Re: [Starpu-devel] CUDA task and GCC plugin

[Mehdi AMINI <mehdi.amini@silkan.com>]

Hi,

> > > I also have an issue related to implicit cpu implementations and -O3
> > > gcc flag which seems to be broken. I'm testing a simple task call with
> > > a CPU and a CUDA implementation. Using -O3 the CPU task is always
> > > executed while without the -O3 flag the CUDA task is correctly
> > > executed. Is it a known issue ?
> >
> > I believe there’s nothing wrong here: the default scheduler (‘eager’)
> > makes non-deterministic scheduling decisions, and just happened to favor
> > CPU over CUDA for some reason.
>
> It seems to be reproducible, I mean 100%
> Moreover StarPU was compiled with --disable-cpu ; and finally I even
> reproduce it using STARPU_NCPUS=0 !
>
> My guess is that -O3 may trigger an inlining of the original implicit
> implementation at call site so that the StarPU wrapper is never called
> and instead there directly the code of the CPU function.


I confirmed it studying the assembly generated for a simple example (see 
attached).

The simple printf in the implicit CPU implementation for the StarPU task 
is inlined and there not call anymore in the main function.

By the way it seems that there is another bug, the plugin segfaults when 
there is a task without any parameter.

Mehdi
#include <stdio.h>


static void launch(double *A) __attribute__ ((task));


static void launch(double *A) {
  printf("CPU");
}

extern void launch_cuda(double *A)
       __attribute__ ((task_implementation ("cuda", launch)));

int main(int argc, char **argv) {
#pragma starpu initialize
  double A[10];
  launch(A);
}
        .file   "test.c"
        .section        .rodata
        .align 8
.LC0:
        .string "test.c:7: error: attempt to use unregistered pointer\n"
        .align 8
.LC1:
        .string "test.c:7: error: failed to insert task `launch': %s\n"
        .text
        .type   launch, @function
launch:
.LFB55:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movq    %rsp, %rbp
        .cfi_def_cfa_register 6
        subq    $32, %rsp
        movq    %rdi, -24(%rbp)
        movq    -24(%rbp), %rax
        movq    %rax, %rdi
        call    starpu_data_lookup
        movq    %rax, -8(%rbp)
        cmpq    $0, -8(%rbp)
        jne     .L2
        movl    $.LC0, %edi
        call    puts
        call    abort
.L2:
        movq    -8(%rbp), %rax
        movl    $0, %ecx
        movq    %rax, %rdx
        movl    $3, %esi
        movl    $launch.codelet, %edi
        movl    $0, %eax
        call    starpu_insert_task
        movl    %eax, -12(%rbp)
        cmpl    $0, -12(%rbp)
        je      .L1
        movl    -12(%rbp), %eax
        negl    %eax
        movl    %eax, %edi
        call    strerror
        movq    %rax, %rsi
        movl    $.LC1, %edi
        movl    $0, %eax
        call    printf
        call    abort
.L1:
        leave
        .cfi_def_cfa 7, 8
        ret
        .cfi_endproc
.LFE55:
        .size   launch, .-launch
        .section        .rodata
.LC2:
        .string "CPU"
        .text
        .type   launch.cpu_implementation, @function
launch.cpu_implementation:
.LFB54:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movq    %rsp, %rbp
        .cfi_def_cfa_register 6
        subq    $16, %rsp
        movq    %rdi, -8(%rbp)
        movl    $.LC2, %eax
        movq    %rax, %rdi
        movl    $0, %eax
        call    printf
        leave
        .cfi_def_cfa 7, 8
        ret
        .cfi_endproc
.LFE54:
        .size   launch.cpu_implementation, .-launch.cpu_implementation
        .section        .rodata
        .align 8
.LC3:
        .string "test.c:15: error: failed to initialize StarPU: %s\n"
        .text
        .globl  main
        .type   main, @function
main:
.LFB56:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movq    %rsp, %rbp
        .cfi_def_cfa_register 6
        subq    $112, %rsp
        movl    %edi, -100(%rbp)
        movq    %rsi, -112(%rbp)
        movl    $0, %edi
        call    starpu_init
        movl    %eax, -4(%rbp)
        cmpl    $0, -4(%rbp)
        je      .L6
        movl    -4(%rbp), %eax
        negl    %eax
        movl    %eax, %edi
        call    strerror
        movq    %rax, %rsi
        movl    $.LC3, %edi
        movl    $0, %eax
        call    printf
        call    abort
.L6:
        leaq    -96(%rbp), %rax
        movq    %rax, %rdi
        call    launch
        leave
        .cfi_def_cfa 7, 8
        ret
        .cfi_endproc
.LFE56:
        .size   main, .-main
        .section        .rodata
.LC4:
        .string "launch"
        .data
        .align 32
        .type   launch.codelet, @object
        .size   launch.codelet, 856
launch.codelet:
        .long   10
        .zero   52
        .quad   launch.cpu_implementation.task_implementation_wrapper
        .quad   0
        .zero   16
        .quad   launch_cuda.task_implementation_wrapper
        .quad   0
        .zero   16
        .quad   0
        .zero   24
        .zero   4
        .long   1
        .long   3
        .zero   28
        .zero   656
        .quad   .LC4
        .text
        .globl  launch_cuda.task_implementation_wrapper
        .type   launch_cuda.task_implementation_wrapper, @function
launch_cuda.task_implementation_wrapper:
.LFB57:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movq    %rsp, %rbp
        .cfi_def_cfa_register 6
        subq    $32, %rsp
        movq    %rdi, -24(%rbp)
        movq    %rsi, -32(%rbp)
        movq    -24(%rbp), %rax
        movq    (%rax), %rax
        movq    (%rax), %rax
        movq    %rax, -8(%rbp)
        movq    -8(%rbp), %rax
        movq    %rax, %rdi
        call    launch_cuda
        leave
        .cfi_def_cfa 7, 8
        ret
        .cfi_endproc
.LFE57:
        .size   launch_cuda.task_implementation_wrapper, 
.-launch_cuda.task_implementation_wrapper
        .type   launch.cpu_implementation.task_implementation_wrapper, 
@function
launch.cpu_implementation.task_implementation_wrapper:
.LFB58:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movq    %rsp, %rbp
        .cfi_def_cfa_register 6
        subq    $32, %rsp
        movq    %rdi, -24(%rbp)
        movq    %rsi, -32(%rbp)
        movq    -24(%rbp), %rax
        movq    (%rax), %rax
        movq    (%rax), %rax
        movq    %rax, -8(%rbp)
        movq    -8(%rbp), %rax
        movq    %rax, %rdi
        call    launch.cpu_implementation
        leave
        .cfi_def_cfa 7, 8
        ret
        .cfi_endproc
.LFE58:
        .size   launch.cpu_implementation.task_implementation_wrapper, 
.-launch.cpu_implementation.task_implementation_wrapper
        .ident  "GCC: (GNU) 4.6.3 20120306 (Red Hat 4.6.3-2)"
        .section        .note.GNU-stack,"",@progbits