- hipDeviceGet/SetGraphMemAttr
- hipDeviceGraphMemTrim
- there is no memory pool for graphs currently
Signed-off-by: sdashmiz <shadi.dashmiz@amd.com>
Change-Id: I11db76ea7ea1c7732175fc93264448052357e8dc
[ROCm/clr commit: e8194dca76]
- Memcpy1D node type is not complying with cuda
Signed-off-by: sdashmiz <shadi.dashmiz@amd.com>
Change-Id: If8113f5e699de0c62d98effc4580a2e0fee9a950
[ROCm/clr commit: 13c875eaf0]
Let more types support hipResourceViewDesc.
Add missing symbols to fix cuda build failing.
Change-Id: Ife694cc6491427093863252e257e820b0bb4fa0f
[ROCm/clr commit: a19b767e8a]
Update HIP's unsafeAtomicAdd to:
- Compile properly even when not compiling for gfx90a
- Fall back to safe atomic add on non-gfx90a architectures
- use flat atomic add for FP64 on gfx90a, instead of dynamically
checking memory spaces.
In addition, when the compiler is passed -munsafe-fp-atomics, it
will define __AMDGCN_UNSAFE_FP_ATOMICS__. When this happens, the
compiler is requesting that the HIP headers force all HIP
atomicAdd() calls on floats or doubles to use their unsafe versions.
This patch thus causes unsafeAtomicAdd() calls when that define
is seen. This call to unsafeAtomicAdd() is also done for atomicSub(),
since that calls atomicAdd underneath. This is not done for
system-scope atomicAdd because, on gfx90a, system-scope atomic FP
add instructions would need to target fine-grained memory, which is
always unsafe.
This patch also creates safeAtomicAdd() functions for float and double.
These functions will create a standalone safe atomic, even when the
application is compiled with -munsafe-fp-atomics.
Finally, this patch adds wrappers in the Nvidia path of HIP so that
these HIP functions call through to atomicAdd there as well.
Change-Id: I8af0621d3d28ea30c9278bfeea7393d03bbdac6d
[ROCm/clr commit: f68149eafd]
align with CUDA to support surf1DLayeredread/write,
surf2DLayeredread/write, surfCubemapread/write functions.
Change-Id: Ie3f6ca75e23accd97cb35cdc8362d88a1e5dbd54
[ROCm/clr commit: bed3995147]
This reverts commit 42be9301d1.
Reason for revert: hip changes not in yet
Change-Id: I89e35d171cf01fb82d361279998efcc90e54ef33
[ROCm/clr commit: 1965dd5795]
HIP apps running on CUDA 11.0 are failing due to some graph APIs/enums
added only in CUDA 11.1 or CUDA 11.3
Change-Id: I0d32b412cb76c42c7b3a9c612d750990f9e89908
[ROCm/clr commit: 087c0c25b2]
Add mapFrom() to map "float __attribute__((ext_vector_type(4)))"
of output of low-level device function to right vector.
Change-Id: Ifed915c87273b784780b2180320ccb746845dafb
[ROCm/clr commit: 7de480760b]
Clang builtins doesnt yet support max/min atomics with float/double
parameters, Hence modify atomicMax/Min to use compare and swap
implementation
Change-Id: Ia351f9b136129fe25f0eddcbbb17549553247182
[ROCm/clr commit: f3c5f12d26]
If the target device doesn't support image,
compiler will send out unavailable error if device texture apis
are called; and host texture apis will return hipErrorNotSupported.
Remove useless include/hip/amd_detail/texture_functions.h.
Add hipDeviceAttributeImageSupport attribute to check image
support in runtime.
Add __HIP_NO_IMAGE_SUPPORT(=1) macro in device code to indicate
texture apis not supported.
Change-Id: Ia891493e681a61980a2444859a90e046a9ebea39
[ROCm/clr commit: 5acc8b5102]
Sarbojit Sarkar
Maneesh Gupta
jaypatel
Satyanvesh Dittakavi
sdashmiz
Tao Sang
Joseph Greathouse
Christophe Paquot
Rakesh Roy
German Andryeyev
haoyuan2
Sourabh Betigeri
pghafari
Payam
Nick Curtis
cjatin