You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: CHANGELOG.md
+62-31Lines changed: 62 additions & 31 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -2,6 +2,41 @@
2
2
3
3
Full documentation for HIP is available at [rocm.docs.amd.com](https://rocm.docs.amd.com/projects/HIP/en/latest/index.html)
4
4
5
+
## HIP 6.3 for ROCm 6.3
6
+
7
+
### Changed
8
+
9
+
* Un-deprecated HIP APIs
10
+
-`hipHostAlloc`
11
+
-`hipFreeHost`
12
+
13
+
* New HIP APIs
14
+
-`hipGraphExecGetFlags` returns the flags on executable graph.
15
+
-`hipGraphNodeSetParams` updates parameters of a created node.
16
+
-`hipGraphExecNodeSetParams` updates parameters of a created node on executable graph.
17
+
-`hipDrvGraphMemcpyNodeGetParams` gets a memcpy node's parameters.
18
+
-`hipDrvGraphMemcpyNodeSetParams` sets a memcpy node's parameters.
19
+
-`hipDrvGraphAddMemFreeNode` creates a memory free node and adds it to a graph.
20
+
-`hipDrvGraphExecMemcpyNodeSetParams` sets the parameters for a memcpy node in the given graphExec.
21
+
-`hipDrvGraphExecMemsetNodeSetParams` sets the parameters for a memset node in the given graphExec.
22
+
-`hipExtHostAlloc` preserves the functionality of `hipHostMalloc`.
23
+
24
+
### Resolved issues
25
+
26
+
27
+
- The `_sync()` version of crosslane builtins such as `shfl_sync()`,
28
+
`__all_sync()` and `__any_sync()`, continue to be hidden behind the
29
+
preprocessor macro `HIP_ENABLE_WARP_SYNC_BUILTINS`, and will be enabled
30
+
unconditionally in the next ROCm release.
31
+
32
+
## HIP 6.2.41134 for ROCm 6.2.1
33
+
34
+
### Resolved issues
35
+
36
+
* Soft hang when use AMD_SERIALIZE_KERNEL.
37
+
* Memory leak in hipIpcCloseMemHandle.
38
+
39
+
5
40
## HIP 6.2 (For ROCm 6.2)
6
41
7
42
### Added
@@ -14,6 +49,12 @@ Full documentation for HIP is available at [rocm.docs.amd.com](https://rocm.docs
14
49
-`hipGetFuncBySymbol` returns the pointer to device entry function that matches entry function symbolPtr.
15
50
-`hipStreamBeginCaptureToGraph` begins graph capture on a stream to an existing graph.
16
51
-`hipGraphInstantiateWithParams` creates an executable graph from a graph.
52
+
-`hipMemcpyAtoA` copies from one 1D array to another.
53
+
-`hipMemcpyDtoA` copies from device memory to a 1D array.
54
+
-`hipMemcpyAtoD` copies from one 1D array to device memory.
55
+
-`hipMemcpyAtoHAsync` copies from one 1D array to host memory.
56
+
-`hipMemcpyHtoAAsync` copies from host memory to a 1D array.
57
+
-`hipMemcpy2DArrayToArray` copies data between host and device.
17
58
18
59
- Added a new flag `integrated` support in device property
19
60
@@ -24,13 +65,12 @@ Full documentation for HIP is available at [rocm.docs.amd.com](https://rocm.docs
24
65
- Added initial support for 8-bit floating point datatype in `amd_hip_fp8.h`. These are accessible via `#include <hip/hip_fp8.h>`
25
66
- Add UUID support for environment variable `HIP_VISIBLE_DEVICES`.
26
67
27
-
### Fixed
68
+
### Resolved issues
28
69
- Stream capture support in HIP graph.
29
70
Prohibited and unhandled operations are fixed during stream capture in HIP runtime.
30
71
- Fix undefined symbol error for hipTexRefGetArray & hipTexRefGetBorderColor.
