1932](https://github.com/lambdaclass/cairo-vm/pull/1932)

cairo_programs/stwo_exclusive_programs/qm31_opcodes_test.cairo

@@ -0,0 +1,183 @@
+from starkware.cairo.common.alloc import alloc
+from starkware.cairo.common.bool import FALSE, TRUE
+
+// Tests the QM31_add_mul opcode runners using specific examples as reference.
+// The test is comprised of 10 test cases, each of which tests a different combination of missing_operand, is_imm, and is_mul.
+// is_mul determines whether the operation is a multiplication or an addition of QM31 elements.
+// is_imm determines whether op1 is an immediate.
+// missing_operand determines which operand is missing and needs to be computed by the VM (0 for dst, 1 for op0, 2 fo op1).
+// the combination of is_imm=TRUE with missig_operand=2 is not tested because we do not use arithmetic opcodes to deduce immediates.
+func main{}() {
+    let qm31_op0_coordinates_a = 0x544b2fba;
+    let qm31_op0_coordinates_b = 0x673cff77;
+    let qm31_op0_coordinates_c = 0x60713d44;
+    let qm31_op0_coordinates_d = 0x499602d2;
+    let qm31_op0 = qm31_op0_coordinates_a + qm31_op0_coordinates_b*(2**36) + qm31_op0_coordinates_c*(2**72) + qm31_op0_coordinates_d*(2**108);
+
+    let qm31_op1_coordinates_a = 0x4b18de99;
+    let qm31_op1_coordinates_b = 0x55f6fb62;
+    let qm31_op1_coordinates_c = 0x6e2290d9;
+    let qm31_op1_coordinates_d = 0x7cd851b9;
+    let qm31_op1 = qm31_op1_coordinates_a + qm31_op1_coordinates_b*(2**36) + qm31_op1_coordinates_c*(2**72) + qm31_op1_coordinates_d*(2**108);
+
+    let qm31_add_dst_coordinates_a = 0x1f640e54;
+    let qm31_add_dst_coordinates_b = 0x3d33fada;
+    let qm31_add_dst_coordinates_c = 0x4e93ce1e;
+    let qm31_add_dst_coordinates_d = 0x466e548c;
+    let qm31_add_dst = qm31_add_dst_coordinates_a + qm31_add_dst_coordinates_b*(2**36) + qm31_add_dst_coordinates_c*(2**72) + qm31_add_dst_coordinates_d*(2**108);
+
+    let qm31_mul_dst_coordinates_a = 0x38810ab4;
+    let qm31_mul_dst_coordinates_b = 0x5a0fd30a;
+    let qm31_mul_dst_coordinates_c = 0x2527b81e;
+    let qm31_mul_dst_coordinates_d = 0x4b1ed1cd;
+    let qm31_mul_dst = qm31_mul_dst_coordinates_a + qm31_mul_dst_coordinates_b*(2**36) + qm31_mul_dst_coordinates_c*(2**72) + qm31_mul_dst_coordinates_d*(2**108);
+
+    let runner_output_mul_dst = run_qm31_operation(missing_operand=0, is_imm=FALSE, is_mul=TRUE, dst_or_op0=qm31_op0, op0_or_op1=qm31_op1);
+    assert runner_output_mul_dst = qm31_mul_dst;
+    let runner_output_add_dst = run_qm31_operation(missing_operand=0, is_imm=FALSE, is_mul=FALSE, dst_or_op0=qm31_op0, op0_or_op1=qm31_op1);
+    assert runner_output_add_dst = qm31_add_dst;
+
+    let runner_output_mul_op0 = run_qm31_operation(missing_operand=1, is_imm=FALSE, is_mul=TRUE, dst_or_op0=qm31_mul_dst, op0_or_op1=qm31_op1);
+    assert runner_output_mul_op0 = qm31_op0;
+    let runner_output_add_op0 = run_qm31_operation(missing_operand=1, is_imm=FALSE, is_mul=FALSE, dst_or_op0=qm31_add_dst, op0_or_op1=qm31_op1);
+    assert runner_output_add_op0 = qm31_op0;
+
+    let runner_output_mul_op1 = run_qm31_operation(missing_operand=2, is_imm=FALSE, is_mul=TRUE, dst_or_op0=qm31_mul_dst, op0_or_op1=qm31_op0);
+    assert runner_output_mul_op1 = qm31_op1;
+    let runner_output_add_op1 = run_qm31_operation(missing_operand=2, is_imm=FALSE, is_mul=FALSE, dst_or_op0=qm31_add_dst, op0_or_op1=qm31_op0);
+    assert runner_output_add_op1 = qm31_op1;
+
+    let runner_output_mul_dst = run_qm31_operation(missing_operand=0, is_imm=TRUE, is_mul=TRUE, dst_or_op0=qm31_op0, op0_or_op1=qm31_op1);
+    assert runner_output_mul_dst = qm31_mul_dst;
+    let runner_output_add_dst = run_qm31_operation(missing_operand=0, is_imm=TRUE, is_mul=FALSE, dst_or_op0=qm31_op0, op0_or_op1=qm31_op1);
+    assert runner_output_add_dst = qm31_add_dst;
+
+    let runner_output_mul_op0 = run_qm31_operation(missing_operand=1, is_imm=TRUE, is_mul=TRUE, dst_or_op0=qm31_mul_dst, op0_or_op1=qm31_op1);
+    assert runner_output_mul_op0 = qm31_op0;
+    let runner_output_add_op0 = run_qm31_operation(missing_operand=1, is_imm=TRUE, is_mul=FALSE, dst_or_op0=qm31_add_dst, op0_or_op1=qm31_op1);
+    assert runner_output_add_op0 = qm31_op0;
+
+    return ();
+}
+
+// Forces the runner to execute the QM31_add_mul opcode with the given operands.
+// missing_operand, is_imm, is_mul determine the configuration of the operation as described above.
+// dst_or_op0 is a felt representing the value of either the op0 (if missing_operand=0) or dst (otherwise) operand.
+// op0_or_op1 is a felt representing the value of either the op0 (if missing_operand=2) or op1 (otherwise) operand.
