Problem 010

Sum of all primes below two million. Linear sieve → portable C. Same algorithm you'd write in C, less noise.

View problem on Project Euler

Performance comparison

Flow source

# Project Euler 010
# Sum of all primes below two million.
#
# Linear sieve → portable C. Same algorithm you'd write in C, less noise.

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

function sum_primes_below(limit: i64) -> i64 {
    let sieve: ptr<i8> = calloc(limit, 1)
    if sieve == null {
        return -1
    }

    # sieve[i] == 0 means prime (calloc zero-fills)
    sieve[0] = 1
    sieve[1] = 1

    let mut p: i64 = 2
    while p * p < limit {
        if sieve[p] == 0 {
            let mut m: i64 = p * p
            while m < limit {
                sieve[m] = 1
                m = m + p
            }
        }
        p = p + 1
    }

    let mut total: i64 = 0
    for i in 2..limit {
        if sieve[i] == 0 {
            total = total + i
        }
    }

    free(sieve)
    return total
}

function main() -> i32 {
    printf("%lld\n", sum_primes_below(2000000))
    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; }

int64_t sum_primes_below_i64(int64_t limit);
int32_t main(void);



int64_t sum_primes_below_i64(int64_t limit) {
    int8_t* sieve = (int8_t*)(calloc(limit, 1));
    if (sieve == NULL) {
        return (-1);
    }
    sieve[0] = 1;
    sieve[1] = 1;
    int64_t p = 2;
    while ((p * p) < limit) {
        if (sieve[p] == 0) {
            int64_t m = (p * p);
            while (m < limit) {
                sieve[m] = 1;
                m = (m + p);
            }
        }
        p = (p + 1);
    }
    int64_t total = 0;
    int32_t __flow_step_1 = 1;
    for (int32_t i = 2; (2 <= limit) ? i < limit : i > limit; i += (2 <= limit) ? 1 : -1) {
        if (sieve[i] == 0) {
            total = (total + i);
        }
    }
    free(sieve);
    return total;
}

