From 3201a6df2c4748ddee05a081128af5ab2cc6435f Mon Sep 17 00:00:00 2001 From: ShuyRoy Date: Sun, 28 Jan 2024 16:36:58 +0800 Subject: [PATCH] add opecnl op func and call --- ggml-opencl.cpp | 250 ++++++++++++++++++++++++++++++++++++++++++++++++ ggml.c | 107 ++++++++++++++++++++- 2 files changed, 354 insertions(+), 3 deletions(-) diff --git a/ggml-opencl.cpp b/ggml-opencl.cpp index 496f9cdca..e79efd72b 100644 --- a/ggml-opencl.cpp +++ b/ggml-opencl.cpp @@ -1338,6 +1338,8 @@ void ggml_cl_free_data(const struct ggml_tensor* tensor) { clReleaseMemObject(mem); } + +//将数据从host端（主机端，即CPU和其内存）复制到device端（设备端，通常是GPU和其内存）上的一个OpenCL缓冲区 static cl_int ggml_cl_h2d_tensor_2d(cl_command_queue queue, cl_mem dst, size_t offset, const struct ggml_tensor * src, uint64_t i3, uint64_t i2, cl_event* ev) { cl_int err; const uint64_t ne0 = src->ne[0]; @@ -1867,6 +1869,254 @@ size_t ggml_cl_mul_mat_get_wsize(const struct ggml_tensor * src0, const struct g return 0; } + + + +// xr +static viod ggml_compute_cl_dup_same_cont( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst){ + +} + +static viod ggml_compute_cl_dup_f16( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst){ + +} + +static viod ggml_compute_cl_dup_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst){ + +} + + + +// xr +static void ggml_cl_softmax(const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + +// wh +static void ggml_compute_cl_abs_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_sgn_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_neg_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_step_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_tanh_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_elu_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_relu_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_gelu_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_gelu_quick_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr +static void ggml_compute_cl_silu_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst) { + +} + +// xr NOP +static void ggml_cl_permute(){ + +} + +// xr NOP +static void ggml_cl_transpose(){ + +} + +// xr NOP +static void ggml_cl_view(const struct ggml_compute_params * params, const struc ggml_tensor * src0, struct ggml_tensor * dst){ + +} + +// xr NOP +static void ggml_cl_reshape(const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst){ + +} + + + +// wsy +static void ggml_cl_rope_f32(const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst, + const bool forward){ + +} + +// wsy +static void ggml_cl_rope_f16(const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst, + const bool forward){ + +} + +// wsy +static void ggml_cl_rms_norm_f32(const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + struct ggml_tensor * dst){ + +} + + + + +// zjl ggml_cl_add +static void ggml_compute_cl_add_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + +// zjl +static void ggml_compute_cl_add_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + +// zjl +static void ggml_compute_cl_add_f16_f16( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + +// zjl +static void ggml_compute_cl_add_f16_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + +// zjl +static void ggml_compute_cl_add_q_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + + + + +// wh ggml_comput_cl_get_rows + +static void ggml_compute_cl_get_rows_q( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + +// wh +static void ggml_compute_cl_get_rows_f16( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + +// wh +static void ggml_compute_cl_get_rows_f32( + const struct ggml_compute_params * params, + const struct ggml_tensor * src0, + const struct ggml_tensor * src1, + struct ggml_tensor * dst){ + +} + + + + + + void ggml_cl_transform_tensor(void * data, ggml_tensor * tensor) { const int64_t ne0 = tensor->ne[0]; const int64_t ne1 = tensor->ne[1]; diff --git a/ggml.c b/ggml.c index 29a2e18a4..f7db08122 100644 --- a/ggml.c +++ b/ggml.c @@ -7115,16 +7115,28 @@ static void ggml_compute_forward_dup( const struct ggml_tensor * src0, struct ggml_tensor * dst) { if (ggml_is_contiguous(src0) && ggml_is_contiguous(dst) && src0->type == dst->type) { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_dup_same_cont(params, src0, dst); + return; + #endif ggml_compute_forward_dup_same_cont(params, src0, dst); return; } switch (src0->type) { case GGML_TYPE_F16: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_dup_f16(params, src0, dst); + return; + #endif ggml_compute_forward_dup_f16(params, src0, dst); } break; case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_dup_f32(params, src0, dst); + return; + #endif ggml_compute_forward_dup_f32(params, src0, dst); } break; default: @@ -7157,6 +7169,11 @@ static void ggml_compute_forward_add_f32( GGML_ASSERT( nb0 == sizeof(float)); GGML_ASSERT(nb00 == sizeof(float)); + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_add_f32(params, src0, src1, dst); + return; + #endif + // rows per thread const int dr = (nr + nth - 1)/nth; @@ -7244,6 +7261,11 @@ static void ggml_compute_forward_add_f16_f32( GGML_ASSERT(nb00 == sizeof(ggml_fp16_t)); + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_add_f16_f32(params, src0, src1, dst); + return; + #endif + // rows per thread const int dr = (nr + nth - 