+// dst_or_op0 and op0_or_op1 are stored within addresses fp-4 and fp-3 respectively, they are passed to the instruction
+// using offsets wrt fp (unless is_imm=TRUE, in which case op1 has offset 1 relative to pc).
+// The missing operand has offset 0 relative to ap.
+// An instruction encoding with the appropriate flags and offsets is built, then written to [pc] and the runner is forced to execute QM31_add_mul.
+// The missing operand is deduced to [ap] and returned.
+func run_qm31_operation(
+    missing_operand: felt,
+    is_imm: felt,
+    is_mul: felt,
+    dst_or_op0: felt,
+    op0_or_op1: felt,
+) -> felt {
+    alloc_locals;
+
+    // Set flags and offsets.
+    let (local offsets) = alloc();
+    let (local flags) = alloc();
+
+    assert offsets[missing_operand] = 2**15; // the missing operand will be written to [ap]
+
+    assert flags[2] = is_imm; // flag_op1_imm = 0;
+    assert flags[5] = 1-is_mul; // flag_res_add = 1-is_mul;
+    assert flags[6] = is_mul; // flag_res_mul = is_mul;
+    assert flags[7] = 0; // flag_PC_update_jump = 0;
+    assert flags[8] = 0; // flag_PC_update_jump_rel = 0;
+    assert flags[9] = 0; // flag_PC_update_jnz = 0;
+    assert flags[10] = 0; // flag_ap_update_add = 0;
+    assert flags[11] = 0; // flag_ap_update_add_1 = 0;
+    assert flags[12] = 0; // flag_opcode_call = 0;
+    assert flags[13] = 0; // flag_opcode_ret = 0;
+    assert flags[14] = 1; // flag_opcode_assert_eq = 1;
+
+    if (missing_operand == 0) {
+        assert offsets[1] = 2**15 - 4;
+        assert offsets[2] = 2**15 - 3 + 4 * is_imm;
+        assert flags[0] = 0; // flag_dst_base_fp
+        assert flags[1] = 1; // flag_op0_base_fp
+    }
+    if (missing_operand == 1) {
+        assert offsets[0] = 2**15 - 4;
+        assert offsets[2] = 2**15 - 3 + 4 * is_imm;
+        assert flags[0] = 1; // flag_dst_base_fp
+        assert flags[1] = 0; // flag_op0_base_fp
+    }
+    if (missing_operand == 2) {
+        assert is_imm = FALSE;
+        assert offsets[0] = 2**15 - 4;
+        assert offsets[1] = 2**15 - 3;
+        assert flags[0] = 1; // flag_dst_base_fp
+        assert flags[1] = 1; // flag_op0_base_fp
+    }
+    assert flags[3] = (2 - flags[0] - flags[1]) * (1 - is_imm); // flag_op1_base_fp
+    assert flags[4] = 1 - is_imm - flags[3]; // flag_op1_base_ap
+
+    // Compute the instruction encoding.
+    let flag_num = flags[0] + flags[1]*(2**1) + flags[2]*(2**2) + flags[3]*(2**3) + flags[4]*(2**4) + flags[5]*(2**5) + flags[6]*(2**6) + flags[14]*(2**14);
+    let qm31_opcode_extension_num = 3;
+    let instruction_encoding = offsets[0] + offsets[1]*(2**16) + offsets[2]*(2**32) + flag_num*(2**48) + qm31_opcode_extension_num*(2**63);
+
+    // Run the instruction and return the result.
+    if (is_imm == TRUE) {
+        assert op0_or_op1 = 0x7cd851b906e2290d9055f6fb6204b18de99;
+        if (missing_operand == 0) {
+            if (is_mul == TRUE) {
+                assert instruction_encoding=0x1c04680017ffc8000;
+                dw 0x1c04680017ffc8000;
+                dw 0x7cd851b906e2290d9055f6fb6204b18de99;
+                return [ap];
+            }
+            assert instruction_encoding=0x1c02680017ffc8000;
+            dw 0x1c02680017ffc8000;
+            dw 0x7cd851b906e2290d9055f6fb6204b18de99;
+            return [ap];
+        }
+        if (missing_operand == 1) {
+            if (is_mul == TRUE) {
+                assert instruction_encoding=0x1c045800180007ffc;
+                dw 0x1c045800180007ffc;
+                dw 0x7cd851b906e2290d9055f6fb6204b18de99;
+                return [ap];
+            }
+            assert instruction_encoding=0x1c025800180007ffc;
+            dw 0x1c025800180007ffc;
+            dw 0x7cd851b906e2290d9055f6fb6204b18de99;
+            return [ap];
+        }
+    }
+
+    if (missing_operand == 0) {
+        if (is_mul == TRUE) {
+            assert instruction_encoding=0x1c04a7ffd7ffc8000;
+            dw 0x1c04a7ffd7ffc8000;
+            return [ap];
+        }
+        assert instruction_encoding=0x1c02a7ffd7ffc8000;
+        dw 0x1c02a7ffd7ffc8000;
+        return [ap];
+    }
+    if (missing_operand == 1) {
+        if (is_mul == TRUE) {
+            assert instruction_encoding=0x1c0497ffd80007ffc;
+            dw 0x1c0497ffd80007ffc;
+            return [ap];
+        }
+        assert instruction_encoding=0x1c0297ffd80007ffc;
+        dw 0x1c0297ffd80007ffc;
+        return [ap];
+    }
+    if (is_mul == TRUE) {
+        assert instruction_encoding=0x1c05380007ffd7ffc;
+        dw 0x1c05380007ffd7ffc;
+        return [ap];
+    }
+    assert instruction_encoding=0x1c03380007ffd7ffc;
+    dw 0x1c03380007ffd7ffc;
+    return [ap];
+}

