maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/sgpr-control-flow.ll

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

236 lines
7.7 KiB
LLVM
Raw Normal View History

AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
AMDGPU: Don't use the default cpu in a few tests Avoids unnecessary test changes in a future commit. llvm-svn: 357539
2019-04-03 08:00:58 +08:00
; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=SI %s
R600/SI: Only select SALU instructions in the entry or exit block SALU instructions ignore control flow, so it is not always safe to use them within branches. This is a partial solution to this problem until we can come up with something better. llvm-svn: 207590
2014-04-30 07:12:48 +08:00
;
; Most SALU instructions ignore control flow, so we need to make sure
; they don't overwrite values from other blocks.
R600/SI: Enable selecting SALU inside branches We can do this now that the FixSGPRLiveRanges pass is working. llvm-svn: 218353
2014-09-24 09:33:28 +08:00
; If the branch decision is made based on a value in an SGPR then all
; threads will execute the same code paths, so we don't need to worry
; about instructions in different blocks overwriting each other.
R600/SI: Only select SALU instructions in the entry or exit block SALU instructions ignore control flow, so it is not always safe to use them within branches. This is a partial solution to this problem until we can come up with something better. llvm-svn: 207590
2014-04-30 07:12:48 +08:00
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sgpr_if_else_salu_br(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) {
AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; SI-LABEL: sgpr_if_else_salu_br:
; SI: ; %bb.0: ; %entry
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
; SI-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xb
; SI-NEXT: s_load_dword s0, s[0:1], 0xf
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s8, 0
; SI-NEXT: s_cbranch_scc0 BB0_2
; SI-NEXT: ; %bb.1: ; %else
; SI-NEXT: s_add_i32 s0, s11, s0
[AMDGPU] Disable 'Skip Uniform Regions' optimization by default for AMDGPU. Reviewers: sameerds, dstuttard Differential Revision: https://reviews.llvm.org/D77228
2020-04-06 21:05:58 +08:00
; SI-NEXT: s_mov_b64 s[2:3], 0
; SI-NEXT: s_andn2_b64 vcc, exec, s[2:3]
; SI-NEXT: s_cbranch_vccz BB0_3
; SI-NEXT: s_branch BB0_4
; SI-NEXT: BB0_2:
; SI-NEXT: s_mov_b64 s[2:3], -1
; SI-NEXT: ; implicit-def: $sgpr0
; SI-NEXT: s_andn2_b64 vcc, exec, s[2:3]
; SI-NEXT: s_cbranch_vccnz BB0_4
; SI-NEXT: BB0_3: ; %if
AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; SI-NEXT: s_sub_i32 s0, s9, s10
[AMDGPU] Disable 'Skip Uniform Regions' optimization by default for AMDGPU. Reviewers: sameerds, dstuttard Differential Revision: https://reviews.llvm.org/D77228
2020-04-06 21:05:58 +08:00
; SI-NEXT: BB0_4: ; %endif
AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; SI-NEXT: s_add_i32 s0, s0, s8
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: v_mov_b32_e32 v0, s0
; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT: s_endpgm
R600/SI: Only select SALU instructions in the entry or exit block SALU instructions ignore control flow, so it is not always safe to use them within branches. This is a partial solution to this problem until we can come up with something better. llvm-svn: 207590
2014-04-30 07:12:48 +08:00
entry:
%0 = icmp eq i32 %a, 0
br i1 %0, label %if, label %else
AMDGPU: Don't required structured CFG The structured CFG is just an aid to inserting exec mask modification instructions, once that is done we don't really need it anymore. We also do not analyze blocks with terminators that modify exec, so this should only be impacting true branches. llvm-svn: 288744
2016-12-06 09:02:51 +08:00
if:
%1 = sub i32 %b, %c
br label %endif
else:
%2 = add i32 %d, %e
br label %endif
endif:
%3 = phi i32 [%1, %if], [%2, %else]
%4 = add i32 %3, %a
store i32 %4, i32 addrspace(1)* %out
ret void
}
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
2018-06-27 03:10:00 +08:00
define amdgpu_kernel void @sgpr_if_else_salu_br_opt(i32 addrspace(1)* %out, [8 x i32], i32 %a, [8 x i32], i32 %b, [8 x i32], i32 %c, [8 x i32], i32 %d, [8 x i32], i32 %e) {
AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; SI-LABEL: sgpr_if_else_salu_br_opt:
; SI: ; %bb.0: ; %entry
; SI-NEXT: s_load_dword s2, s[0:1], 0x13
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s2, 0
; SI-NEXT: s_cbranch_scc0 BB1_2
; SI-NEXT: ; %bb.1: ; %else
; SI-NEXT: s_load_dword s3, s[0:1], 0x2e
[AMDGPU] Disable 'Skip Uniform Regions' optimization by default for AMDGPU. Reviewers: sameerds, dstuttard Differential Revision: https://reviews.llvm.org/D77228
2020-04-06 21:05:58 +08:00
; SI-NEXT: s_load_dword s6, s[0:1], 0x37
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_add_i32 s3, s3, s6
; SI-NEXT: s_mov_b64 s[6:7], 0
; SI-NEXT: s_andn2_b64 vcc, exec, s[6:7]
; SI-NEXT: s_cbranch_vccz BB1_3
; SI-NEXT: s_branch BB1_4
; SI-NEXT: BB1_2:
; SI-NEXT: s_mov_b64 s[6:7], -1
; SI-NEXT: ; implicit-def: $sgpr3
; SI-NEXT: s_andn2_b64 vcc, exec, s[6:7]
; SI-NEXT: s_cbranch_vccnz BB1_4
; SI-NEXT: BB1_3: ; %if
AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; SI-NEXT: s_load_dword s3, s[0:1], 0x1c
; SI-NEXT: s_load_dword s0, s[0:1], 0x25
; SI-NEXT: s_waitcnt lgkmcnt(0)
[AMDGPU] Disable 'Skip Uniform Regions' optimization by default for AMDGPU. Reviewers: sameerds, dstuttard Differential Revision: https://reviews.llvm.org/D77228
2020-04-06 21:05:58 +08:00
; SI-NEXT: s_add_i32 s3, s3, s0
; SI-NEXT: BB1_4: ; %endif
; SI-NEXT: s_add_i32 s0, s3, s2
AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: v_mov_b32_e32 v0, s0
; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT: s_endpgm
AMDGPU: Don't required structured CFG The structured CFG is just an aid to inserting exec mask modification instructions, once that is done we don't really need it anymore. We also do not analyze blocks with terminators that modify exec, so this should only be impacting true branches. llvm-svn: 288744
2016-12-06 09:02:51 +08:00
entry:
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
2018-06-27 03:10:00 +08:00
%cmp0 = icmp eq i32 %a, 0
br i1 %cmp0, label %if, label %else
AMDGPU: Don't required structured CFG The structured CFG is just an aid to inserting exec mask modification instructions, once that is done we don't really need it anymore. We also do not analyze blocks with terminators that modify exec, so this should only be impacting true branches. llvm-svn: 288744
2016-12-06 09:02:51 +08:00
R600/SI: Only select SALU instructions in the entry or exit block SALU instructions ignore control flow, so it is not always safe to use them within branches. This is a partial solution to this problem until we can come up with something better. llvm-svn: 207590
2014-04-30 07:12:48 +08:00
if:
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
2018-06-27 03:10:00 +08:00
%add0 = add i32 %b, %c
R600/SI: Only select SALU instructions in the entry or exit block SALU instructions ignore control flow, so it is not always safe to use them within branches. This is a partial solution to this problem until we can come up with something better. llvm-svn: 207590
2014-04-30 07:12:48 +08:00
br label %endif
else:
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
2018-06-27 03:10:00 +08:00
%add1 = add i32 %d, %e
R600/SI: Only select SALU instructions in the entry or exit block SALU instructions ignore control flow, so it is not always safe to use them within branches. This is a partial solution to this problem until we can come up with something better. llvm-svn: 207590
2014-04-30 07:12:48 +08:00
br label %endif
endif:
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
2018-06-27 03:10:00 +08:00
%phi = phi i32 [%add0, %if], [%add1, %else]
%add2 = add i32 %phi, %a
store i32 %add2, i32 addrspace(1)* %out
R600/SI: Only select SALU instructions in the entry or exit block SALU instructions ignore control flow, so it is not always safe to use them within branches. This is a partial solution to this problem until we can come up with something better. llvm-svn: 207590
2014-04-30 07:12:48 +08:00
ret void
}
R600/SI: Enable selecting SALU inside branches We can do this now that the FixSGPRLiveRanges pass is working. llvm-svn: 218353
2014-09-24 09:33:28 +08:00
; The two S_ADD instructions should write to different registers, since
; different threads will take different control flow paths.
