Commit Graph

25 Commits

Author SHA1 Message Date
Jan Vesely 8fa100dfe3 amdgcn/fmin: Fix typos that reduced precision
Not sure how these sneaked in.
Fixes fminD and few other tests(fractD, cosD) on carrizo
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Aaron Watry <awatry@gmail.com>

llvm-svn: 330198
2018-04-17 18:11:29 +00:00
Jan Vesely fd11db19c2 amdgcn: Update datalayout after LLVM r328656
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Aaron Watry <awatry@gmail.com>
llvm-svn: 329290
2018-04-05 14:47:44 +00:00
Jan Vesely f96b1b88f8 amdgcn/fmax: fcanonicalize operands
v_max instruction needs canonicalized operands.
Passes CTS on carrizo

Reviewer: Aaron Watry <awatry@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 327076
2018-03-08 23:01:01 +00:00
Jan Vesely e724e346ab amdgcn/fmin: fcanonicalize operands
v_min instruction needs canonicalized operands.
Passes CTS on carrizo

Reviewer: Aaron Watry <awatry@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 327075
2018-03-08 23:00:58 +00:00
Jan Vesely 04a46bf0a2 amdgcn,popcount: Workaround broken llvm.ctpop intrinsic on some GCN ASICs
This is only really needed for VI+ ASICs. However, llvm would cast the value to
i32 for older asics anyway. The proper fix is in LLVM-7 (r326535).
Fixes CTS popcount on carrizo.

Reviewer: Aaron Watry <awatry@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 327044
2018-03-08 18:58:07 +00:00
Jan Vesely 1ad6a94676 amdgcn: Fix build after GDS/const AS swap in r325030
Acked-by: Aaron Watry <awatry@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 325866
2018-02-23 07:37:01 +00:00
Jan Vesely eda1872d04 amdgcn: Fix datalayout after addition of 32bit const AS in r324747
Acked-by: Aaron Watry <awatry@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 325865
2018-02-23 07:36:54 +00:00
Jan Vesely 911666f3fa amdgcn: Fix datalayout after clang r324101
r324101 switched around AS numbering

Acked-by: Aaron Watry <awatry@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 325863
2018-02-23 07:36:39 +00:00
Jan Vesely c420b61b26 amdgcn: Add missing datalayout info to .ll files
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Acked-by: Aaron Watry <awatry@gmail.com>
llvm-svn: 316239
2017-10-20 21:10:18 +00:00
Jeroen Ketema fe9fa89854 Let get_work_dim take exactly 0 arguments
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 314634
2017-10-01 20:11:46 +00:00
Jan Vesely ce29e8cde1 Restore support for llvm-3.9
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Acked-by: Aaron Watry <awatry@gmail.com>
llvm-svn: 314543
2017-09-29 19:06:41 +00:00
Jan Vesely c9bbbe2403 Implement cl_khr_int64_extended_atomics builtins
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Aaron Watry <awatry@gmail.com>
Tested-by: Aaron Watry <awatry@gmail.com>
llvm-svn: 313811
2017-09-20 20:42:19 +00:00
Jan Vesely 3d1db3de74 amdgcn,waitcnt: Add datalayout info
This file is only compiled for GCN which all share the same layout

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Aaron Watry <awatry@gmail.com>
llvm-svn: 312493
2017-09-04 15:52:07 +00:00
Jan Vesely 999b1d9426 amdgcn: rewrite barrier() using fence and clang __builtin_amdgcn_s_barrier
Specs require using fences when barrier() is invoked:
"The barrier function will either flush any variables stored in local memory
or queue a memory fence to ensure correct ordering of memory operations to local memory."
and
"The barrier function will queue a memory fence to ensure correct ordering
of memory operations to global memory."

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Aaron Watry <awatry@gmail.com>
Tested-by: Aaron Watry <awatry@gmail.com>
llvm-svn: 311022
2017-08-16 17:09:00 +00:00
Jan Vesely 1977092dc3 amdgcn: Implement {read_,write_,}mem_fence builtin
v2: add more detailed comment about waitcnt instruction

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Aaron Watry <awatry@gmail.com>
Tested-by: Aaron Watry <awatry@gmail.com>
llvm-svn: 311021
2017-08-16 17:08:56 +00:00
Matt Arsenault 958fce3192 amdgcn: Fix return type of get_num_groups
llvm-svn: 279723
2016-08-25 07:31:40 +00:00
Matt Arsenault 26d9c41ff6 amdgcn: Fix return type for get_global_size
llvm-svn: 279644
2016-08-24 17:52:04 +00:00
Matt Arsenault 314364cbd2 amdgpu: Fix default case value for get_local_size
llvm-svn: 279359
2016-08-20 04:17:17 +00:00
Matt Arsenault 220268d177 amdgcn: Fix get_local_size IR return type
llvm-svn: 279350
2016-08-20 00:01:21 +00:00
Matt Arsenault 2ce3d94a01 amdgcn: Correct return types to be size_t
llvm-svn: 279343
2016-08-19 22:49:39 +00:00
Jan Vesely a82e080b57 AMDGPU: Implement get_global_offset builtin
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
2016-07-22 17:24:24 +00:00
Jan Vesely 74f02db922 AMDGPU: Use clang intrinsics for workitem builtins
v2: split into 2 patches
    use clang builtins for other intrinsics as well

v3: Fix warnings
    Switch r600 to use implictarg.ptr

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 276442
2016-07-22 17:24:20 +00:00
Matt Arsenault b456c6dd56 Replace llvm.AMDGPU.ldexp with llvm.amdgcn.ldexp
It didn't really work on r600 to begin with, which should
get its own intrinsic.

llvm-svn: 275813
2016-07-18 16:42:50 +00:00
Matt Arsenault 45e6eaaa05 amdgcn: Use new workitem intrinsics
llvm-svn: 261042
2016-02-17 00:27:27 +00:00
Matt Arsenault a48e15c6cb Split sources for amdgcn and r600
Most files remain in a common amdgpu directory.

Also switches barriers to to use convergent,
and use llvm.amdgcn.s.barrier.

This now requires 3.9/trunk to build amdgcn.

llvm-svn: 260777
2016-02-13 01:01:59 +00:00