Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322899
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322897
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322896
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322895
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322894
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322893
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322892
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322891
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322890
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322889
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322888
Passes CTS on carrizo
v2: Use full precision implementation
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322887
Passes piglit on turks and carrizo
fp64 passes ctx on carrizo
v2: fix formatting
check fp32 denormal support at runtime
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322763
Passes piglit on turks and carrizo
fp64 passes cts on carrizo
v2: fix formatting
check fp32 denormal support at runtime
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322762
Passes piglit on turks and carrizo
fp64 passes CTS on carrizo
v2: fix formatting
check fp32 denormal support at runtime
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322761
Passes piglit on turks and carrizo
fp64 passes CTS on carrizo
v2: fix formatting
check fp32 denormal support at runtime
Reviewer: Jeroen Ketema <j.ketema@xs4all.nl>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 322760
v2: don't use assume
check only for x<0, the other conditions are handled transparently
v3: don't check inputs at all, nan propagation works as expected
Reviewer: Jeroen Ketema
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 318204
v2: Add __CLC_XCONCAT instead of function name redirection
Use __CLC_XCONCAT for intrinsic functions as well
Reviewer: Jeroen Ketema
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 317932
Use llvm instrinsic by default
Provide amdgpu workaround
v2: drop old amd copyrights
Reviewer: Aaron Watry
Reviewed-by: Vedran Miletić <vedran@miletic.net>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 316588
The implementation uses r600 sepcific intrinsics
LLVM-4 switched to _ro_t and _rw_t image types
Portions of the code can be moved back as more targets/llvm versions add image support
Reviewer: Aaron Watry
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 315341
The generated llvm IR mostly identical. char/uchar case is a bit worse.
reviewer: Tom Stellard
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 314701
v2: add vload(half) as well
make helpers amdgpu specific (NVPTX uses different private AS numbering)
use clang builtin on clang >= 6
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Tom Stellard <tstellar@redhat.com>
llvm-svn: 312839
Add missing undefs
Make helpers amdgpu specific (NVPTX uses different numbering for private AS)
Use clang builtins on clang >= 6
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Tom Stellard <tstellar@redhat.com>
llvm-svn: 312838
This was added in CL 1.1
Tested with a Radeon HD 7850 (Pitcairn) using the CL CTS via:
test_conformance/relationals/test_relationals shuffle_built_in_dual_input
v2: Add half support to shuffle2
Move shuffle2 to misc/
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 312404
This was added in CL 1.1
Tested with a Radeon HD 7850 (Pitcairn) using the CL CTS via:
test_conformance/relationals/test_relationals shuffle_built_in
v2: Add half-precision support to shuffle when available.
Move to misc/ and add section 6.12.12 to clc.h
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 312403
Ported from the amd-builtins branch.
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Matt Arsenault <Matthew.Arsenault@amd.com>
CC: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 292335
Ported from the amd-builtins branch.
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Matt Arsenault <Matthew.Arsenault@amd.com>
CC: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 292334
Just use lgamma_r and ignore the value returned in the second argument
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 281565
Ported from the amd-builtins branch, which is itself based on the
Sun Microsystems implementation.
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 281564
This macro is currently unused, but I plan to use it shortly.
The previous form did casts of pointers without an address space, which
doesn't work so well for CL 1.x.
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 281563
clang (since r280553) allows pointer casts in function overloads,
so we need to disambiguate the second argument.
clang might be smarter about overloads in the future
see https://reviews.llvm.org/D24113, but let's be safe in libclc anyway.
llvm-svn: 280871
OpenCL 1.0: "Returns y if y < x, otherwise it returns x. If x *and* y
are infinite or NaN, the return values are undefined."
OpenCL 1.1+: "Returns y if y < x, otherwise it returns x. If x *or* y
are infinite or NaN, the return values are undefined."
The 1.0 version is stricter so use that one.
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 276704
Also fix get_global_id to consider offset
No idea how to add this for ptx, so they are stuck with the old get_global_id
implementation.
v2: split to a separate patch
v3: Switch R600 to use implictarg.ptr
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 276443
Fixes fdim piglit on Turks
v2: use CL fmax instead of __builtin
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Tom Stellard <tom.stellard@amd.com>
llvm-svn: 269807
The scalar float/double function bodies are a direct copy/paste,
aside from the removed (optional) code in float function body that
requires subnormals.
reviewers: jvesely
Patch by: Vedran Miletić <rivanvx@gmail.com>
llvm-svn: 268766
Based on the amd-builtin, but explicitly vectorized for all sizes (not just
float4), and includes a vectorized double implementation.
Passes piglit (float) tests on pitcairn.
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 268708
The scalar float/double function bodies are a direct copy/paste
with usage of the CLC wrappers to vectorize them.
This commit also adds in the FP_ILOGB0 and FP_ILOGBNAN macros which are
equal to the results of ilogb(0.0f) and ilogb(float nan) respectively.
v2: Add FP_ILOGB0 and FP_ILOGBNAN definitions
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
v1 Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 261639
The float implementation is almost a direct port from the amd-builtins,
but instead of just having a scalar and float4 implementation, it has
a scalar and arbitrary width vector implementation.
The double scalar is also a direct port from AMD's builtin release.
The double vector implementation copies the logic in the float vector
implementation using the values from the double scalar version.
Both have been tested in piglit using tests sent to that project's
mailing list.
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 260114
Use the implementation was ported from the AMD builtin library rather
than LLVM Intrinsics.
This has been tested with piglit, OpenCV, and the ocl conformance tests.
llvm-svn: 243131
Passing values less than 0 to the llvm.sqrt() intrinsic results in
undefined behavior, so we need to check the input and return NaN if
is is less than 0.
v2:
- Fix build failures.
llvm-svn: 241906
Using exp2(x * M_LOG2E_F) does not give us accurate enough results for
OpenCL. If you look at the new exp implementation you'll see that
it does multiply the input by M_LOG2E_F, but it still uses the original
input in part of the calculation.
This exp implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.
llvm-svn: 237229
Not all targets support the intrinsic, so it's better to have a
generic implementation which does not use it.
This exp2 implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.
llvm-svn: 237228
This implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.
v2:
- Remove f suffix from constant in double implementations.
- Consolidate implementations using the .cl/.inc approach.
v3:
- Use __CLC_FPSIZE instead of __CLC_FP{32,64}
v4 (Jan Vesely):
- Limit to single precision.
llvm-svn: 236920
This is a generic implementation which just calls rsqrt.
Targets should override this if they want a faster implementation.
v2:
- Alphabettize SOURCES
v3 (Jan Vesely):
Limit to single precision types.
llvm-svn: 236915
Ported from AMD builtin library, passes piglit on Turks.
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 236647
The new implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.
llvm-svn: 236608
Jan Vesely