1)/nth; @@ -7315,6 +7337,11 @@ static void ggml_compute_forward_add_f16_f16( GGML_ASSERT( nb0 == sizeof(ggml_fp16_t)); GGML_ASSERT(nb00 == sizeof(ggml_fp16_t)); + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_add_f16_f16(params, src0, src1, dst); + return; + #endif + // rows per thread const int dr = (nr + nth - 1)/nth; @@ -7379,6 +7406,11 @@ static void ggml_compute_forward_add_q_f32( GGML_ASSERT(ggml_is_quantized(src0->type)); GGML_ASSERT(src1->type == GGML_TYPE_F32); + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_add_q_f32(params, src0, src1, dst); + return; + #endif + // rows per thread const int dr = (nr + nth - 1)/nth; @@ -7429,12 +7461,12 @@ static void ggml_compute_forward_add( struct ggml_tensor * dst) { switch (src0->type) { case GGML_TYPE_F32: - { + { ggml_compute_forward_add_f32(params, src0, src1, dst); } break; case GGML_TYPE_F16: { - if (src1->type == GGML_TYPE_F16) { + if (src1->type == GGML_TYPE_F16) { ggml_compute_forward_add_f16_f16(params, src0, src1, dst); } else if (src1->type == GGML_TYPE_F32) { @@ -7455,6 +7487,7 @@ static void ggml_compute_forward_add( case GGML_TYPE_Q5_K: case GGML_TYPE_Q6_K: { + ggml_compute_forward_add_q_f32(params, src0, src1, dst); } break; default: @@ -8765,6 +8798,10 @@ static void ggml_compute_forward_abs( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_abs_f32(params, src0, dst); + return; + #endif ggml_compute_forward_abs_f32(params, src0, dst); } break; default: @@ -8807,6 +8844,10 @@ static void ggml_compute_forward_sgn( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_sgn_f32(params, src0, dst); + return; + #endif ggml_compute_forward_sgn_f32(params, src0, dst); } break; default: @@ -8849,6 +8890,10 @@ static void ggml_compute_forward_neg( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_neg_f32(params, src0, dst); + return; + #endif ggml_compute_forward_neg_f32(params, src0, dst); } break; default: @@ -8891,6 +8936,10 @@ static void ggml_compute_forward_step( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_step_f32(params, src0, dst); + return; + #endif ggml_compute_forward_step_f32(params, src0, dst); } break; default: @@ -8933,6 +8982,10 @@ static void ggml_compute_forward_tanh( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_tanh_f32(params, src0, dst); + return; + #endif ggml_compute_forward_tanh_f32(params, src0, dst); } break; default: @@ -8975,6 +9028,10 @@ static void ggml_compute_forward_elu( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_elu_f32(params, src0, dst); + return; + #endif ggml_compute_forward_elu_f32(params, src0, dst); } break; default: @@ -9017,6 +9074,10 @@ static void ggml_compute_forward_relu( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_relu_f32(params, src0, dst); + return; + #endif ggml_compute_forward_relu_f32(params, src0, dst); } break; default: @@ -9076,6 +9137,10 @@ static void ggml_compute_forward_gelu( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_gelu_f32(params, src0, dst); + return; + #endif ggml_compute_forward_gelu_f32(params, src0, dst); } break; default: @@ -9135,6 +9200,10 @@ static void ggml_compute_forward_gelu_quick( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_gelu_quick_f32(params, src0, dst); + return; + #endif ggml_compute_forward_gelu_quick_f32(params, src0, dst); } break; default: @@ -9194,6 +9263,10 @@ static void ggml_compute_forward_silu( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_silu_f32(params, src0, dst); + return; + #endif ggml_compute_forward_silu_f32(params, src0, dst); } break; default: @@ -9440,7 +9513,11 @@ static void ggml_compute_forward_rms_norm( struct ggml_tensor * dst) { switch (src0->type) { case GGML_TYPE_F32: - { + { + #if defined(GGML_USE_CLBLAST) + ggml_cl_rms_norm_f32(params, src0, dst); + return; + #endif ggml_compute_forward_rms_norm_f32(params, src0, dst); } break; default: @@ -10839,14 +10916,26 @@ static void ggml_compute_forward_get_rows( case GGML_TYPE_Q5_K: case GGML_TYPE_Q6_K: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_get_rows_q(params, src0, src1, dst); + return; + #endif ggml_compute_forward_get_rows_q(params, src0, src1, dst); } break; case GGML_TYPE_F16: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_get_rows_f16(params, src0, src1, dst); + return; + #endif ggml_compute_forward_get_rows_f16(params, src0, src1, dst); } break; case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_compute_cl_get_rows_f32(params, src0, src1, dst); + return; + #endif ggml_compute_forward_get_rows_f32(params, src0, src1, dst); } break; default: @@ -11222,6 +11311,10 @@ static void ggml_compute_forward_soft_max( switch (src0->type) { case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_cl_softmax(params,src0,src1,dst); + return; + #endif ggml_compute_forward_soft_max_f32(params, src0, src1, dst); } break; default: @@ -11946,10 +12039,18 @@ static void ggml_compute_forward_rope( switch (src0->type) { case GGML_TYPE_F16: { + #if defined(GGML_USE_CLBLAST) + ggml_cl_rope_f16(params, src0, src1, dst, true); + return; + #endif ggml_compute_forward_rope_f16(params, src0, src1, dst, true); } break; case GGML_TYPE_F32: { + #if defined(GGML_USE_CLBLAST) + ggml_cl_rope_f32(params, src0, src1, dst, true); + return; + #endif ggml_compute_forward_rope_f32(params, src0, src1, dst, true); } break; default: