annotate src/gpu/ptx/vm/ptxKernelArguments.cpp @ 12653:1a7e7011a341

* PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler. * Change dynamic loading of CUDA driver API functions to load 32-bit or 64-bit versions of depending on the the host architecture. * Add ability to generate PTX kernels to be launched both on 32-bit and 64-bit hosts. * Use Unified Virtual Memory APIs to perform array argument marshalling. * PTX array storage test runs on the device and returns correct results. * More integer test failures on GPU fixed.
author S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
date Fri, 01 Nov 2013 18:34:03 -0400
parents 11b086b1bae4
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
1 /*
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
2 * Copyright (c) 2013, Oracle and/or its affiliates. All rights reserved.
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
4 *
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
5 * This code is free software; you can redistribute it and/or modify it
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
6 * under the terms of the GNU General Public License version 2 only, as
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
7 * published by the Free Software Foundation.
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
8 *
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
9 * This code is distributed in the hope that it will be useful, but WITHOUT
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
12 * version 2 for more details (a copy is included in the LICENSE file that
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
13 * accompanied this code).
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
14 *
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
15 * You should have received a copy of the GNU General Public License version
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
16 * 2 along with this work; if not, write to the Free Software Foundation,
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
18 *
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
20 * or visit www.oracle.com if you need additional information or have any
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
21 * questions.
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
22 *
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
23 */
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
24
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
25 #include "precompiled.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: 11485
diff changeset
26 #include "ptxKernelArguments.hpp"
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
27 #include "runtime/javaCalls.hpp"
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
28
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
29 gpu::Ptx::cuda_cu_memalloc_func_t gpu::Ptx::_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:
diff changeset
30 gpu::Ptx::cuda_cu_memcpy_htod_func_t gpu::Ptx::_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:
diff changeset
31
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
32 // Get next java argument
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
33 oop PTXKernelArguments::next_arg(BasicType expectedType) {
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
34 assert(_index < _args->length(), "out of bounds");
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
35 oop arg = ((objArrayOop) (_args))->obj_at(_index++);
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
36 assert(expectedType == T_OBJECT ||
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
37 java_lang_boxing_object::is_instance(arg, expectedType), "arg type mismatch");
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
38 return arg;
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
39 }
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
40
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
41 /*
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
42 * Pad kernel argument buffer to naturally align for given size.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
43 */
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
44 void PTXKernelArguments::pad_kernel_argument_buffer(size_t dataSz) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
45 while ((_bufferOffset % dataSz) != 0) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
46 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = (char) 0;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
47 _bufferOffset += sizeof(char);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
48 }
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
49 return;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
50 }
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
51 void PTXKernelArguments::do_int() {
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
52 // If the parameter is a return value,
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
53 if (is_return_type()) {
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
54 // Allocate device memory for T_INT return value pointer on device. Size in bytes
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
55 int status = gpu::Ptx::_cuda_cu_memalloc(&_dev_return_value, T_INT_BYTE_SIZE);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
56 if (status != GRAAL_CUDA_SUCCESS) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
57 tty->print_cr("[CUDA] *** Error (%d) Failed to allocate memory for return value pointer on device", status);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
58 _success = false;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
59 return;
11894
c7abc8411011 Fixed BasicPTXTest and IntegerPTXTest
Morris Meyer <morris.meyer@oracle.com>
parents: 11821
diff changeset
60 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
61
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
62 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
63 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
64 pad_kernel_argument_buffer(sizeof(_dev_return_value));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
65 // Push _dev_return_value to _kernelBuffer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
66 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = _dev_return_value;
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
67 _bufferOffset += sizeof(_dev_return_value);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
68 } else {
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
69 // Get the next java argument and its value which should be a T_INT
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
70 oop arg = next_arg(T_INT);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
71 // Copy the java argument value to kernelArgBuffer
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
72 jvalue intval;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
73 if (java_lang_boxing_object::get_value(arg, &intval) != T_INT) {
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
74 tty->print_cr("[CUDA] *** Error: Unexpected argument type; expecting T_INT");
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
75 _success = false;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
76 return;
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
77 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
78
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
79 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
80 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
81 pad_kernel_argument_buffer(sizeof(intval.i));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
82 // Push _dev_return_value to _kernelBuffer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
83 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = intval.i;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
