maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/polly/test/Isl/CodeGen/OpenMP/new_multidim_access.ll

78 lines
3.3 KiB
LLVM
Raw Normal View History

[tests] Set -polly-import-jscop-dir=%S always This simplifies the test cases. llvm-svn: 307645
2017-07-11 18:39:01 +08:00
; RUN: opt %loadPolly -polly-import-jscop \
OpenMP codegen: support generation of multi-dimensional access functions When computing the index expressions for new, multi-dimensional memory accesses these new index expressions may reference original llvm::Values that are not transfered into the OpenMP subfunction. Using GlobalMap we now replace references to such values with the rewritten values that have e.g. been passed to the OpenMP subfunction. llvm-svn: 246923
2015-09-05 18:32:56 +08:00
; RUN: -analyze < %s | FileCheck %s
[tests] Set -polly-import-jscop-dir=%S always This simplifies the test cases. llvm-svn: 307645
2017-07-11 18:39:01 +08:00
; RUN: opt %loadPolly -polly-import-jscop \
tests: Drop -polly-detect-unprofitable and -polly-no-early-exit These flags are now always passed to all tests and need to be disabled if not needed. Disabling these flags, rather than passing them to almost all tests, significantly simplfies our RUN: lines. llvm-svn: 249422
2015-10-06 23:36:44 +08:00
; RUN: -polly-codegen -S < %s \
OpenMP codegen: support generation of multi-dimensional access functions When computing the index expressions for new, multi-dimensional memory accesses these new index expressions may reference original llvm::Values that are not transfered into the OpenMP subfunction. Using GlobalMap we now replace references to such values with the rewritten values that have e.g. been passed to the OpenMP subfunction. llvm-svn: 246923
2015-09-05 18:32:56 +08:00
; RUN: -polly-parallel \
; RUN: | FileCheck %s -check-prefix=IR
; void new_multidim_access(long n, long m, float A[][m]) {
; for (long i = 0; i < n; i++)
; for (long j = 0; j < 100; j++)
; A[i][2 * j] += i + j;
; }
; CHECK: ReadAccess := [Reduction Type: NONE] [Scalar: 0]
; CHECK: [n, m] -> { Stmt_bb4[i0, i1] -> MemRef_A[i0, 2i1] };
; CHECK: new: [n, m] -> { Stmt_bb4[i0, i1] -> MemRef_A[i0, 13 + i1] };
; CHECK: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]
; CHECK: [n, m] -> { Stmt_bb4[i0, i1] -> MemRef_A[i0, 2i1] };
; CHECK: new: [n, m] -> { Stmt_bb4[i0, i1] -> MemRef_A[i0, 43 + i1] };
Check overflows in RTCs and bail accordingly We utilize assumptions on the input to model IR in polyhedral world. To verify these assumptions we version the code and guard it with a runtime-check (RTC). However, since the RTCs are themselves generated from the polyhedral representation we generate them under the same assumptions that they should verify. In other words, the guarantees that we try to provide with the RTCs do not hold for the RTCs themselves. To this end it is necessary to employ a different check for the RTCs that will verify the assumptions did hold for them too. Differential Revision: http://reviews.llvm.org/D20165 llvm-svn: 269299
2016-05-12 23:12:43 +08:00
; IR: %polly.access.mul.polly.subfunc.arg.A = mul nsw i64 %polly.indvar, %polly.subfunc.arg.m
This reverts recent expression type changes The recent expression type changes still need more discussion, which will happen on phabricator or on the mailing list. The precise list of commits reverted are: - "Refactor division generation code" - "[NFC] Generate runtime checks after the SCoP" - "[FIX] Determine insertion point during SCEV expansion" - "Look through IntToPtr & PtrToInt instructions" - "Use minimal types for generated expressions" - "Temporarily promote values to i64 again" - "[NFC] Avoid unnecessary comparison for min/max expressions" - "[Polly] Fix -Wunused-variable warnings (NFC)" - "[NFC] Simplify min/max expression generation" - "Simplify the type adjustment in the IslExprBuilder" Some of them are just reverted as we would otherwise get conflicts. I will try to re-commit them if possible. llvm-svn: 272483
2016-06-12 03:17:15 +08:00
; IR: %6 = add nsw i64 %polly.indvar5, 13
; IR: %polly.access.add.polly.subfunc.arg.A = add nsw i64 %polly.access.mul.polly.subfunc.arg.A, %6
OpenMP: Name the values passed to the subfunciton according to the original llvm::Values llvm-svn: 246924
2015-09-05 18:41:19 +08:00
; IR: %polly.access.polly.subfunc.arg.A = getelementptr float, float* %polly.subfunc.arg.A, i64 %polly.access.add.polly.subfunc.arg.A