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sgpr_if_else_valu_br(i32 addrspace(1)* %out, float %a, i32 %b, i32 %c, i32 %d, i32 %e) {
AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; SI-LABEL: sgpr_if_else_valu_br:
; SI: ; %bb.0: ; %entry
; SI-NEXT: v_cvt_f32_u32_e32 v0, v0
; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0xc
; SI-NEXT: ; implicit-def: $sgpr6
; SI-NEXT: v_cmp_lg_f32_e32 vcc, 0, v0
; SI-NEXT: s_and_saveexec_b64 s[8:9], vcc
; SI-NEXT: s_xor_b64 s[8:9], exec, s[8:9]
; SI-NEXT: s_cbranch_execz BB2_2
; SI-NEXT: ; %bb.1: ; %else
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_add_i32 s6, s2, s3
; SI-NEXT: BB2_2: ; %Flow
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_or_saveexec_b64 s[2:3], s[8:9]
; SI-NEXT: v_mov_b32_e32 v0, s6
; SI-NEXT: s_xor_b64 exec, exec, s[2:3]
; SI-NEXT: ; %bb.3: ; %if
; SI-NEXT: s_add_i32 s0, s0, s1
; SI-NEXT: v_mov_b32_e32 v0, s0
; SI-NEXT: ; %bb.4: ; %endif
; SI-NEXT: s_or_b64 exec, exec, s[2:3]
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT: s_endpgm
R600/SI: Enable selecting SALU inside branches We can do this now that the FixSGPRLiveRanges pass is working. llvm-svn: 218353
2014-09-24 09:33:28 +08:00
entry:
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Enable selecting SALU inside branches We can do this now that the FixSGPRLiveRanges pass is working. llvm-svn: 218353
2014-09-24 09:33:28 +08:00
%tid_f = uitofp i32 %tid to float
%tmp1 = fcmp ueq float %tid_f, 0.0
br i1 %tmp1, label %if, label %else
if:
%tmp2 = add i32 %b, %c
br label %endif
else:
%tmp3 = add i32 %d, %e
br label %endif
endif:
%tmp4 = phi i32 [%tmp2, %if], [%tmp3, %else]
store i32 %tmp4, i32 addrspace(1)* %out
ret void
}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sgpr_if_else_valu_cmp_phi_br(i32 addrspace(1)* %out, i32 addrspace(1)* %a, i32 addrspace(1)* %b) {
AMDGPU: Switch some tests to use generated checks Control flow tests are particularly annoying, and it's probably better to be have comprehensive check lines for them.