31
72
32
73
## HIP 6.1 (For ROCm 6.1)
33
-
### Optimizations
34
74
35
75
### Added
36
76
- New environment variable HIP_LAUNCH_BLOCKING
@@ -41,7 +81,7 @@ The default value is 0 (disable), kernel will execute normally as defined in the
41
81
### Changed
42
82
- HIPRTC now assumes WGP mode for gfx10+. CU mode can be enabled by passing `-mcumode` to the compile options from `hiprtcCompileProgram`.
43
83
44
-
### Fixed
84
+
### Resolved issues
45
85
- HIP complex vector type multiplication and division operations.
46
86
On AMD platform, some duplicated complex operators are removed to avoid compilation failures.
47
87
In HIP, hipFloatComplex and hipDoubleComplex are defined as complex data types,
@@ -53,12 +93,8 @@ Any application uses complex multiplication and division operations, need to rep
53
93
54
94
Note: These complex operations are equivalent to corresponding types/functions on NVIDIA platform.
55
95
56
-
### Known Issues
57
-
58
96
## HIP 6.0 (For ROCm 6.0)
59
97
60
-
### Optimizations
61
-
62
98
### Added
63
99
- Addition of hipExtGetLastError
64
100
- AMD backend specific API, to return error code from last HIP API called from the active host thread
@@ -112,7 +148,7 @@ This header exists alongside the older bfloat16 header in`amd_hip_bfloat16.h` wh
112
148
- hipGraphicsGLRegisterImage
113
149
- With ROCm 6.0, the HIP version is 6.0. As the HIP runtime binary suffix is updated in every major ROCm release, in ROCm 6.0, the new filename is libamdhip64.so.6. Furthermore, in ROCm 6.0 release, the libamdhip64.so.5 binary from ROCm 5.7 is made available to maintain binary backward compatibility with ROCm 5.x.
114
150
115
-
### Changes Impacting Backward Compatibility
151
+
### Changed Impacting Backward Compatibility
116
152
- Data types for members in HIP_MEMCPY3D structure are changed from "unsigned int" to "size_t".
117
153
- The value of the flag hipIpcMemLazyEnablePeerAccess is changed to “0x01”, which was previously defined as “0”.
118
154
- Some device property attributes are not currently support in HIP runtime, in order to maintain consistency, the following related enumeration names are changed in hipDeviceAttribute_t
@@ -125,13 +161,7 @@ This header exists alongside the older bfloat16 header in`amd_hip_bfloat16.h` wh
125
161
- hipArray_t replaces hipArray*, as the pointer to array.
126
162
- This allows hipMemcpyAtoH and hipMemcpyHtoA to have the correct array type which is equivalent to coresponding CUDA driver APIs.
127
163
128
-
### Fixed
129
-
- Kernel launch maximum dimension validation is added specifically on gridY and gridZ in the HIP API hipModule-LaunchKernel. As a result,when hipGetDeviceAttribute is called for the value of hipDeviceAttributeMaxGrid-Dim, the behavior on the AMD platform is equivalent to NVIDIA.
130
-
- The HIP stream synchronisation behaviour is changed in internal stream functions, in which a flag "wait" is added and set when the current stream is null pointer while executing stream synchronisation on other explicitly created streams. This change avoids blocking of execution on null/default stream.
131
-
The change won't affect usage of applications, and makes them behave the same on the AMD platform as NVIDIA.
132
-
- Error handling behavior on unsupported GPU is fixed, HIP runtime will log out error message, instead of creating signal abortion error which is invisible to developers but continued kernel execution process. This is for the case when developers compile any application via hipcc, setting the option --offload-arch with GPU ID which is different from the one on the system.
133
-
134
-
### Deprecated And Removed
164
+
### Removed
135
165
- Deprecated Heterogeneous Compute (HCC) symbols and flags are removed from the HIP source code, including,
136
166
- Build options on obsolete HCC_OPTIONS was removed from cmake.
137
167
- Micro definitions are removed.
@@ -152,6 +182,12 @@ The change won't affect usage of applications, and makes them behave the same on
152
182
- Deprecated HIT based tests are removed from HIP project
153
183
- Catch tests are available [hip-tests] (https://github.com/ROCm/hip-tests) project
154
184
185
+
### Resolved issues
186
+
- Kernel launch maximum dimension validation is added specifically on gridY and gridZ in the HIP API hipModule-LaunchKernel. As a result,when hipGetDeviceAttribute is called for the value of hipDeviceAttributeMaxGrid-Dim, the behavior on the AMD platform is equivalent to NVIDIA.
