Mercurial > hg > truffle
annotate src/gpu/ptx/vm/gpu_ptx.cpp @ 11634:bff2b88444f5
Start of PTX array passing
| author | Morris Meyer <morris.meyer@oracle.com> |
|---|---|
| date | Sat, 14 Sep 2013 17:31:25 -0400 |
| parents | 4f69a5189e77 |
| children | d8659ad83fcc |
| rev | line source |
|---|---|
| 10879 | 1 /* |
| 2 * Copyright (c) 2013, Oracle and/or its affiliates. All rights reserved. | |
| 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
| 4 * | |
| 5 * This code is free software; you can redistribute it and/or modify it | |
| 6 * under the terms of the GNU General Public License version 2 only, as | |
| 7 * published by the Free Software Foundation. | |
| 8 * | |
| 9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
| 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 12 * version 2 for more details (a copy is included in the LICENSE file that | |
| 13 * accompanied this code). | |
| 14 * | |
| 15 * You should have received a copy of the GNU General Public License version | |
| 16 * 2 along with this work; if not, write to the Free Software Foundation, | |
| 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
| 18 * | |
| 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA | |
| 20 * or visit www.oracle.com if you need additional information or have any | |
| 21 * questions. | |
| 22 * | |
| 23 */ | |
| 24 | |
| 25 #include "precompiled.hpp" | |
|
11284
d876002b98e6
ptx: fix include for non-precompiledheader build in gpu_ptx.cpp
Bernhard Urban <bernhard.urban@jku.at>
parents:
11283
diff
changeset
|
26 #include "runtime/javaCalls.hpp" |
| 10879 | 27 #include "runtime/gpu.hpp" |
| 28 #include "utilities/globalDefinitions.hpp" | |
| 29 #include "utilities/ostream.hpp" | |
|
10883
5fcb30bcb90a
Fix compilation without precompiled headers.
Roland Schatz <roland.schatz@oracle.com>
parents:
10882
diff
changeset
|
30 #include "memory/allocation.hpp" |
|
5fcb30bcb90a
Fix compilation without precompiled headers.
Roland Schatz <roland.schatz@oracle.com>
parents:
10882
diff
changeset
|
31 #include "memory/allocation.inline.hpp" |
|
11596
91e5f927af63
Initial implementation of PTXRuntime (RegisterConfig, PTX description etc); guarded with new flag UseGPU. Specify -XX:+UseGPU to exercise this new implementation.
bharadwaj
parents:
11527
diff
changeset
|
32 #include "ptxKernelArguments.hpp" |
| 10879 | 33 |
| 34 void * gpu::Ptx::_device_context; | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
35 int gpu::Ptx::_cu_device = 0; |
| 10879 | 36 |
| 37 gpu::Ptx::cuda_cu_init_func_t gpu::Ptx::_cuda_cu_init; | |
| 38 gpu::Ptx::cuda_cu_ctx_create_func_t gpu::Ptx::_cuda_cu_ctx_create; | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
39 gpu::Ptx::cuda_cu_ctx_destroy_func_t gpu::Ptx::_cuda_cu_ctx_destroy; |
| 10879 | 40 gpu::Ptx::cuda_cu_ctx_synchronize_func_t gpu::Ptx::_cuda_cu_ctx_synchronize; |
|
11527
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
41 gpu::Ptx::cuda_cu_ctx_set_current_func_t gpu::Ptx::_cuda_cu_ctx_set_current; |
| 10879 | 42 gpu::Ptx::cuda_cu_device_get_count_func_t gpu::Ptx::_cuda_cu_device_get_count; |
| 43 gpu::Ptx::cuda_cu_device_get_name_func_t gpu::Ptx::_cuda_cu_device_get_name; | |
| 44 gpu::Ptx::cuda_cu_device_get_func_t gpu::Ptx::_cuda_cu_device_get; | |
| 45 gpu::Ptx::cuda_cu_device_compute_capability_func_t gpu::Ptx::_cuda_cu_device_compute_capability; | |
| 46 gpu::Ptx::cuda_cu_device_get_attribute_func_t gpu::Ptx::_cuda_cu_device_get_attribute; | |
| 47 gpu::Ptx::cuda_cu_launch_kernel_func_t gpu::Ptx::_cuda_cu_launch_kernel; | |
| 48 gpu::Ptx::cuda_cu_module_get_function_func_t gpu::Ptx::_cuda_cu_module_get_function; | |
| 49 gpu::Ptx::cuda_cu_module_load_data_ex_func_t gpu::Ptx::_cuda_cu_module_load_data_ex; | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
50 gpu::Ptx::cuda_cu_memcpy_dtoh_func_t gpu::Ptx::_cuda_cu_memcpy_dtoh; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
51 gpu::Ptx::cuda_cu_memfree_func_t gpu::Ptx::_cuda_cu_memfree; |
| 10879 | 52 |
| 53 bool gpu::Ptx::initialize_gpu() { | |
| 54 | |
| 55 /* Initialize CUDA driver API */ | |
| 56 int status = _cuda_cu_init(0); | |
| 57 if (status != GRAAL_CUDA_SUCCESS) { | |
| 58 tty->print_cr("Failed to initialize CUDA device"); | |
| 59 return false; | |
| 60 } | |
|
11527
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
61 |
| 10879 | 62 if (TraceGPUInteraction) { |
| 63 tty->print_cr("CUDA driver initialization: Success"); | |
| 64 } | |
| 65 | |
| 66 /* Get the number of compute-capable device count */ | |
| 67 int device_count = 0; | |
| 68 status = _cuda_cu_device_get_count(&device_count); | |
| 69 if (status != GRAAL_CUDA_SUCCESS) { | |