2020-02-01 07:14:50 +08:00
; SI-LABEL: sgpr_if_else_valu_cmp_phi_br:
; SI: ; %bb.0: ; %entry
; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT: s_mov_b32 s14, 0
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0
; SI-NEXT: ; implicit-def: $sgpr2_sgpr3
; SI-NEXT: s_and_saveexec_b64 s[8:9], vcc
; SI-NEXT: s_xor_b64 s[8:9], exec, s[8:9]
; SI-NEXT: s_cbranch_execz BB3_2
; SI-NEXT: ; %bb.1: ; %else
; SI-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0xd
; SI-NEXT: v_lshlrev_b32_e32 v1, 2, v0
; SI-NEXT: v_mov_b32_e32 v2, 0
; SI-NEXT: s_mov_b32 s15, 0xf000
; SI-NEXT: s_andn2_b64 s[0:1], s[0:1], exec
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: buffer_load_dword v1, v[1:2], s[12:15], 0 addr64
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_cmp_gt_i32_e32 vcc, 0, v1
; SI-NEXT: s_and_b64 s[2:3], vcc, exec
; SI-NEXT: s_or_b64 s[2:3], s[0:1], s[2:3]
; SI-NEXT: BB3_2: ; %Flow
; SI-NEXT: s_or_saveexec_b64 s[0:1], s[8:9]
; SI-NEXT: s_xor_b64 exec, exec, s[0:1]
; SI-NEXT: s_cbranch_execz BB3_4
; SI-NEXT: ; %bb.3: ; %if
; SI-NEXT: s_mov_b32 s11, 0xf000
; SI-NEXT: s_mov_b32 s10, 0
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b64 s[8:9], s[6:7]
; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; SI-NEXT: v_mov_b32_e32 v1, 0
; SI-NEXT: buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
; SI-NEXT: s_andn2_b64 s[2:3], s[2:3], exec
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; SI-NEXT: s_and_b64 s[6:7], vcc, exec
; SI-NEXT: s_or_b64 s[2:3], s[2:3], s[6:7]
; SI-NEXT: BB3_4: ; %endif
; SI-NEXT: s_or_b64 exec, exec, s[0:1]
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[2:3]
; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT: s_endpgm
R600/SI: Remove i1 pseudo VALU ops Select i1 logical ops directly to 64-bit SALU instructions. Vector i1 values are always really in SGPRs, with each bit for each item in the wave. This saves about 4 instructions when and/or/xoring any condition, and also helps write conditions that need to be passed in vcc. This should work correctly now that the SGPR live range fixing pass works. More work is needed to eliminate the VReg_1 pseudo regclass and possibly the entire SILowerI1Copies pass. llvm-svn: 223206
2014-12-03 13:22:35 +08:00
entry:
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Remove i1 pseudo VALU ops Select i1 logical ops directly to 64-bit SALU instructions. Vector i1 values are always really in SGPRs, with each bit for each item in the wave. This saves about 4 instructions when and/or/xoring any condition, and also helps write conditions that need to be passed in vcc. This should work correctly now that the SGPR live range fixing pass works. More work is needed to eliminate the VReg_1 pseudo regclass and possibly the entire SILowerI1Copies pass. llvm-svn: 223206
2014-12-03 13:22:35 +08:00
%tmp1 = icmp eq i32 %tid, 0
br i1 %tmp1, label %if, label %else
if:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%gep.if = getelementptr i32, i32 addrspace(1)* %a, i32 %tid
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%a.val = load i32, i32 addrspace(1)* %gep.if
R600/SI: Remove i1 pseudo VALU ops Select i1 logical ops directly to 64-bit SALU instructions. Vector i1 values are always really in SGPRs, with each bit for each item in the wave. This saves about 4 instructions when and/or/xoring any condition, and also helps write conditions that need to be passed in vcc. This should work correctly now that the SGPR live range fixing pass works. More work is needed to eliminate the VReg_1 pseudo regclass and possibly the entire SILowerI1Copies pass. llvm-svn: 223206
2014-12-03 13:22:35 +08:00
%cmp.if = icmp eq i32 %a.val, 0
br label %endif
else:
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%gep.else = getelementptr i32, i32 addrspace(1)* %b, i32 %tid
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%b.val = load i32, i32 addrspace(1)* %gep.else
R600/SI: Remove i1 pseudo VALU ops Select i1 logical ops directly to 64-bit SALU instructions. Vector i1 values are always really in SGPRs, with each bit for each item in the wave. This saves about 4 instructions when and/or/xoring any condition, and also helps write conditions that need to be passed in vcc. This should work correctly now that the SGPR live range fixing pass works. More work is needed to eliminate the VReg_1 pseudo regclass and possibly the entire SILowerI1Copies pass. llvm-svn: 223206
2014-12-03 13:22:35 +08:00
%cmp.else = icmp slt i32 %b.val, 0
br label %endif
endif:
%tmp4 = phi i1 [%cmp.if, %if], [%cmp.else, %else]
%ext = sext i1 %tmp4 to i32
store i32 %ext, i32 addrspace(1)* %out
ret void
}
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
declare i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Enable selecting SALU inside branches We can do this now that the FixSGPRLiveRanges pass is working. llvm-svn: 218353
2014-09-24 09:33:28 +08:00
attributes #0 = { readnone }