caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
cf699ad9ab458ed94de8a4e772534ff4a23c8933
clang-p2996/llvm/test/CodeGen/AMDGPU/setcc-opt.ll
Matt Arsenault 8c4a35237a AMDGPU: Add pass to lower kernel arguments to loads 
This replaces most argument uses with loads, but for
now not all.

The code in SelectionDAG for calling convention lowering
is actively harmful for amdgpu_kernel. It attempts to
split the argument types into register legal types, which
results in low quality code for arbitary types. Since
all kernel arguments are passed in memory, we just want the
raw types.

I've tried a couple of methods of mitigating this in SelectionDAG,
but it's easier to just bypass this problem alltogether. It's
possible to hack around the problem in the initial lowering,
but the real problem is the DAG then expects to be able to use
CopyToReg/CopyFromReg for uses of the arguments outside the block.

Exposing the argument loads in the IR also has the advantage
that the LoadStoreVectorizer can merge them.

I'm not sure the best approach to dealing with the IR
argument list is. The patch as-is just leaves the IR arguments
in place, so all the existing code will still compute the same
kernarg size and pointlessly lowers the arguments.

Arguably the frontend should emit kernels with an empty argument
list in the first place. Alternatively a dummy array could be
inserted as a single argument just to reserve space.

This does have some disadvantages. Local pointer kernel arguments can
no longer have AssertZext placed  on them as the equivalent !range
metadata is not valid on pointer  typed loads. This is mostly bad
for SI which needs to know about the known bits in order to use the
DS instruction offset, so in this case this is not done.

More importantly, this skips noalias arguments since this pass
does not yet convert this to the equivalent !alias.scope and !noalias
metadata. Producing this metadata correctly seems to be tricky,
although this logically is the same as inlining into a function which
doesn't exist. Additionally, exposing these loads to the vectorizer
may result in degraded aliasing information if a pointer load is
merged with another argument load.

I'm also not entirely sure this is preserving the current clover
ABI, although I would greatly prefer if it would stop widening
arguments and match the HSA ABI. As-is I think it is extending
< 4-byte arguments to 4-bytes but doesn't align them to 4-bytes.