84
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
85 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
86 _bufferOffset += sizeof(intval.i);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
87 }
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
88 return;
11894
c7abc8411011 Fixed BasicPTXTest and IntegerPTXTest
Morris Meyer <morris.meyer@oracle.com>
parents: 11821
diff changeset
89 }
c7abc8411011 Fixed BasicPTXTest and IntegerPTXTest
Morris Meyer <morris.meyer@oracle.com>
parents: 11821
diff changeset
90
c7abc8411011 Fixed BasicPTXTest and IntegerPTXTest
Morris Meyer <morris.meyer@oracle.com>
parents: 11821
diff changeset
91 void PTXKernelArguments::do_float() {
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
92 // If the parameter is a return value,
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
93 if (is_return_type()) {
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
94 // Allocate device memory for T_FLOAT return value pointer on device. Size in bytes
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
95 int status = gpu::Ptx::_cuda_cu_memalloc(&_dev_return_value, T_FLOAT_BYTE_SIZE);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
96 if (status != GRAAL_CUDA_SUCCESS) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
97 tty->print_cr("[CUDA] *** Error (%d) Failed to allocate memory for return value pointer on device", status);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
98 _success = false;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
99 return;
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
100 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
101 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
102 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
103 pad_kernel_argument_buffer(sizeof(_dev_return_value));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
104 // Push _dev_return_value to _kernelBuffer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
105 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = _dev_return_value;
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
106 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
107 _bufferOffset += sizeof(_dev_return_value);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
108 } else {
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
109 // Get the next java argument and its value which should be a T_FLOAT
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
110 oop arg = next_arg(T_FLOAT);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
111 // Copy the java argument value to kernelArgBuffer
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
112 jvalue floatval;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
113 if (java_lang_boxing_object::get_value(arg, &floatval) != T_FLOAT) {
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
114 tty->print_cr("[CUDA] *** Error: Unexpected argument type; expecting T_FLOAT");
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
115 _success = false;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
116 return;
11894
c7abc8411011 Fixed BasicPTXTest and IntegerPTXTest
Morris Meyer <morris.meyer@oracle.com>
parents: 11821
diff changeset
117 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
118 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
119 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
120 pad_kernel_argument_buffer(sizeof(floatval.f));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
121 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = (gpu::Ptx::CUdeviceptr) floatval.f;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
122
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
123 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
124 _bufferOffset += sizeof(floatval.f);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
125 }
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
126 return;
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
127 }
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
128
11902
67a1e27a8dbb PTX initial float and double
Morris Meyer <morris.meyer@oracle.com>
parents: 11901
diff changeset
129 void PTXKernelArguments::do_double() {
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
130 // If the parameter is a return value,
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
131 jvalue doubleval;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
132 if (is_return_type()) {
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
133 // Allocate device memory for T_DOUBLE return value pointer on device. Size in bytes
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
134 int status = gpu::Ptx::_cuda_cu_memalloc(&_dev_return_value, T_DOUBLE_BYTE_SIZE);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
135 if (status != GRAAL_CUDA_SUCCESS) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
136 tty->print_cr("[CUDA] *** Error (%d) Failed to allocate memory for return value pointer on device", status);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
137 _success = false;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
138 return;
11902
67a1e27a8dbb PTX initial float and double
Morris Meyer <morris.meyer@oracle.com>
parents: 11901
diff changeset
139 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
140 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
141 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
142 pad_kernel_argument_buffer(sizeof(_dev_return_value));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
143 // Push _dev_return_value to _kernelBuffer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
144 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = _dev_return_value;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
145 // Advance _bufferOffset.
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
146 _bufferOffset += sizeof(doubleval.d);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
147 } else {
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
148 // Get the next java argument and its value which should be a T_INT
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
149 oop arg = next_arg(T_FLOAT);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
150 // Copy the java argument value to kernelArgBuffer
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
151 if (java_lang_boxing_object::get_value(arg, &doubleval) != T_DOUBLE) {
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
152 tty->print_cr("[CUDA] *** Error: Unexpected argument type; expecting T_INT");
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
153 _success = false;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
154 return;
11902
67a1e27a8dbb PTX initial float and double
Morris Meyer <morris.meyer@oracle.com>
parents: 11901
diff changeset
155 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
156 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
157 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
158 pad_kernel_argument_buffer(sizeof(doubleval.d));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
159 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = (gpu::Ptx::CUdeviceptr) doubleval.d;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
160
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
161 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
162 _bufferOffset += sizeof(doubleval.d);
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
163 // For a 64-bit host, since size of double is 8, there is no need
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
164 // to pad the kernel argument buffer to ensure 8-byte alignment of
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
165 // the next potential argument to be pushed.