| 70 tty->print_cr("[CUDA] Failed to get compute-capable device count"); | |
| 71 return false; | |
| 72 } | |
| 73 | |
| 74 if (device_count == 0) { | |
| 75 tty->print_cr("[CUDA] Found no device supporting CUDA"); | |
| 76 return false; | |
| 77 } | |
| 78 | |
| 79 if (TraceGPUInteraction) { | |
| 80 tty->print_cr("[CUDA] Number of compute-capable devices found: %d", device_count); | |
| 81 } | |
|
11527
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
82 |
| 10879 | 83 /* Get the handle to the first compute device */ |
| 84 int device_id = 0; | |
| 85 /* Compute-capable device handle */ | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
86 status = _cuda_cu_device_get(&_cu_device, device_id); |
| 10879 | 87 |
| 88 if (status != GRAAL_CUDA_SUCCESS) { | |
| 89 tty->print_cr("[CUDA] Failed to get handle of first compute-capable device i.e., the one at ordinal: %d", device_id); | |
| 90 return false; | |
| 91 } | |
| 92 | |
| 93 if (TraceGPUInteraction) { | |
| 94 tty->print_cr("[CUDA] Got the handle of first compute-device"); | |
| 95 } | |
| 96 | |
| 97 /* Get device attributes */ | |
|
11283
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
98 int minor, major, unified_addressing; |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
99 status = _cuda_cu_device_get_attribute(&minor, GRAAL_CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, _cu_device); |
| 10879 | 100 |
| 101 if (status != GRAAL_CUDA_SUCCESS) { | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
102 tty->print_cr("[CUDA] Failed to get minor attribute of device: %d", _cu_device); |
| 10879 | 103 return false; |
| 104 } | |
| 105 | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
106 status = _cuda_cu_device_get_attribute(&major, GRAAL_CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, _cu_device); |
| 10879 | 107 |
| 108 if (status != GRAAL_CUDA_SUCCESS) { | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
109 tty->print_cr("[CUDA] Failed to get major attribute of device: %d", _cu_device); |
| 10879 | 110 return false; |
| 111 } | |
| 112 | |
| 113 if (TraceGPUInteraction) { | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
114 tty->print_cr("[CUDA] Compatibility version of device %d: %d.%d", _cu_device, major, minor); |
| 10879 | 115 } |
| 116 | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
117 status = _cuda_cu_device_get_attribute(&unified_addressing, GRAAL_CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING, _cu_device); |
|
11283
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
118 |
|
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
119 if (status != GRAAL_CUDA_SUCCESS) { |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
120 tty->print_cr("[CUDA] Failed to query unified addressing mode of device: %d", _cu_device); |
|
11283
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
121 return false; |
|
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
122 } |
|
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
123 |
|
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
124 if (TraceGPUInteraction) { |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
125 tty->print_cr("[CUDA] Unified addressing support on device %d: %d", _cu_device, unified_addressing); |
|
11283
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
126 } |
|
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
127 |
|
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
128 |
| 10879 | 129 /* Get device name */ |
| 130 char device_name[256]; | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
131 status = _cuda_cu_device_get_name(device_name, 256, _cu_device); |
| 10879 | 132 |
| 133 if (status != GRAAL_CUDA_SUCCESS) { | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
134 tty->print_cr("[CUDA] Failed to get name of device: %d", _cu_device); |
| 10879 | 135 return false; |
| 136 } | |
| 137 | |
| 138 if (TraceGPUInteraction) { | |
| 139 tty->print_cr("[CUDA] Using %s", device_name); | |
| 140 } | |
| 141 | |
| 142 return true; | |
| 143 } | |
| 144 | |
| 145 void *gpu::Ptx::generate_kernel(unsigned char *code, int code_len, const char *name) { | |
| 146 | |
| 147 struct CUmod_st * cu_module; | |
| 148 // Use three JIT compiler options | |
| 149 const unsigned int jit_num_options = 3; | |
| 150 int *jit_options = NEW_C_HEAP_ARRAY(int, jit_num_options, mtCompiler); | |
| 151 void **jit_option_values = NEW_C_HEAP_ARRAY(void *, jit_num_options, mtCompiler); | |
| 152 | |
| 153 // Set up PTX JIT compiler options | |
| 154 // 1. set size of compilation log buffer | |
| 155 int jit_log_buffer_size = 1024; | |
| 156 jit_options[0] = GRAAL_CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES; | |