187
+
- The HIP stream synchronisation behaviour is changed in internal stream functions, in which a flag "wait" is added and set when the current stream is null pointer while executing stream synchronisation on other explicitly created streams. This change avoids blocking of execution on null/default stream.
188
+
The change won't affect usage of applications, and makes them behave the same on the AMD platform as NVIDIA.
189
+
- Error handling behavior on unsupported GPU is fixed, HIP runtime will log out error message, instead of creating signal abortion error which is invisible to developers but continued kernel execution process. This is for the case when developers compile any application via hipcc, setting the option --offload-arch with GPU ID which is different from the one on the system.
190
+
155
191
### Known Issues
156
192
- Dynamically loaded HIP runtime library references incorrect version of hipDeviceGetProperties and hipChooseDevice APIs
157
193
@@ -161,13 +197,11 @@ As a workaround, while compiling with ROCm 6.0, use the string "hipDeviceGetProp
161
197
162
198
## HIP 5.7.1 (For ROCm 5.7.1)
163
199
164
-
### Fixed
200
+
### Resolved issues
165
201
- hipPointerGetAttributes API returns the correct HIP memory type as hipMemoryTypeManaged for managed memory.
166
202
167
203
## HIP 5.7 (For ROCm 5.7)
168
204
169
-
### Optimizations
170
-
171
205
### Added
172
206
- Added meta_group_size/rank for getting the number of tiles and rank of a tile in the partition
173
207
- Added new APIs supporting Windows only, under development on Linux
@@ -184,15 +218,11 @@ As a workaround, while compiling with ROCm 6.0, use the string "hipDeviceGetProp
184
218
185
219
- hipMipmappedArrayGetLevel for getting a mipmapped array on a mipmapped level
186
220
187
-
### Changed
188
-
189
-
### Fixed
190
-
191
221
### Known Issues
192
222
- HIP memory type enum values currently don't support equivalent value to cudaMemoryTypeUnregistered, due to HIP functionality backward compatibility.
193
223
- HIP API hipPointerGetAttributes could return invalid value in case the input memory pointer was not allocated through any HIP API on device or host.
194
224
195
-
### Upcoming changes in ROCm 6.0 release
225
+
### Upcoming changes
196
226
- Removal of gcnarch from hipDeviceProp_t structure
197
227
- Addition of new fields in hipDeviceProp_t structure
198
228
- maxTexture1D
@@ -218,17 +248,14 @@ As a workaround, while compiling with ROCm 6.0, use the string "hipDeviceGetProp
218
248
- Remove hiparray* and make it opaque with hipArray_t
219
249
220
250
## HIP 5.6.1 (For ROCm 5.6.1)
221
-
### Fixed
251
+
252
+
### Resolved issues
222
253
- Enabled xnack+ check in HIP catch2 tests hang while tests execution
223
254
- Memory leak when code object files are loaded/unloaded via hipModuleLoad/hipModuleUnload APIs
224
-
-Fixed a crash happening while using hipGraphAddMemFreeNode
255
+
-Resolved an issue of crash while using hipGraphAddMemFreeNode
225
256
226
257
## HIP 5.6 (For ROCm 5.6)
227
258
228
-
### Optimizations
229
-
- Consolidation of hipamd, rocclr and OpenCL projects in clr
230
-
- Optimized lock for graph global capture mode
231
-
232
259
### Added
233
260
- Added hipRTC support for amd_hip_fp16
234
261
- Added hipStreamGetDevice implementation to get the device assocaited with the stream
@@ -243,7 +270,11 @@ As a workaround, while compiling with ROCm 6.0, use the string "hipDeviceGetProp
243
270
- Consolidation of hipamd, ROCclr, and OpenCL repositories into a single repository called clr. Instructions are updated to build HIP from sources in the HIP Installation guide
244
271
- Removed hipBusBandwidth and hipCommander samples from hip-tests
245
272
246
-
### Fixed
273
+
### Optimized
274
+
- Consolidation of hipamd, rocclr and OpenCL projects in clr