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
166 }
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
167 return;
11902
67a1e27a8dbb PTX initial float and double
Morris Meyer <morris.meyer@oracle.com>
parents: 11901
diff changeset
168 }
67a1e27a8dbb PTX initial float and double
Morris Meyer <morris.meyer@oracle.com>
parents: 11901
diff changeset
169
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
170 void PTXKernelArguments::do_long() {
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: 11485
diff changeset
171 // If the parameter is a return value,
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
172 if (is_return_type()) {
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
173 // Allocate device memory for T_LONG return value pointer on device. Size in bytes
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
174 int status = gpu::Ptx::_cuda_cu_memalloc(&_dev_return_value, T_LONG_BYTE_SIZE);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
175 if (status != GRAAL_CUDA_SUCCESS) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
176 tty->print_cr("[CUDA] *** Error (%d) Failed to allocate memory for return value pointer on device", status);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
177 _success = false;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
178 return;
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
179 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
180 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
181 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
182 pad_kernel_argument_buffer(sizeof(_dev_return_value));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
183 // Push _dev_return_value to _kernelBuffer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
184 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = _dev_return_value;
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
185 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
186 _bufferOffset += sizeof(_dev_return_value);
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
187 } else {
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
188 // Get the next java argument and its value which should be a T_LONG
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
189 oop arg = next_arg(T_LONG);
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
190 // Copy the java argument value to kernelArgBuffer
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
191 jvalue val;
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
192 if (java_lang_boxing_object::get_value(arg, &val) != T_LONG) {
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
193 tty->print_cr("[CUDA] *** Error: Unexpected argument type; expecting T_LONG");
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
194 _success = false;
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
195 return;
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
196 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
197 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
198 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
199 pad_kernel_argument_buffer(sizeof(val.j));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
200 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = val.j;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
201
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
202 // Advance _bufferOffset
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
203 _bufferOffset += sizeof(val.j);
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
204 // For a 64-bit host, since size of long is 8, there is no need
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
205 // to pad the kernel argument buffer to ensure 8-byte alignment of
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
206 // the next potential argument to be pushed.
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
207 }
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
208 return;
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
209 }
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
210
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
211 void PTXKernelArguments::do_byte() {
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
212 // If the parameter is a return value,
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
213 if (is_return_type()) {
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
214 // Allocate device memory for T_BYTE return value pointer on device. Size in bytes
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
215 int status = gpu::Ptx::_cuda_cu_memalloc(&_dev_return_value, T_BYTE_SIZE);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
216 if (status != GRAAL_CUDA_SUCCESS) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
217 tty->print_cr("[CUDA] *** Error (%d) Failed to allocate memory for return value pointer on device", status);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
218 _success = false;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
219 return;
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
220 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
221 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
222 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
223 pad_kernel_argument_buffer(sizeof(_dev_return_value));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
224 // Push _dev_return_value to _kernelBuffer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
225 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = _dev_return_value;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
226
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
227 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
228 _bufferOffset += sizeof(_dev_return_value);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
229 } else {
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
230 // Get the next java argument and its value which should be a T_BYTE
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
231 oop arg = next_arg(T_BYTE);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
232 // Copy the java argument value to kernelArgBuffer
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
233 jvalue val;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
234 if (java_lang_boxing_object::get_value(arg, &val) != T_BYTE) {
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
235 tty->print_cr("[CUDA] *** Error: Unexpected argument type; expecting T_BYTE");
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
236 _success = false;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
237 return;
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
238 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
239 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
240 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
241 pad_kernel_argument_buffer(sizeof(val.b));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
242 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = val.b;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
243
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
244 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
245 _bufferOffset += sizeof(val.b);
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
246 // For a 64-bit host, since size of T_BYTE is 8, there is no need
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
247 // to pad the kernel argument buffer to ensure 8-byte alignment of
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
248 // the next potential argument to be pushed.