; IR: %tmp10_p_scalar_ = load float, float* %polly.access.polly.subfunc.arg.A, align 4, !alias.scope !0, !noalias !2, !llvm.mem.parallel_loop_access !3
Check overflows in RTCs and bail accordingly We utilize assumptions on the input to model IR in polyhedral world. To verify these assumptions we version the code and guard it with a runtime-check (RTC). However, since the RTCs are themselves generated from the polyhedral representation we generate them under the same assumptions that they should verify. In other words, the guarantees that we try to provide with the RTCs do not hold for the RTCs themselves. To this end it is necessary to employ a different check for the RTCs that will verify the assumptions did hold for them too. Differential Revision: http://reviews.llvm.org/D20165 llvm-svn: 269299
2016-05-12 23:12:43 +08:00
; IR: %polly.access.mul.polly.subfunc.arg.A8 = mul nsw i64 %polly.indvar, %polly.subfunc.arg.m
This reverts recent expression type changes The recent expression type changes still need more discussion, which will happen on phabricator or on the mailing list. The precise list of commits reverted are: - "Refactor division generation code" - "[NFC] Generate runtime checks after the SCoP" - "[FIX] Determine insertion point during SCEV expansion" - "Look through IntToPtr & PtrToInt instructions" - "Use minimal types for generated expressions" - "Temporarily promote values to i64 again" - "[NFC] Avoid unnecessary comparison for min/max expressions" - "[Polly] Fix -Wunused-variable warnings (NFC)" - "[NFC] Simplify min/max expression generation" - "Simplify the type adjustment in the IslExprBuilder" Some of them are just reverted as we would otherwise get conflicts. I will try to re-commit them if possible. llvm-svn: 272483
2016-06-12 03:17:15 +08:00
; IR: %7 = add nsw i64 %polly.indvar5, 43
; IR: %polly.access.add.polly.subfunc.arg.A9 = add nsw i64 %polly.access.mul.polly.subfunc.arg.A8, %7
Reapply "BlockGenerator: Generate synthesisable instructions only on-demand" Instructions which we can synthesis from a SCEV expression are not generated directly, but only when they are used as an operand of another instruction. This avoids generating unnecessary instructions and works more reliably than first inserting them and then deleting them later on. This commit was reverted in r248860 due to a remaining miscompile, where we forgot to synthesis the operand values that were referenced from scalar writes. test/Isl/CodeGen/scalar-store-from-same-bb.ll tests that we do this now correctly. llvm-svn: 248900
2015-09-30 21:36:54 +08:00
; IR: %polly.access.polly.subfunc.arg.A10 = getelementptr float, float* %polly.subfunc.arg.A, i64 %polly.access.add.polly.subfunc.arg.A9
; IR: store float %p_tmp11, float* %polly.access.polly.subfunc.arg.A10, align 4, !alias.scope !0, !noalias !2, !llvm.mem.parallel_
OpenMP codegen: support generation of multi-dimensional access functions When computing the index expressions for new, multi-dimensional memory accesses these new index expressions may reference original llvm::Values that are not transfered into the OpenMP subfunction. Using GlobalMap we now replace references to such values with the rewritten values that have e.g. been passed to the OpenMP subfunction. llvm-svn: 246923
2015-09-05 18:32:56 +08:00
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
define void @new_multidim_access(i64 %n, i64 %m, float* %A) {
bb:
br label %bb1
bb1: ; preds = %bb15, %bb
%i.0 = phi i64 [ 0, %bb ], [ %tmp16, %bb15 ]
%tmp = icmp slt i64 %i.0, %n
br i1 %tmp, label %bb2, label %bb17
bb2: ; preds = %bb1
br label %bb3
bb3: ; preds = %bb12, %bb2
%j.0 = phi i64 [ 0, %bb2 ], [ %tmp13, %bb12 ]
%exitcond = icmp ne i64 %j.0, 100
br i1 %exitcond, label %bb4, label %bb14
bb4: ; preds = %bb3
%tmp5 = add nsw i64 %i.0, %j.0
%tmp6 = sitofp i64 %tmp5 to float
%tmp7 = shl nsw i64 %j.0, 1
%tmp8 = mul nsw i64 %i.0, %m
%.sum = add i64 %tmp8, %tmp7
%tmp9 = getelementptr inbounds float, float* %A, i64 %.sum
%tmp10 = load float, float* %tmp9, align 4
%tmp11 = fadd float %tmp10, %tmp6
store float %tmp11, float* %tmp9, align 4
br label %bb12
bb12: ; preds = %bb4
%tmp13 = add nsw i64 %j.0, 1
br label %bb3
bb14: ; preds = %bb3
br label %bb15
bb15: ; preds = %bb14
%tmp16 = add nsw i64 %i.0, 1
br label %bb1
bb17: ; preds = %bb1
ret void
}