| 157 jit_option_values[0] = (void *)(size_t)jit_log_buffer_size; | |
| 158 | |
| 159 // 2. set pointer to compilation log buffer | |
| 160 char *jit_log_buffer = NEW_C_HEAP_ARRAY(char, jit_log_buffer_size, mtCompiler); | |
| 161 jit_options[1] = GRAAL_CU_JIT_INFO_LOG_BUFFER; | |
| 162 jit_option_values[1] = jit_log_buffer; | |
| 163 | |
| 164 // 3. set pointer to set the Maximum # of registers (32) for the kernel | |
| 165 int jit_register_count = 32; | |
| 166 jit_options[2] = GRAAL_CU_JIT_MAX_REGISTERS; | |
| 167 jit_option_values[2] = (void *)(size_t)jit_register_count; | |
| 168 | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
169 /* Create CUDA context to compile and execute the kernel */ |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
170 int status = _cuda_cu_ctx_create(&_device_context, 0, _cu_device); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
171 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
172 if (status != GRAAL_CUDA_SUCCESS) { |
|
11634
bff2b88444f5
Start of PTX array passing
Morris Meyer <morris.meyer@oracle.com>
parents:
11606
diff
changeset
|
173 tty->print_cr("[CUDA] Failed to create CUDA context for device(%d): %d", _cu_device, status); |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
174 return NULL; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
175 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
176 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
177 if (TraceGPUInteraction) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
178 tty->print_cr("[CUDA] Success: Created context for device: %d", _cu_device); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
179 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
180 |
|
11527
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
181 status = _cuda_cu_ctx_set_current(_device_context); |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
182 |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
183 if (status != GRAAL_CUDA_SUCCESS) { |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
184 tty->print_cr("[CUDA] Failed to set current context for device: %d", _cu_device); |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
185 return NULL; |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
186 } |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
187 |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
188 if (TraceGPUInteraction) { |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
189 tty->print_cr("[CUDA] Success: Set current context for device: %d", _cu_device); |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
190 } |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
191 |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
192 if (TraceGPUInteraction) { |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
193 tty->print_cr("[CUDA] PTX Kernel\n%s", code); |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
194 tty->print_cr("[CUDA] Function name : %s", name); |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
195 |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
196 } |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
197 |
| 10879 | 198 /* Load module's data with compiler options */ |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
199 status = _cuda_cu_module_load_data_ex(&cu_module, (void*) code, jit_num_options, |
| 10879 | 200 jit_options, (void **)jit_option_values); |
| 201 if (status != GRAAL_CUDA_SUCCESS) { | |
| 202 if (status == GRAAL_CUDA_ERROR_NO_BINARY_FOR_GPU) { | |
| 203 tty->print_cr("[CUDA] Check for malformed PTX kernel or incorrect PTX compilation options"); | |
| 204 } | |
|
11527
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
205 tty->print_cr("[CUDA] *** Error (%d) Failed to load module data with online compiler options for method %s", |
| 10879 | 206 status, name); |
| 207 return NULL; | |
| 208 } | |
| 209 | |
| 210 if (TraceGPUInteraction) { | |
| 211 tty->print_cr("[CUDA] Loaded data for PTX Kernel"); | |
| 212 } | |
| 213 | |
| 214 struct CUfunc_st * cu_function; | |
| 215 | |
| 216 status = _cuda_cu_module_get_function(&cu_function, cu_module, name); | |
| 217 | |
| 218 if (status != GRAAL_CUDA_SUCCESS) { | |
|
11283
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
219 tty->print_cr("[CUDA] *** Error: Failed to get function %s", name); |
| 10879 | 220 return NULL; |
| 221 } | |
| 222 | |
| 223 if (TraceGPUInteraction) { | |
| 224 tty->print_cr("[CUDA] Got function handle for %s", name); | |
| 225 } | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
226 |
| 10879 | 227 return cu_function; |
| 228 } | |
| 229 | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