vm/src/lib.rs

@@ -61,6 +61,7 @@ pub mod hint_processor;
 pub mod math_utils;
 pub mod program_hash;
 pub mod serde;
+pub mod typed_operations;
 pub mod types;
 pub mod utils;
 pub mod vm;

vm/src/math_utils/mod.rs

@@ -24,7 +24,7 @@ lazy_static! {
             .collect::<Vec<_>>();
 }
 
-const STWO_PRIME: u64 = (1 << 31) - 1;
+pub const STWO_PRIME: u64 = (1 << 31) - 1;
 const STWO_PRIME_U128: u128 = STWO_PRIME as u128;
 const MASK_36: u64 = (1 << 36) - 1;
 const MASK_8: u64 = (1 << 8) - 1;
@@ -101,7 +101,7 @@ fn qm31_packed_reduced_read_coordinates(felt: Felt252) -> Result<[u64; 4], MathE
 /// Reduces four u64 coordinates and packs them into a single Felt252.
 /// STWO_PRIME fits in 36 bits, hence each coordinate can be represented by 36 bits and a QM31
 /// element can be stored in the first 144 bits of a Felt252.
-fn qm31_coordinates_to_packed_reduced(coordinates: [u64; 4]) -> Felt252 {
+pub fn qm31_coordinates_to_packed_reduced(coordinates: [u64; 4]) -> Felt252 {
     let bytes_part1 = ((coordinates[0] % STWO_PRIME) as u128
         + (((coordinates[1] % STWO_PRIME) as u128) << 36))
         .to_le_bytes();

vm/src/tests/cairo_run_test.rs

@@ -576,6 +576,14 @@ fn blake2s_opcode_test() {
     run_program_simple(program_data.as_slice());
 }
 
+#[test]
+#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)]
+fn qm31_opcodes_test() {
+    let program_data =
+        include_bytes!("../../../cairo_programs/stwo_exclusive_programs/qm31_opcodes_test.json");
+    run_program_simple(program_data.as_slice());
+}
+
 #[test]
 #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)]
 fn relocate_segments() {