Problem 014

Starting number under one million that produces the longest Collatz chain. Memoized in a heap buffer — Flow malloc → same C performance.

View problem on Project Euler

Performance comparison

Flow source

# Project Euler 014
# Starting number under one million that produces the longest Collatz chain.
# Memoized in a heap buffer — Flow malloc → same C performance.

extern {
    function calloc(n: i64, size: i64) -> ptr<void>
    function free(p: ptr<void>) -> void
}

function collatz_len(n0: i64, memo: ptr<i32>, limit: i64) -> i32 {
    let mut n: i64 = n0
    let mut steps: i32 = 0
    while n != 1 {
        if n < limit && memo[n] != 0 {
            return steps + memo[n]
        }
        if n % 2 == 0 {
            n = n / 2
        } else {
            n = 3 * n + 1
        }
        steps = steps + 1
    }
    return steps + 1
}

function solve(limit: i64) -> i64 {
    let memo: ptr<i32> = calloc(limit, 4)
    if memo == null {
        return -1
    }
    memo[1] = 1

    let mut best_n: i64 = 1
    let mut best_len: i32 = 1
    for n in 2..limit {
        let len: i32 = collatz_len(n, memo, limit)
        memo[n] = len
        if len > best_len {
            best_len = len
            best_n = n
        }
    }
    free(memo)
    return best_n
}

function main() -> i32 {
    printf("%lld\n", solve(1000000))
    return 0
}

Generated C

#include <stdint.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

/* Flow runtime helpers */
typedef struct flow_temp_node { struct flow_temp_node* next; } flow_temp_node;
static flow_temp_node* flow_temp_head = NULL;
static int flow_temp_atexit_set = 0;
__attribute__((unused)) static void flow_temp_free_all(void) {
    while (flow_temp_head) {
        flow_temp_node* n = flow_temp_head;
        flow_temp_head = n->next;
        free(n);
    }
}
__attribute__((unused)) static void* flow_temp_alloc(size_t nbytes) {
    flow_temp_node* node = (flow_temp_node*)malloc(sizeof(flow_temp_node) + nbytes);
    if (!node) return NULL;
    node->next = flow_temp_head;
    flow_temp_head = node;
    if (!flow_temp_atexit_set) {
        flow_temp_atexit_set = 1;
        atexit(flow_temp_free_all);
    }
    return (void*)(node + 1);
}
#ifndef FLOW_DIAG
#define FLOW_DIAG(msg) fprintf(stderr, "%s", (msg))
#endif
#ifndef FLOW_LOG
#define FLOW_LOG(fmt, ...) printf(fmt, __VA_ARGS__)
#endif
#ifndef FLOW_LOG_EMPTY
#define FLOW_LOG_EMPTY(fmt) printf(fmt)
#endif
static char* flow_strcat(const char* a, const char* b) {
    size_t la = strlen(a ? a : ""), lb = strlen(b ? b : "");
    char* r = (char*)flow_temp_alloc(la + lb + 1);
    if (!r) return NULL;
    if (la) memcpy(r, a, la);
    if (lb) memcpy(r + la, b, lb);
    r[la + lb] = '\0';
    return r;
}

#define __flow_in_arr(arr, val) __extension__ ({ \
    int _found = 0; \
    size_t _n = sizeof(arr)/sizeof((arr)[0]); \
    for (size_t _i = 0; _i < _n; _i++) { \
        if ((arr)[_i] == (val)) { _found = 1; break; } \
    } _found; })

/* Unified fault handler (MISRA #279) — override with -DFLOW_FAULT_HANDLER=fn */
#ifndef FLOW_FAULT_HANDLER
__attribute__((unused)) static inline void flow_fault_handler(const char* msg) {
    fprintf(stderr, "flow: %s\n", msg ? msg : "fault");
    abort();
#if defined(__GNUC__) || defined(__clang__)
    __builtin_unreachable();
#endif
}
#else
#define flow_fault_handler FLOW_FAULT_HANDLER
#endif
#define flow_div_by_zero_handler() flow_fault_handler("division by zero")
#define flow_shift_ub_handler() flow_fault_handler("invalid shift (amount out of range or left-shift of negative)")

#ifndef FLOW_CHECKED_DIV
#define FLOW_CHECKED_DIV(L, R) (((R) != 0) ? ((L) / (R)) : (flow_div_by_zero_handler(), (L) * 0))
#endif
#ifndef FLOW_CHECKED_MOD
#define FLOW_CHECKED_MOD(L, R) (((R) != 0) ? ((L) % (R)) : (flow_div_by_zero_handler(), (L) * 0))
#endif
#ifndef FLOW_CHECKED_SHL
#define FLOW_CHECKED_SHL(L, R) ((((R) >= 0) && ((unsigned long long)(R) < (sizeof(L) * 8ull)) && ((L) >= 0)) ? ((L) << (R)) : (flow_shift_ub_handler(), (L) * 0))
#endif
#ifndef FLOW_CHECKED_SHR
#define FLOW_CHECKED_SHR(L, R) ((((R) >= 0) && ((unsigned long long)(R) < (sizeof(L) * 8ull))) ? ((L) >> (R)) : (flow_shift_ub_handler(), (L) * 0))
#endif