275
+
- Optimized lock for graph global capture mode
276
+
277
+
### Resolved issues
247
278
- Fixed regression in hipMemCpyParam3D when offset is applied
248
279
249
280
### Known Issues
@@ -252,7 +283,7 @@ As a workaround, while compiling with ROCm 6.0, use the string "hipDeviceGetProp
252
283
- hipSetDevice and hipSetDeviceFlags APIs return hipErrorInvalidDevice instead of hipErrorNoDevice, on a system without GPU
253
284
- Known memory leak when code object files are loaded/unloaded via hipModuleLoad/hipModuleUnload APIs. Issue will be fixed in future release
254
285
255
-
### Upcoming changes in future release
286
+
### Upcoming changes
256
287
- Removal of gcnarch from hipDeviceProp_t structure
257
288
- Addition of new fields in hipDeviceProp_t structure
Copy file name to clipboardExpand all lines: CONTRIBUTING.md
+2-2Lines changed: 2 additions & 2 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -23,7 +23,7 @@ Please use the [GitHub Issue](https://github.com/ROCm/clr/issues) tab to notify
23
23
24
24
## Acceptance Criteria ##
25
25
26
-
clr is Common Language Runtime contains C++ codes for the implementation of HIP runtime APIs on the AMD platform.
26
+
clr Compute Language Runtime contains C++ codes for the implementation of HIP runtime APIs on the AMD platform.
27
27
Bug fixes and performance are both important goals in clr. Because of this, when a pull request is created, the owner of the repository will review, and put it in automated testing to make sure,
28
28
* The change will build on various OS platforms (Ubuntu, RHEL, etc.)
29
29
* The build package will install and run the code on different GPU architectures (MI-series, Radeon series cards, etc.),
@@ -38,7 +38,7 @@ clr contains three parts of codes,
38
38
39
39
-`opencl` - contains implementation of OpenCL on the AMD platform.
40
40
41
-
-`rocclr` - contains common runtime used in HIP and OpenCL, which includes
41
+
-`rocclr` - contains compute runtime used in HIP and OpenCL, which includes
42
42
-`include`, header files,
43
43
-`device`, implementation of GPU device related interfaces to the backend support,
44
44
-`cimpiler`, implementation of interfaces with compiler,
Copy file name to clipboardExpand all lines: README.md
+3-3Lines changed: 3 additions & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,12 +1,12 @@
1
1
# AMD CLR - Compute Language Runtimes
2
2
3
-
AMD CLR (Common Language Runtime) contains source codes for AMD's compute languages runtimes: `HIP` and `OpenCL™`.
3
+
AMD CLR (Compute Language Runtime) contains source codes for AMD's compute languages runtimes: `HIP` and `OpenCL™`.
4
4
5
5
## Project Organisation
6
6
7
7
-`hipamd` - contains implementation of `HIP` language on AMD platform. It is hosted at [ROCm/clr/hipamd](https://github.com/ROCm/clr/tree/develop/hipamd)
8
8
-`opencl` - contains implementation of [OpenCL™](https://www.khronos.org/opencl/) on AMD platform. Now it is hosted at [ROCm/clr/opencl](https://github.com/ROCm/clr/tree/develop/opencl)
9
-
-`rocclr` - contains common runtime used in `HIP` and `OpenCL™`. This is hosted at [ROCm/clr/rocclr](https://github.com/ROCm/clr/tree/develop/rocclr)
9
+
-`rocclr` - contains compute runtime used in `HIP` and `OpenCL™`. This is hosted at [ROCm/clr/rocclr](https://github.com/ROCm/clr/tree/develop/rocclr)
10
10
11
11
## How to build/install
12
12
@@ -56,7 +56,7 @@ To run `hip-tests` please go to the repository and follow the steps.
56
56
57
57
## Release notes
58
58
59
-
HIP provides release notes in [CLR change log](https://github.com/ROCm/clr/blob/develop/CHANGELOG.md), which has the records of changes in each release.
59
+
HIP provides release notes in [CLR change log](./CHANGELOG.md), which has the records of changes in each release.
0 commit comments