capi预测出core，当样本是dense_vector_sequence类型时，请教正确的调用方式

[fsfszongming256]

问题：当输入样本是sequence，sequence中每一个元素是定长向量时，输入参数如何构造？sequence_start_pos如何构造？ 看capi的样例代码没有找到摸索到答案

输入样本类型：
dense_vector_sequence(64)

相关代码：

        paddle_gradient_machine thread_local_machine;
        CHECK(paddle_gradient_machine_create_shared_param(_machine, _config_bin, _size, &thread_local_machine));
        gettimeofday(&tv_begin, NULL);
        paddle_arguments in_args = paddle_arguments_create_none();
        CHECK(paddle_arguments_resize(in_args, 1));
        //paddle_matrix sentence = paddle_matrix_create(count, 64, useGPU);
        paddle_matrix sentence = paddle_matrix_create(1, count * 64, useGPU);
        for (int i = 0; i < count; ++i) {
            CHECK(paddle_matrix_set_row(sentence, i, sep[i]));
        }    
        CHECK(paddle_arguments_set_value(in_args, 0, sentence));
        uint64_t h = 0;
        uint64_t w = 0;
        CHECK(paddle_matrix_get_shape(sentence, &h, &w));
        fprintf(stderr, "shape[%lu,%lu]\n", h, w);
        // paddle input matrix end

        int seq_pos_array[] = {0, count};
        paddle_ivector seq_pos = paddle_ivector_create(seq_pos_array, count, false, false);
        CHECK(paddle_arguments_set_sequence_start_pos(in_args, 0, 0, seq_pos));
        // paddle sequence pos info end

        paddle_arguments out_args = paddle_arguments_create_none();
        CHECK(paddle_gradient_machine_forward(thread_local_machine, in_args, out_args, false)); //predict end

出core日志：
I0908 11:01:07.981760 22947 Util.cpp:166] commandline:
I0908 11:01:07.983686 22947 GradientMachine.cpp:74] Loading parameters from ./model/param
shape[9,64]
F0908 11:01:07.984719 22947 Matrix.h:262] Check failed: startRow + numRows <= getHeight() (10 vs. 9)
*** Check failure stack trace: ***
@ 0x7f3a3041287d google::LogMessage::Fail()
@ 0x7f3a3041632c google::LogMessage::SendToLog()
@ 0x7f3a304123a3 google::LogMessage::Flush()
@ 0x7f3a3041783e google::LogMessageFatal::~LogMessageFatal()
@ 0x7f3a302ad461 paddle::Matrix::subMatrix()
@ 0x7f3a30297746 paddle::ContextProjectionForward<>()
@ 0x7f3a3029b642 paddle::ContextProjectionForwardFunc<>::calc()
@ 0x7f3a30157c0e paddle::ContextProjection::forward()
@ 0x7f3a301ed41e paddle::MixedLayer::forward()
@ 0x7f3a30289530 paddle::NeuralNetwork::forward()
@ 0x7f3a301404e6 paddle_gradient_machine_forward
@ 0x40321d main
@ 0x7f3a2f16dbd5 __libc_start_main
@ 0x403e99 (unknown)
@ (nil) (unknown)
Aborted (core dumped)