#include <math.h>

void* _ui_state = NULL;

static inline float i32_to_f32(int32_t v) { return (float)v; }

/* Host stub for @gpu kernels (device codegen replaces this). */
static inline int32_t gpu_thread_id(void) { return 0; }

int32_t collatz_len_i64_ptr_i32_i64(int64_t n0, int32_t* memo, int64_t limit);
int64_t solve_i64(int64_t limit);
int32_t main(void);



int32_t collatz_len_i64_ptr_i32_i64(int64_t n0, int32_t* memo, int64_t limit) {
    int64_t n = n0;
    int32_t steps = 0;
    while (n != 1) {
        if ((n < limit && memo[n] != 0)) {
            return (steps + memo[n]);
        }
        if (FLOW_CHECKED_MOD((n), (2)) == 0) {
            n = FLOW_CHECKED_DIV((n), (2));
        } else {
            n = ((3 * n) + 1);
        }
        steps = (steps + 1);
    }
    return (steps + 1);
}

int64_t solve_i64(int64_t limit) {
    int32_t* memo = (int32_t*)(calloc(limit, 4));
    if (memo == NULL) {
        return (-1);
    }
    memo[1] = 1;
    int64_t best_n = 1;
    int32_t best_len = 1;
    int32_t __flow_step_1 = 1;
    for (int32_t n = 2; (2 <= limit) ? n < limit : n > limit; n += (2 <= limit) ? 1 : -1) {
        int32_t len = collatz_len_i64_ptr_i32_i64(n, memo, limit);
        memo[n] = len;
        if (len > best_len) {
            best_len = len;
            best_n = n;
        }
    }
    free(memo);
    return best_n;
}