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
249 }
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
250 return;
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
251 }
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
252
11901
61767ccd4600 PTX boolean return value, emitIntegerTestMove, warnings
Morris Meyer <morris.meyer@oracle.com>
parents: 11894
diff changeset
253 void PTXKernelArguments::do_bool() {
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
254 // If the parameter is a return value,
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
255 if (is_return_type()) {
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
256 // Allocate device memory for T_BYTE return value pointer on device. Size in bytes
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
257 int status = gpu::Ptx::_cuda_cu_memalloc(&_dev_return_value, T_BOOLEAN_SIZE);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
258 if (status != GRAAL_CUDA_SUCCESS) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
259 tty->print_cr("[CUDA] *** Error (%d) Failed to allocate memory for return value pointer on device", status);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
260 _success = false;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
261 return;
11901
61767ccd4600 PTX boolean return value, emitIntegerTestMove, warnings
Morris Meyer <morris.meyer@oracle.com>
parents: 11894
diff changeset
262 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
263 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
264 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
265 pad_kernel_argument_buffer(sizeof(_dev_return_value));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
266 // Push _dev_return_value to _kernelBuffer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
267 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = _dev_return_value;
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
268 _bufferOffset += sizeof(_dev_return_value);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
269 } else {
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
270 // Get the next java argument and its value which should be a T_BOOLEAN
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
271 oop arg = next_arg(T_BOOLEAN);
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
272 // Copy the java argument value to kernelArgBuffer
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
273 jvalue val;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
274 if (java_lang_boxing_object::get_value(arg, &val) != T_BOOLEAN) {
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
275 tty->print_cr("[CUDA] *** Error: Unexpected argument type; expecting T_BOOLEAN");
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
276 _success = false;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
277 return;
11901
61767ccd4600 PTX boolean return value, emitIntegerTestMove, warnings
Morris Meyer <morris.meyer@oracle.com>
parents: 11894
diff changeset
278 }
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
279 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
280 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
281 pad_kernel_argument_buffer(sizeof(val.z));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
282 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = val.z;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
283
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
284 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
285 _bufferOffset += sizeof(val.z);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
286 }
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
287 return;
11901
61767ccd4600 PTX boolean return value, emitIntegerTestMove, warnings
Morris Meyer <morris.meyer@oracle.com>
parents: 11894
diff changeset
288 }
61767ccd4600 PTX boolean return value, emitIntegerTestMove, warnings
Morris Meyer <morris.meyer@oracle.com>
parents: 11894
diff changeset
289
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
290 void PTXKernelArguments::do_array(int begin, int end) {
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
291 // Get the next java argument and its value which should be a T_ARRAY
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
292 oop arg = next_arg(T_OBJECT);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
293 assert(arg->is_array(), "argument value not an array");
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
294 // Size of array argument
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
295 int argSize = arg->size() * HeapWordSize;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
296 // Device pointer to array argument.
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
297 gpu::Ptx::CUdeviceptr arrayArgOnDev;
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
298 int status;
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
299
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
300 // Register host memory for use by the device. Size in bytes
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
301 status = gpu::Ptx::_cuda_cu_mem_host_register(arg, argSize, GRAAL_CU_MEMHOSTREGISTER_DEVICEMAP);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
302 if (status != GRAAL_CUDA_SUCCESS) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
303 tty->print_cr("[CUDA] *** Error (%d) Failed to register host memory for array argument on device",
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
304 status);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
305 _success = false;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
306 return;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
307 }
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
308 // Get device pointer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
309 status = gpu::Ptx::_cuda_cu_mem_host_get_device_pointer(&arrayArgOnDev, arg, 0);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
310 if (status != GRAAL_CUDA_SUCCESS) {
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
311 tty->print_cr("[CUDA] *** Error (%d) Failed to get device pointer of mapped pinned memory of array argument.",
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
312 status);
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
313 _success = false;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
314 return;
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
315 }
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
316
12653
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
317 // Kernel arguments are expected to be naturally aligned.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
318 // Insert padding into kernel argument buffer, if needed.
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
319 pad_kernel_argument_buffer(sizeof(arrayArgOnDev));
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
320 // Push device array argument to _kernelBuffer
1a7e7011a341 * PTX kernel argument buffer now has naturally aligned arguments as required by PTX JIT compiler.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12545
diff changeset
321 *((gpu::Ptx::CUdeviceptr*) &_kernelArgBuffer[_bufferOffset]) = arrayArgOnDev;
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
322
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
323 // Advance _bufferOffset
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
324 _bufferOffset += sizeof(arrayArgOnDev);
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
325 return;
11821
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
326 }
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
327
d8659ad83fcc PTX single-threaded array store, Warp annotation
Morris Meyer <morris.meyer@oracle.com>
parents: 11596
diff changeset
328 void PTXKernelArguments::do_void() {
12519
f020e149c1b6 PTX codegen enhancements; fixes to PTX test regressions.
S.Bharadwaj Yadavalli <bharadwaj.yadavalli@oracle.com>
parents: 12360
diff changeset
329 return;
11485
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
330 }
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
331
49bb1bc983c6 Implement several missing PTX codegen features; return value capture and method args passing of java method executed on GPU.
bharadwaj
parents:
diff changeset
332 // TODO implement other do_*