llvm-svn: 335650
2018-06-26 19:10:00 +00:00

286 lines
9.6 KiB
LLVM
Raw Blame History

; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s
; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN -check-prefix=FUNC %s
; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
; FUNC-LABEL: {{^}}sext_bool_icmp_eq_0:
; GCN-NOT: v_cmp
; GCN: v_cmp_ne_u32_e32 vcc,
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN-NEXT:buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
; EG: SETNE_INT * [[CMP:T[0-9]+]].[[CMPCHAN:[XYZW]]], KC0[2].Z, KC0[2].W
; EG: AND_INT T{{[0-9]+.[XYZW]}}, PS, 1
define amdgpu_kernel void @sext_bool_icmp_eq_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp eq i32 %a, %b
%ext = sext i1 %icmp0 to i32
%icmp1 = icmp eq i32 %ext, 0
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}sext_bool_icmp_ne_0:
; GCN-NOT: v_cmp
; GCN: v_cmp_ne_u32_e32 vcc,
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN-NEXT: buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
; EG: SETNE_INT * [[CMP:T[0-9]+]].[[CMPCHAN:[XYZW]]], KC0[2].Z, KC0[2].W
; EG: AND_INT T{{[0-9]+.[XYZW]}}, PS, 1
define amdgpu_kernel void @sext_bool_icmp_ne_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = sext i1 %icmp0 to i32
%icmp1 = icmp ne i32 %ext, 0
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}sext_bool_icmp_eq_neg1:
; GCN-NOT: v_cmp
; GCN: v_cmp_eq_u32_e32 vcc,
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN-NEXT: buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @sext_bool_icmp_eq_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp eq i32 %a, %b
%ext = sext i1 %icmp0 to i32
%icmp1 = icmp eq i32 %ext, -1
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}sext_bool_icmp_ne_neg1:
; GCN-NOT: v_cmp
; GCN: v_cmp_eq_u32_e32 vcc,
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN-NEXT: buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @sext_bool_icmp_ne_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = sext i1 %icmp0 to i32
%icmp1 = icmp ne i32 %ext, -1
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zext_bool_icmp_eq_0:
; GCN-NOT: v_cmp
; GCN: v_cmp_ne_u32_e32 vcc,
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN-NEXT: buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @zext_bool_icmp_eq_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp eq i32 %a, %b
%ext = zext i1 %icmp0 to i32
%icmp1 = icmp eq i32 %ext, 0
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zext_bool_icmp_ne_0:
; GCN-NOT: v_cmp
; GCN: v_cmp_ne_u32_e32 vcc,
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN-NEXT: buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @zext_bool_icmp_ne_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = zext i1 %icmp0 to i32
%icmp1 = icmp ne i32 %ext, 0
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zext_bool_icmp_eq_1:
; GCN-NOT: v_cmp
; GCN: v_cmp_eq_u32_e32 vcc,
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN-NEXT: buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @zext_bool_icmp_eq_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp eq i32 %a, %b
%ext = zext i1 %icmp0 to i32
%icmp1 = icmp eq i32 %ext, 1
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zext_bool_icmp_ne_1:
; GCN-NOT: v_cmp
; GCN: v_cmp_eq_u32_e32 vcc,
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN-NEXT: buffer_store_byte [[RESULT]]
define amdgpu_kernel void @zext_bool_icmp_ne_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = zext i1 %icmp0 to i32
%icmp1 = icmp ne i32 %ext, 1
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; Reduces to false:
; FUNC-LABEL: {{^}}zext_bool_icmp_eq_neg1:
; GCN: v_mov_b32_e32 [[TMP:v[0-9]+]], 0{{$}}
; GCN: buffer_store_byte [[TMP]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @zext_bool_icmp_eq_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp eq i32 %a, %b
%ext = zext i1 %icmp0 to i32
%icmp1 = icmp eq i32 %ext, -1
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; Reduces to true:
; FUNC-LABEL: {{^}}zext_bool_icmp_ne_neg1:
; GCN: v_mov_b32_e32 [[TMP:v[0-9]+]], 1{{$}}
; GCN: buffer_store_byte [[TMP]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @zext_bool_icmp_ne_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = zext i1 %icmp0 to i32
%icmp1 = icmp ne i32 %ext, -1
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}cmp_zext_k_i8max:
; SI: s_load_dword [[VALUE:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb
; VI: s_load_dword [[VALUE:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c