int32_t main(void) {
    printf("%lld\n", sum_primes_below_i64(2000000));
    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 @sum_primes_below(%arg0: i64) -> i64 {
    %1 = arith.constant 1 : i32
    %2 = arith.extsi %1 : i32 to i64
    %0 = func.call @calloc(%arg0, %2) : (i64, i64) -> !llvm.ptr
    %3 = llvm.mlir.zero : !llvm.ptr
    %4 = llvm.icmp "eq" %0, %3 : !llvm.ptr
    cf.cond_br %4, ^bb0, ^bb1
    ^bb0:
      %5 = arith.constant 1 : i32
      %7 = arith.constant 0 : i32
      %6 = arith.subi %7, %5 : i32
      %8 = arith.extsi %6 : i32 to i64
      func.return %8 : i64
    ^bb1:
      cf.br ^bb2
    ^bb2:
    %9 = arith.constant 1 : i32
    %10 = arith.constant 0 : i32
    %11 = arith.trunci %9 : i32 to i8
    %12 = arith.extsi %10 : i32 to i64
    %13 = llvm.getelementptr %0[%12] : (!llvm.ptr, i64) -> !llvm.ptr, i8
    llvm.store %11, %13 : i8, !llvm.ptr
    %14 = arith.constant 1 : i32
    %15 = arith.constant 1 : i32
    %16 = arith.trunci %14 : i32 to i8
    %17 = arith.extsi %15 : i32 to i64
    %18 = llvm.getelementptr %0[%17] : (!llvm.ptr, i64) -> !llvm.ptr, i8
    llvm.store %16, %18 : i8, !llvm.ptr
    %19 = arith.constant 2 : i32
    %20 = arith.extsi %19 : i32 to i64
    %21 = llvm.mlir.constant(1 : i64) : i64
    %22 = llvm.alloca %21 x i64 : (i64) -> !llvm.ptr
    llvm.store %20, %22 : i64, !llvm.ptr
    cf.br ^bb3
    ^bb3:
    %23 = llvm.load %22 : !llvm.ptr -> i64
    %24 = llvm.load %22 : !llvm.ptr -> i64
    %25 = arith.muli %23, %24 : i64
    %26 = arith.cmpi slt, %25, %arg0 : i64
    cf.cond_br %26, ^bb4, ^bb5
    ^bb4:
      %28 = llvm.load %22 : !llvm.ptr -> i64
      %29 = llvm.getelementptr %0[%28] : (!llvm.ptr, i64) -> !llvm.ptr, i8
      %27 = llvm.load %29 : !llvm.ptr -> i8
      %30 = arith.constant 0 : i32
      %32 = arith.extsi %27 : i8 to i32
      %31 = arith.cmpi eq, %32, %30 : i32
      cf.cond_br %31, ^bb6, ^bb7
      ^bb6:
        %33 = llvm.load %22 : !llvm.ptr -> i64
        %34 = llvm.load %22 : !llvm.ptr -> i64
        %35 = arith.muli %33, %34 : i64
        %36 = llvm.mlir.constant(1 : i64) : i64
        %37 = llvm.alloca %36 x i64 : (i64) -> !llvm.ptr
        llvm.store %35, %37 : i64, !llvm.ptr
        cf.br ^bb9
        ^bb9:
        %38 = llvm.load %37 : !llvm.ptr -> i64
        %39 = arith.cmpi slt, %38, %arg0 : i64
        cf.cond_br %39, ^bb10, ^bb11
        ^bb10:
          %40 = arith.constant 1 : i32
          %41 = llvm.load %37 : !llvm.ptr -> i64
          %42 = arith.trunci %40 : i32 to i8
          %43 = llvm.getelementptr %0[%41] : (!llvm.ptr, i64) -> !llvm.ptr, i8
          llvm.store %42, %43 : i8, !llvm.ptr
          %44 = llvm.load %37 : !llvm.ptr -> i64
          %45 = llvm.load %22 : !llvm.ptr -> i64
          %46 = arith.addi %44, %45 : i64
          llvm.store %46, %37 : i64, !llvm.ptr
          cf.br ^bb9
        ^bb11:
        cf.br ^bb8
      ^bb7:
        cf.br ^bb8
      ^bb8:
      %47 = llvm.load %22 : !llvm.ptr -> i64
      %48 = arith.constant 1 : i32
      %50 = arith.extsi %48 : i32 to i64
      %49 = arith.addi %47, %50 : i64
      llvm.store %49, %22 : i64, !llvm.ptr
      cf.br ^bb3
    ^bb5:
    %51 = arith.constant 0 : i32
    %52 = arith.extsi %51 : i32 to i64
    %53 = llvm.mlir.constant(1 : i64) : i64
    %54 = llvm.alloca %53 x i64 : (i64) -> !llvm.ptr
    llvm.store %52, %54 : i64, !llvm.ptr
    %55 = arith.constant 2 : i32
    %56 = arith.index_cast %55 : i32 to index
    %57 = arith.index_cast %arg0 : i32 to index
    %59 = arith.constant 1 : index
    %60 = arith.constant -1 : index
    %61 = arith.cmpi sle, %56, %57 : index
    %58 = arith.select %61, %59, %60 : index
    cf.br ^bb12(%56 : index)
    ^bb12(%62: index):
    %63 = arith.cmpi slt, %62, %57 : index
    %64 = arith.cmpi sgt, %62, %57 : index
    %65 = arith.select %61, %63, %64 : i1
    cf.cond_br %65, ^bb13(%62 : index), ^bb14(%62 : index)
    ^bb13(%66: index):
      %68 = arith.index_cast %66 : index to i64
      %69 = llvm.getelementptr %0[%68] : (!llvm.ptr, i64) -> !llvm.ptr, i8
      %67 = llvm.load %69 : !llvm.ptr -> i8
      %70 = arith.constant 0 : i32
      %72 = arith.extsi %67 : i8 to i32
      %71 = arith.cmpi eq, %72, %70 : i32
      cf.cond_br %71, ^bb15, ^bb16
      ^bb15:
        %73 = llvm.load %54 : !llvm.ptr -> i64
        %75 = arith.trunci %73 : i64 to i32
        %76 = arith.index_cast %66 : index to i32
        %74 = arith.addi %75, %76 : i32
        %77 = arith.extsi %74 : i32 to i64
        llvm.store %77, %54 : i64, !llvm.ptr
        cf.br ^bb17
      ^bb16:
        cf.br ^bb17
      ^bb17:
      %78 = arith.addi %66, %58 : index
      cf.br ^bb12(%78 : index)
    ^bb14(%79: index):
    func.call @free(%0) : (!llvm.ptr) -> ()
    %81 = llvm.load %54 : !llvm.ptr -> i64
    func.return %81 : i64
  }
  func.func @main() -> i32 {
    %82 = llvm.mlir.addressof @str_0 : !llvm.ptr
    %84 = arith.constant 2000000 : i32
    %85 = arith.extsi %84 : i32 to i64
    %83 = func.call @sum_primes_below(%85) : (i64) -> i64
    %86 = llvm.call @printf(%82, %83) vararg(!llvm.func<i32 (ptr, ...)>) : (!llvm.ptr, i64) -> i32
    %87 = arith.constant 0 : i32
    func.return %87 : i32
  }
}