230 bool gpu::Ptx::execute_kernel(address kernel, PTXKernelArguments &ptxka, JavaValue &ret) { |
| 10879 | 231 // grid dimensionality |
| 232 unsigned int gridX = 1; | |
| 233 unsigned int gridY = 1; | |
| 234 unsigned int gridZ = 1; | |
| 235 | |
| 236 // thread dimensionality | |
| 237 unsigned int blockX = 1; | |
| 238 unsigned int blockY = 1; | |
| 239 unsigned int blockZ = 1; | |
|
11527
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
240 |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
241 struct CUfunc_st* cu_function = (struct CUfunc_st*) kernel; |
| 10879 | 242 |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
243 void * config[5] = { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
244 GRAAL_CU_LAUNCH_PARAM_BUFFER_POINTER, ptxka._kernelArgBuffer, |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
245 GRAAL_CU_LAUNCH_PARAM_BUFFER_SIZE, &(ptxka._bufferOffset), |
|
11283
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
246 GRAAL_CU_LAUNCH_PARAM_END |
|
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
247 }; |
|
1cd1f8ff70a1
CR-20: PTX kernel invocation with arguments - from Bharadwaj
Morris Meyer <morris.meyer@oracle.com>
parents:
10884
diff
changeset
|
248 |
| 10879 | 249 if (kernel == NULL) { |
| 250 return false; | |
| 251 } | |
| 252 | |
| 253 if (TraceGPUInteraction) { | |
| 254 tty->print_cr("[CUDA] launching kernel"); | |
| 255 } | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
256 |
| 10879 | 257 int status = _cuda_cu_launch_kernel(cu_function, |
| 258 gridX, gridY, gridZ, | |
| 259 blockX, blockY, blockZ, | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
260 0, NULL, NULL, (void **) &config); |
| 10879 | 261 if (status != GRAAL_CUDA_SUCCESS) { |
| 262 tty->print_cr("[CUDA] Failed to launch kernel"); | |
| 263 return false; | |
| 264 } | |
| 265 | |
| 266 if (TraceGPUInteraction) { | |
| 267 tty->print_cr("[CUDA] Success: Kernel Launch"); | |
| 268 } | |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
269 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
270 status = _cuda_cu_ctx_synchronize(); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
271 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
272 if (status != GRAAL_CUDA_SUCCESS) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
273 tty->print_cr("[CUDA] Failed to synchronize launched kernel (%d)", status); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
274 return false; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
275 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
276 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
277 if (TraceGPUInteraction) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
278 tty->print_cr("[CUDA] Success: Synchronized launch kernel"); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
279 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
280 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
281 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
282 // Get the result. TODO: Move this code to get_return_oop() |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
283 BasicType return_type = ptxka.get_ret_type(); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
284 switch (return_type) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
285 case T_INT : |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
286 { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
287 int return_val; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
288 status = gpu::Ptx::_cuda_cu_memcpy_dtoh(&return_val, ptxka._return_value_ptr, T_INT_BYTE_SIZE); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
289 if (status != GRAAL_CUDA_SUCCESS) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
290 tty->print_cr("[CUDA] *** Error (%d) Failed to copy value to device argument", status); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
291 return false; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
292 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
293 ret.set_jint(return_val); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
294 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
295 break; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
296 case T_LONG : |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
297 { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
298 long return_val; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
299 status = gpu::Ptx::_cuda_cu_memcpy_dtoh(&return_val, ptxka._return_value_ptr, T_LONG_BYTE_SIZE); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
300 if (status != GRAAL_CUDA_SUCCESS) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