int32_t main(void) {
    printf("%lld\n", solve_i64(1000000));
    return 0;
}

Generated MLIR

module {
  llvm.func @printf(!llvm.ptr, ...) -> i32
  llvm.mlir.global internal constant @str_0("%lld\n\00") {addr_space = 0 : i32} : !llvm.array<6 x i8>
  func.func private @calloc(i64, i64) -> !llvm.ptr
  func.func private @free(!llvm.ptr) -> ()
  func.func @collatz_len(%arg0: i64, %arg1: !llvm.ptr, %arg2: i64) -> i32 {
    %0 = llvm.mlir.constant(1 : i64) : i64
    %1 = llvm.alloca %0 x i64 : (i64) -> !llvm.ptr
    llvm.store %arg0, %1 : i64, !llvm.ptr
    %2 = arith.constant 0 : i32
    %3 = llvm.mlir.constant(1 : i64) : i64
    %4 = llvm.alloca %3 x i32 : (i64) -> !llvm.ptr
    llvm.store %2, %4 : i32, !llvm.ptr
    cf.br ^bb0
    ^bb0:
    %5 = llvm.load %1 : !llvm.ptr -> i64
    %6 = arith.constant 1 : i32
    %8 = arith.extsi %6 : i32 to i64
    %7 = arith.cmpi ne, %5, %8 : i64
    cf.cond_br %7, ^bb1, ^bb2
    ^bb1:
      %9 = llvm.load %1 : !llvm.ptr -> i64
      %10 = arith.cmpi slt, %9, %arg2 : i64
      %11 = scf.if %10 -> (i1) {
        %13 = llvm.load %1 : !llvm.ptr -> i64
        %14 = llvm.getelementptr %arg1[%13] : (!llvm.ptr, i64) -> !llvm.ptr, i32
        %12 = llvm.load %14 : !llvm.ptr -> i32
        %15 = arith.constant 0 : i32
        %16 = arith.cmpi ne, %12, %15 : i32
        scf.yield %16 : i1
      } else {
        %17 = arith.constant false
        scf.yield %17 : i1
      }
      cf.cond_br %11, ^bb3, ^bb4
      ^bb3:
        %18 = llvm.load %4 : !llvm.ptr -> i32
        %20 = llvm.load %1 : !llvm.ptr -> i64
        %21 = llvm.getelementptr %arg1[%20] : (!llvm.ptr, i64) -> !llvm.ptr, i32
        %19 = llvm.load %21 : !llvm.ptr -> i32
        %22 = arith.addi %18, %19 : i32
        func.return %22 : i32
      ^bb4:
        cf.br ^bb5
      ^bb5:
      %23 = llvm.load %1 : !llvm.ptr -> i64
      %24 = arith.constant 2 : i32
      %26 = arith.extsi %24 : i32 to i64
      %25 = arith.remsi %23, %26 : i64
      %27 = arith.constant 0 : i32
      %29 = arith.extsi %27 : i32 to i64
      %28 = arith.cmpi eq, %25, %29 : i64
      cf.cond_br %28, ^bb6, ^bb7
      ^bb6:
        %30 = llvm.load %1 : !llvm.ptr -> i64
        %31 = arith.constant 2 : i32
        %33 = arith.extsi %31 : i32 to i64
        %32 = arith.divsi %30, %33 : i64
        llvm.store %32, %1 : i64, !llvm.ptr
        cf.br ^bb8
      ^bb7:
        %34 = arith.constant 3 : i32
        %35 = llvm.load %1 : !llvm.ptr -> i64
        %37 = arith.extsi %34 : i32 to i64
        %36 = arith.muli %37, %35 : i64
        %38 = arith.constant 1 : i32
        %40 = arith.extsi %38 : i32 to i64
        %39 = arith.addi %36, %40 : i64
        llvm.store %39, %1 : i64, !llvm.ptr
        cf.br ^bb8
      ^bb8:
      %41 = llvm.load %4 : !llvm.ptr -> i32
      %42 = arith.constant 1 : i32
      %43 = arith.addi %41, %42 : i32
      llvm.store %43, %4 : i32, !llvm.ptr
      cf.br ^bb0
    ^bb2:
    %44 = llvm.load %4 : !llvm.ptr -> i32
    %45 = arith.constant 1 : i32
    %46 = arith.addi %44, %45 : i32
    func.return %46 : i32
  }
  func.func @solve(%arg0: i64) -> i64 {
    %48 = arith.constant 4 : i32
    %49 = arith.extsi %48 : i32 to i64
    %47 = func.call @calloc(%arg0, %49) : (i64, i64) -> !llvm.ptr
    %50 = llvm.mlir.zero : !llvm.ptr
    %51 = llvm.icmp "eq" %47, %50 : !llvm.ptr
    cf.cond_br %51, ^bb9, ^bb10
    ^bb9:
      %52 = arith.constant 1 : i32
      %54 = arith.constant 0 : i32
      %53 = arith.subi %54, %52 : i32
      %55 = arith.extsi %53 : i32 to i64
      func.return %55 : i64
    ^bb10:
      cf.br ^bb11
    ^bb11:
    %56 = arith.constant 1 : i32
    %57 = arith.constant 1 : i32
    %58 = arith.extsi %57 : i32 to i64
    %59 = llvm.getelementptr %47[%58] : (!llvm.ptr, i64) -> !llvm.ptr, i32
    llvm.store %56, %59 : i32, !llvm.ptr
    %60 = arith.constant 1 : i32
    %61 = arith.extsi %60 : i32 to i64
    %62 = llvm.mlir.constant(1 : i64) : i64
    %63 = llvm.alloca %62 x i64 : (i64) -> !llvm.ptr
    llvm.store %61, %63 : i64, !llvm.ptr
    %64 = arith.constant 1 : i32
    %65 = llvm.mlir.constant(1 : i64) : i64
    %66 = llvm.alloca %65 x i32 : (i64) -> !llvm.ptr
    llvm.store %64, %66 : i32, !llvm.ptr
    %67 = arith.constant 2 : i32
    %68 = arith.index_cast %67 : i32 to index
    %69 = arith.index_cast %arg0 : i32 to index
    %71 = arith.constant 1 : index
    %72 = arith.constant -1 : index
    %73 = arith.cmpi sle, %68, %69 : index
    %70 = arith.select %73, %71, %72 : index
    cf.br ^bb12(%68 : index)
    ^bb12(%74: index):
    %75 = arith.cmpi slt, %74, %69 : index
    %76 = arith.cmpi sgt, %74, %69 : index
    %77 = arith.select %73, %75, %76 : i1
    cf.cond_br %77, ^bb13(%74 : index), ^bb14(%74 : index)
    ^bb13(%78: index):
      %80 = arith.index_cast %78 : index to i64
      %79 = func.call @collatz_len(%80, %47, %arg0) : (i64, !llvm.ptr, i64) -> i32
      %81 = arith.index_cast %78 : index to i64
      %82 = llvm.getelementptr %47[%81] : (!llvm.ptr, i64) -> !llvm.ptr, i32
      llvm.store %79, %82 : i32, !llvm.ptr
      %83 = llvm.load %66 : !llvm.ptr -> i32
      %84 = arith.cmpi sgt, %79, %83 : i32
      cf.cond_br %84, ^bb15, ^bb16
      ^bb15:
        llvm.store %79, %66 : i32, !llvm.ptr
        %85 = arith.index_cast %78 : index to i64
        llvm.store %85, %63 : i64, !llvm.ptr
        cf.br ^bb17
      ^bb16:
        cf.br ^bb17
      ^bb17:
      %86 = arith.addi %78, %70 : index
      cf.br ^bb12(%86 : index)
    ^bb14(%87: index):
    func.call @free(%47) : (!llvm.ptr) -> ()
    %89 = llvm.load %63 : !llvm.ptr -> i64
    func.return %89 : i64
  }
  func.func @main() -> i32 {
    %90 = llvm.mlir.addressof @str_0 : !llvm.ptr
    %92 = arith.constant 1000000 : i32
    %93 = arith.extsi %92 : i32 to i64
    %91 = func.call @solve(%93) : (i64) -> i64
    %94 = llvm.call @printf(%90, %91) vararg(!llvm.func<i32 (ptr, ...)>) : (!llvm.ptr, i64) -> i32
    %95 = arith.constant 0 : i32
    func.return %95 : i32
  }
}