Introduce two new subtarget features:
- WMMA256bInsts for GFX11 WMMA instructions and
- WMMA128bInsts for GFX1170 and GFX12 WMMA and SWMMAC instructions
Some WMMA instructions have changed from GFX 11.0 to GFX 11.7 so new
Real versions were added with "_gfx1170" suffix. For consistency all
WMMA and SWMMAC GFX11.7 instructions use this suffix.
To resolve decoding issues between different formats for some WMMA
instructions between GFX 11 and GFX 11.7, new decoding tables were
added.
We should not allow `-wavefrontsize32` and `-wavefrontsize64` to be
specified at the same time. We should also not allow `-wavefrontsize32`
on a target that only supports `wavefrontsize32`, and the vice versa.
Xnack processing is essential and performed at the frontend to enable
ASan instrumentation for AMDGPU device code. Certain AMDGPU subtargets
like gfx1250 && gfx1251 don't have to enable 'xnack+' explictly in
'--offload-arch=' for device ASan instrumentation.
If a wavefrontsize32 or wavefrontsize64 is the only possible value
insert it into feature list by default and use that value as an
indication that another wavefront size is not legal.
1. The PR proceeds with a backend target hook to allow front-ends to
determine what target features are available in a compilation based on
the CPU name.
2. Fix a backend target feature bug that supports HTM for
Power8/9/10/11. However, HTM is only supported on Power8/9 according to
the ISA.
3. All target features that are hardcoded in PPC.cpp can be retrieved
from the backend target feature. I have double-checked that the
hardcoded logic for inferring target features from the CPU in the
frontend(PPC.cpp) is the same as in PPC.td.
The reland patch addressed the comment
https://github.com/llvm/llvm-project/pull/137670#discussion_r2143541120
This reverts commit 9208b343e962b9f1140ee345c0050a3920bdcbf2.
TargetParser shouldn't re-run the PPC subtarget tablegen target, it
should define its own `-gen-ppc-target-def` rule like all the other
targets do in llvm/include/llvm/TargetParser/CMakeLists.txt .
One user reported that there are incorrect CMake dependencies after this
change, so I will roll this back in the meantime.
1. The PR proceeds with a backend target hook to allow front-ends to
determine what target features are available in a compilation based on
the CPU name.
2. Fix a backend target feature bug that supports HTM for
Power8/9/10/11. However, HTM is only supported on Power8/9 according to
the ISA.
3. All target features that are hardcoded in PPC.cpp can be retrieved
from the backend target feature. I have double-checked that the
hardcoded logic for inferring target features from the CPU in the
frontend(PPC.cpp) is the same as in PPC.td.
This patch introduces the `vmem-to-lds-load-insts` target feature, which
can be used to enable builtins `__builtin_amdgcn_global_load_lds` and
`__builtin_amdgcn_raw_ptr_buffer_load_lds` on platforms which have this
feature.
This feature is only available on gfx9/10.
A limitation of using a common target feature for both builtins is that
we could have made `__builtin_amdgcn_raw_ptr_buffer_load_lds` available
on gfx6,7,8.
gfx940 and gfx941 are no longer supported. This is one of a series of
PRs to remove them from the code base.
This PR removes all non-documentation occurrences of gfx940/gfx941 from
the llvm directory, and the remaining occurrences in clang.
Documentation changes will follow.
For SWDEV-512631
When we did the initial AMDGCNSPIRV commits we left the initialisation
of the feature map in a relatively disorderly state. This change
corrects that oversight:
- We make sure that AMDGCNSPIRV actually advertises the union of all
AMDGCN features, as some were not included;
- We keep feature initialisation in sorted order to make it easy to pick
an insertion point when features are added in the future.
OPSEL ASM Syntax for v_cvt_scalef32_pk_f32_fp4 : opsel:[x,y,z]
where, x & y i.e. OPSEL[1 : 0] selects which src_byte to read.
OPSEL ASM Syntax for v_cvt_scalef32_pk_fp4_f32 : opsel:[a,b,c,d]
where, c & d i.e. OPSEL[3 : 2] selects which dst_byte to write.
Co-authored-by: Pravin Jagtap <Pravin.Jagtap@amd.com>
OPSEL[1:0] collectively decide which byte to read
from src input.
Builtin takes additional imm argument which
represents index (with valid values:[0:3]) of src
byte read. Out of bounds checks will added in next
patch.
OPSEL ASM Syntax: opsel:[x,y,z]
where,
opsel[x] = Inst{11} = src0_modifier{2}
opsel[y] = Inst{12} = src1_modifier{2}
opsel[z] = Inst{14} = src0_modifier{3}
Note: Inst{13} i.e. OPSEL[2] is ignored in
asm syntax and opsel[z] is meaningless
for v_cvt_scalef32_f32_{fp|bf}8
Co-authored-by: Pravin Jagtap <Pravin.Jagtap@amd.com>
The encoding of v_dot2c_f32_bf16 opcode is same as v_mac_f32 in gfx90a,
both from gfx9 series. This required a new decoderNameSpace GFX950_DOT.
Co-authored-by: Sirish Pande <Sirish.Pande@amd.com>
v_dot2_f32_bf16 was added in gfx11 along with v_dot2_f16_f16 and v_dot2_bf16_bf16.
All three instructions were part of Dot9 instructions in the compiler.
This patch will split existing dot9 (v_dot2_f16_f16, v_dot2_bf16_bf16, v_dot2_f32_bf16)
into new dot9 (v_dot2_f16_f16 and v_dot2_bf16_bf16), and dot12 (v_dot2_f32_bf16).
All necessary changes to gfx11 and gfx12 are updated to reflect this change.
Co-authored-by: Sirish Pande <Sirish.Pande@amd.com>
This was a bit annoying because these introduce a new special case
encoding usage. op_sel is repurposed as a subset of dpp controls,
and is eligible for VOP3->VOP1 shrinking. For some reason fi also
uses an enum value, so we need to convert the raw boolean to 1 instead
of -1.
The 2 registers are swapped, so this has 2 defs. Ideally the builtin
would return a pair, but that's difficult so return a vector instead.
This would make a hypothetical builtin that supports v2f16 directly
uglier.