gdb信息：

#0  0x00007f3a2f1813f7 in raise () from /home/opt/gcc-4.8.2.bpkg-r2/gcc-4.8.2.bpkg-r2/lib64/libc-2.18.so
#1  0x00007f3a2f1827d8 in abort () from /home/opt/gcc-4.8.2.bpkg-r2/gcc-4.8.2.bpkg-r2/lib64/libc-2.18.so
#2  0x00007f3a3041b715 in google::DumpStackTraceAndExit() () from /home/img/zongming/rs/dump_embdding_capi/libpaddle_capi_shared.so
#3  0x00007f3a3041287d in google::LogMessage::Fail() () from /home/img/zongming/rs/dump_embdding_capi/libpaddle_capi_shared.so
#4  0x00007f3a3041632c in google::LogMessage::SendToLog() () from /home/img/zongming/rs/dump_embdding_capi/libpaddle_capi_shared.so
#5  0x00007f3a304123a3 in google::LogMessage::Flush() () from /home/img/zongming/rs/dump_embdding_capi/libpaddle_capi_shared.so
#6  0x00007f3a3041783e in google::LogMessageFatal::~LogMessageFatal() ()
   from /home/img/zongming/rs/dump_embdding_capi/libpaddle_capi_shared.so
#7  0x00007f3a302ad461 in paddle::Matrix::subMatrix (this=0x7fff8f1541e0, startRow=9, numRows=1)
    at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/math/Matrix.h:262
#8  0x00007f3a30297746 in paddle::ContextProjectionForward<(paddle::DeviceType)1> (out_mat=..., input_mat=..., weight_mat=..., 
    seq_vec=..., context_length=3, context_start=-1, begin_pad=1)
    at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/function/ContextProjectionOp.cpp:43
#9  0x00007f3a3029b642 in paddle::ContextProjectionForwardFunc<(paddle::DeviceType)1>::calc (this=0x4718f9b0, inputs=..., 
    outputs=...) at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/function/ContextProjectionOp.cpp:140
#10 0x00007f3a30157c0e in paddle::ContextProjection::forward (this=0x47190200)
    at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/gserver/layers/ContextProjection.cpp:130
#11 0x00007f3a301ed41e in forward (passType=paddle::enumeration_wrapper::PASS_TEST, out=0x4718b878, in=<optimized out>, 
    this=<optimized out>) at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/gserver/layers/Projection.h:69
#12 paddle::MixedLayer::forward (this=0x4718b6a0, passType=paddle::enumeration_wrapper::PASS_TEST)
    at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/gserver/layers/MixedLayer.cpp:126
#13 0x00007f3a30289530 in paddle::NeuralNetwork::forward (this=0x47181210, inArgs=..., outArgs=0x61ed98, 
    passType=paddle::enumeration_wrapper::PASS_TEST)
    at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/gserver/gradientmachines/NeuralNetwork.cpp:250
#14 0x00007f3a301404e6 in paddle_gradient_machine_forward (machine=<optimized out>, inArgs=<optimized out>, 
    outArgs=<optimized out>, isTrain=<optimized out>)

[QiJune]

请检查一下输入的设置
例如，对于长度为3的dense_vector_sequence [ [1,2,3], [4,5,6], [7 8 9] ]，那么就应该设置seq_pos_array为 [0, 3, 6, 9]

[fsfszongming256]

第一个问题已解决，又出现新问题:
cpu版本 预测结果符合预期
gpu版本 出core

I0911 01:17:28.140764 22298 Util.cpp:166] commandline:  
I0911 01:17:28.142509 22298 GradientMachine.cpp:74] Loading parameters from ./model/param
F0911 01:17:28.143321 22298 Allocator.h:83] Check failed: ptr Fail to allocate GPU memory 3072 bytes
*** Check failure stack trace: ***
    @     0x7fe0d31c887d  google::LogMessage::Fail()
    @     0x7fe0d31cc32c  google::LogMessage::SendToLog()
    @     0x7fe0d31c83a3  google::LogMessage::Flush()
    @     0x7fe0d31cd83e  google::LogMessageFatal::~LogMessageFatal()
    @     0x7fe0d313820e  paddle::GpuAllocator::alloc()
    @     0x7fe0d31372e6  paddle::PoolAllocator::alloc()
    @     0x7fe0d31360af  paddle::GpuMemoryHandle::GpuMemoryHandle()
    @     0x7fe0d3117b81  paddle::GpuMatrix::GpuMatrix()
    @     0x7fe0d311842b  paddle::Matrix::create()
    @     0x7fe0d2ef7dd0  paddle_matrix_create
    @           0x40324e  main
    @     0x7fe0d1f23bd5  __libc_start_main
    @           0x404009  (unknown)
    @              (nil)  (unknown)
Aborted (core dumped)

机器上有gpu，python训练 和 预测都正常

[QiJune]

如果需要GPU预测，那么capi需要使用GPU版本的paddle，请检查编译的版本是否正确