; GCN: s_movk_i32 [[K255:s[0-9]+]], 0xff
; GCN-DAG: v_mov_b32_e32 [[VK255:v[0-9]+]], [[K255]]
; SI-DAG: s_and_b32 [[B:s[0-9]+]], [[VALUE]], [[K255]]
; SI: v_cmp_ne_u32_e32 vcc, [[B]], [[VK255]]
; VI-DAG: v_and_b32_e32 [[B:v[0-9]+]], [[VALUE]], [[VK255]]
; VI: v_cmp_ne_u16_e32 vcc, [[K255]], [[B]]
; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN: buffer_store_byte [[RESULT]]
; GCN: s_endpgm
define amdgpu_kernel void @cmp_zext_k_i8max(i1 addrspace(1)* %out, i8 %b) nounwind {
%b.ext = zext i8 %b to i32
%icmp0 = icmp ne i32 %b.ext, 255
store i1 %icmp0, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}cmp_sext_k_neg1:
; GCN: buffer_load_sbyte [[B:v[0-9]+]]
; GCN: v_cmp_ne_u32_e32 vcc, -1, [[B]]{{$}}
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN: buffer_store_byte [[RESULT]]
; GCN: s_endpgm
define amdgpu_kernel void @cmp_sext_k_neg1(i1 addrspace(1)* %out, i8 addrspace(1)* %b.ptr) nounwind {
%b = load i8, i8 addrspace(1)* %b.ptr
%b.ext = sext i8 %b to i32
%icmp0 = icmp ne i32 %b.ext, -1
store i1 %icmp0, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}v_cmp_sext_k_neg1_i8_sext_arg:
; GCN: v_cmp_ne_u32_e32 vcc, -1, v0
; GCN-NEXT: v_cndmask_b32_e64 [[SELECT:v[0-9]+]], 0, 1, vcc
; GCN: buffer_store_byte [[SELECT]]
define void @v_cmp_sext_k_neg1_i8_sext_arg(i8 signext %b) nounwind {
%b.ext = sext i8 %b to i32
%icmp0 = icmp ne i32 %b.ext, -1
store i1 %icmp0, i1 addrspace(1)* undef
ret void
}
; FIXME: This ends up doing a buffer_load_ubyte, and and compare to
; 255. Seems to be because of ordering problems when not allowing load widths to be reduced.
; Should do a buffer_load_sbyte and compare with -1
; FUNC-LABEL: {{^}}cmp_sext_k_neg1_i8_arg:
; SI: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xb
; VI: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c
; GCN: s_movk_i32 [[K:s[0-9]+]], 0xff
; GCN-DAG: s_and_b32 [[B:s[0-9]+]], [[VAL]], [[K]]
; GCN-DAG: v_mov_b32_e32 [[VK:v[0-9]+]], [[K]]
; GCN: v_cmp_ne_u32_e32 vcc, [[B]], [[VK]]{{$}}
; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
; GCN: buffer_store_byte [[RESULT]]
; GCN: s_endpgm
define amdgpu_kernel void @cmp_sext_k_neg1_i8_arg(i1 addrspace(1)* %out, i8 %b) nounwind {
%b.ext = sext i8 %b to i32
%icmp0 = icmp ne i32 %b.ext, -1
store i1 %icmp0, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}cmp_zext_k_neg1:
; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}}
; GCN: buffer_store_byte [[RESULT]]
; GCN: s_endpgm
define amdgpu_kernel void @cmp_zext_k_neg1(i1 addrspace(1)* %out, i8 %b) nounwind {
%b.ext = zext i8 %b to i32
%icmp0 = icmp ne i32 %b.ext, -1
store i1 %icmp0, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zext_bool_icmp_ne_k:
; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}}
; GCN: buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @zext_bool_icmp_ne_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = zext i1 %icmp0 to i32
%icmp1 = icmp ne i32 %ext, 2
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zext_bool_icmp_eq_k:
; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 0{{$}}
; GCN: buffer_store_byte [[RESULT]]
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @zext_bool_icmp_eq_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = zext i1 %icmp0 to i32
%icmp1 = icmp eq i32 %ext, 2
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FIXME: These cases should really be able fold to true/false in
; DAGCombiner
; This really folds away to false
; FUNC-LABEL: {{^}}sext_bool_icmp_eq_1:
; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 0{{$}}
; GCN: buffer_store_byte [[K]]
define amdgpu_kernel void @sext_bool_icmp_eq_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp eq i32 %a, %b
%ext = sext i1 %icmp0 to i32
%icmp1 = icmp eq i32 %ext, 1
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}sext_bool_icmp_ne_1:
; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 1{{$}}
; GCN: buffer_store_byte [[K]]
define amdgpu_kernel void @sext_bool_icmp_ne_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = sext i1 %icmp0 to i32
%icmp1 = icmp ne i32 %ext, 1
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}sext_bool_icmp_ne_k:
; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 1{{$}}
; GCN: buffer_store_byte [[K]]
define amdgpu_kernel void @sext_bool_icmp_ne_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
%icmp0 = icmp ne i32 %a, %b
%ext = sext i1 %icmp0 to i32
%icmp1 = icmp ne i32 %ext, 2
store i1 %icmp1, i1 addrspace(1)* %out
ret void
}
Reference in New Issue View Git Blame Copy Permalink