301 tty->print_cr("[CUDA] *** Error (%d) Failed to copy value to device argument", status); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
302 return false; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
303 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
304 ret.set_jlong(return_val); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
305 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
306 break; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
307 default: |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
308 tty->print_cr("[CUDA] TODO *** Unhandled return type"); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
309 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
310 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
311 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
312 // Free device memory allocated for result |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
313 status = gpu::Ptx::_cuda_cu_memfree(ptxka._return_value_ptr); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
314 if (status != GRAAL_CUDA_SUCCESS) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
315 tty->print_cr("[CUDA] *** Error (%d) Failed to free device memory of return value", status); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
316 return false; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
317 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
318 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
319 if (TraceGPUInteraction) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
320 tty->print_cr("[CUDA] Success: Freed device memory of return value"); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
321 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
322 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
323 // Destroy context |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
324 status = gpu::Ptx::_cuda_cu_ctx_destroy(_device_context); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
325 if (status != GRAAL_CUDA_SUCCESS) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
326 tty->print_cr("[CUDA] *** Error (%d) Failed to destroy context", status); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
327 return false; |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
328 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
329 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
330 if (TraceGPUInteraction) { |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
331 tty->print_cr("[CUDA] Success: Destroy context"); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
332 } |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
333 |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
334 return (status == GRAAL_CUDA_SUCCESS); |
| 10879 | 335 } |
| 336 | |
| 337 #if defined(LINUX) | |
| 338 static const char cuda_library_name[] = "libcuda.so"; | |
| 339 #elif defined(__APPLE__) | |
| 340 static char const cuda_library_name[] = "/usr/local/cuda/lib/libcuda.dylib"; | |
| 341 #else | |
| 342 static char const cuda_library_name[] = ""; | |
| 343 #endif | |
| 344 | |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
345 #define STD_BUFFER_SIZE 1024 |
|
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
346 |
| 10879 | 347 bool gpu::Ptx::probe_linkage() { |
| 348 if (cuda_library_name != NULL) { | |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
349 char *buffer = (char*)malloc(STD_BUFFER_SIZE); |
|
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
350 void *handle = os::dll_load(cuda_library_name, buffer, STD_BUFFER_SIZE); |
|
11527
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
351 free(buffer); |
| 10879 | 352 if (handle != NULL) { |
| 353 _cuda_cu_init = | |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
354 CAST_TO_FN_PTR(cuda_cu_init_func_t, os::dll_lookup(handle, "cuInit")); |
| 10879 | 355 _cuda_cu_ctx_create = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
356 CAST_TO_FN_PTR(cuda_cu_ctx_create_func_t, os::dll_lookup(handle, "cuCtxCreate")); |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
357 _cuda_cu_ctx_destroy = |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
358 CAST_TO_FN_PTR(cuda_cu_ctx_destroy_func_t, os::dll_lookup(handle, "cuCtxDestroy")); |
| 10879 | 359 _cuda_cu_ctx_synchronize = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
360 CAST_TO_FN_PTR(cuda_cu_ctx_synchronize_func_t, os::dll_lookup(handle, "cuCtxSynchronize")); |
|
11527
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
361 _cuda_cu_ctx_set_current = |
|
c99e65785936
Improvements to PTX codegen; allows more PTX tests that run on the device to pass.
bharadwaj
parents:
11485
diff
changeset
|
362 CAST_TO_FN_PTR(cuda_cu_ctx_set_current_func_t, os::dll_lookup(handle, "cuCtxSetCurrent")); |
| 10879 | 363 _cuda_cu_device_get_count = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
364 CAST_TO_FN_PTR(cuda_cu_device_get_count_func_t, os::dll_lookup(handle, "cuDeviceGetCount")); |
| 10879 | 365 _cuda_cu_device_get_name = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
366 CAST_TO_FN_PTR(cuda_cu_device_get_name_func_t, os::dll_lookup(handle, "cuDeviceGetName")); |
| 10879 | 367 _cuda_cu_device_get = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
368 CAST_TO_FN_PTR(cuda_cu_device_get_func_t, os::dll_lookup(handle, "cuDeviceGet")); |
| 10879 | 369 _cuda_cu_device_compute_capability = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
370 CAST_TO_FN_PTR(cuda_cu_device_compute_capability_func_t, os::dll_lookup(handle, "cuDeviceComputeCapability")); |
| 10879 | 371 _cuda_cu_device_get_attribute = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
372 CAST_TO_FN_PTR(cuda_cu_device_get_attribute_func_t, os::dll_lookup(handle, "cuDeviceGetAttribute")); |
| 10879 | 373 _cuda_cu_module_get_function = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
374 CAST_TO_FN_PTR(cuda_cu_module_get_function_func_t, os::dll_lookup(handle, "cuModuleGetFunction")); |
| 10879 | 375 _cuda_cu_module_load_data_ex = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
376 CAST_TO_FN_PTR(cuda_cu_module_load_data_ex_func_t, os::dll_lookup(handle, "cuModuleLoadDataEx")); |
| 10879 | 377 _cuda_cu_launch_kernel = |
|
10882
6a2d65cb5d7d
fix: gpu_ptx.cpp did not compile on windows
Christian Wirth <christian.wirth@oracle.com>
parents:
10879
diff
changeset
|
378 CAST_TO_FN_PTR(cuda_cu_launch_kernel_func_t, os::dll_lookup(handle, "cuLaunchKernel")); |
|
11485
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
379 _cuda_cu_memalloc = |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
380 CAST_TO_FN_PTR(cuda_cu_memalloc_func_t, os::dll_lookup(handle, "cuMemAlloc")); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
381 _cuda_cu_memfree = |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
382 CAST_TO_FN_PTR(cuda_cu_memfree_func_t, os::dll_lookup(handle, "cuMemFree")); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
383 _cuda_cu_memcpy_htod = |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
384 CAST_TO_FN_PTR(cuda_cu_memcpy_htod_func_t, os::dll_lookup(handle, "cuMemcpyHtoD")); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
385 _cuda_cu_memcpy_dtoh = |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
386 CAST_TO_FN_PTR(cuda_cu_memcpy_dtoh_func_t, os::dll_lookup(handle, "cuMemcpyDtoH")); |
|
49bb1bc983c6
Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
11284
diff
changeset
|
387 |
| 10879 | 388 if (TraceGPUInteraction) { |
| 389 tty->print_cr("[CUDA] Success: library linkage"); | |
| 390 } | |
| 391 return true; | |
| 392 } else { | |
| 393 // Unable to dlopen libcuda | |
| 394 return false; | |
| 395 } | |
| 396 } else { | |
| 397 tty->print_cr("Unsupported CUDA platform"); | |
| 398 return false; | |
| 399 } | |
| 400 tty->print_cr("Failed to find CUDA linkage"); | |
| 401 return false; | |
| 402 } |