llvm-project/polly/test/ScopInfo/multidim_fortran_2d_params.ll

87 lines
3.0 KiB
LLVM
Raw Normal View History

; RUN: opt %loadPolly -polly-scops -analyze \
[ScopInfo] Disable memory folding in case it results in multi-disjunct relations Multi-disjunct access maps can easily result in inbound assumptions which explode in case of many memory accesses and many parameters. This change reduces compilation time of some larger kernel from over 15 minutes to less than 16 seconds. Interesting is the test case test/ScopInfo/multidim_param_in_subscript.ll which has a memory access [n] -> { Stmt_for_body3[i0, i1] -> MemRef_A[i0, -1 + n - i1] } which requires folding, but where only a single disjunct remains. We can still model this test case even when only using limited memory folding. For people only reading commit messages, here the comment that explains what memory folding is: To recover memory accesses with array size parameters in the subscript expression we post-process the delinearization results. We would normally recover from an access A[exp0(i) * N + exp1(i)] into an array A[][N] the 2D access A[exp0(i)][exp1(i)]. However, another valid delinearization is A[exp0(i) - 1][exp1(i) + N] which - depending on the range of exp1(i) - may be preferrable. Specifically, for cases where we know exp1(i) is negative, we want to choose the latter expression. As we commonly do not have any information about the range of exp1(i), we do not choose one of the two options, but instead create a piecewise access function that adds the (-1, N) offsets as soon as exp1(i) becomes negative. For a 2D array such an access function is created by applying the piecewise map: [i,j] -> [i, j] : j >= 0 [i,j] -> [i-1, j+N] : j < 0 After this patch we generate only the first case, except for situations where we can proove the first case to be invalid and can consequently select the second without introducing disjuncts. llvm-svn: 296679
2017-03-02 05:11:27 +08:00
; RUN: -polly-precise-fold-accesses \
; RUN: -polly-invariant-load-hoisting=true < %s | FileCheck %s
; subroutine init_array(ni, nj, pi, pj, a)
; implicit none
; double precision, dimension(nj, ni) :: a
; integer*8 :: ni, nj
; integer*8 :: pi, pj
; integer*8 :: i, j
; do i = 1, ni
; do j = 1, nj
; a(j + pi, i + pj) = i + j
; end do
; end do
; end subroutine
2016-01-15 23:54:45 +08:00
; CHECK: [tmp9, nj_loaded2, tmp20, tmp19] -> { Stmt_bb17[i0, i1] -> MemRef_a[-1 + tmp20 + i0, nj_loaded2 + tmp19 + i1] : i1 < -tmp19; Stmt_bb17[i0, i1] -> MemRef_a[tmp20 + i0, tmp19 + i1] : i1 >= -tmp19 };
target datalayout = "e-p:64:64:64-S128-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f16:16:16-f32:32:32-f64:64:64-f128:128:128-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
target triple = "x86_64-unknown-linux-gnu"
module asm "\09.ident\09\22GCC: (GNU) 4.6.4 LLVM: 3.3.1\22"
; Function Attrs: nounwind uwtable
define void @blam(i64* noalias %arg, i64* noalias %nj, i64* noalias %arg2, i64* noalias %arg3, [0 x double]* noalias %a) unnamed_addr #0 {
bb:
br label %bb5
bb5: ; preds = %bb
%nj_loaded = load i64, i64* %nj, align 8
%tmp6 = icmp slt i64 %nj_loaded, 0
%tmp7 = select i1 %tmp6, i64 0, i64 %nj_loaded
%tmp8 = xor i64 %tmp7, -1
%tmp9 = load i64, i64* %arg, align 8
%tmp10 = icmp sgt i64 %tmp9, 0
br i1 %tmp10, label %bb11, label %bb36
bb11: ; preds = %bb5
br label %bb12
bb12: ; preds = %bb32, %bb11
%tmp13 = phi i64 [ %tmp34, %bb32 ], [ 1, %bb11 ]
%nj_loaded2 = load i64, i64* %nj, align 8
%tmp15 = icmp sgt i64 %nj_loaded2, 0
br i1 %tmp15, label %bb16, label %bb32
bb16: ; preds = %bb12
br label %bb17
bb17: ; preds = %bb17, %bb16
%tmp18 = phi i64 [ %tmp30, %bb17 ], [ 1, %bb16 ]
%tmp19 = load i64, i64* %arg2, align 8
%tmp20 = load i64, i64* %arg3, align 8
%tmp21 = add i64 %tmp20, %tmp13
%tmp22 = mul i64 %tmp21, %tmp7
%tmp23 = add i64 %tmp18, %tmp8
%tmp24 = add i64 %tmp23, %tmp19
%tmp25 = add i64 %tmp24, %tmp22
%tmp26 = add i64 %tmp18, %tmp13
%tmp27 = sitofp i64 %tmp26 to double
%tmp28 = getelementptr [0 x double], [0 x double]* %a, i64 0, i64 %tmp25
store double %tmp27, double* %tmp28, align 8
%tmp29 = icmp eq i64 %tmp18, %nj_loaded2
%tmp30 = add i64 %tmp18, 1
br i1 %tmp29, label %bb31, label %bb17
bb31: ; preds = %bb17
br label %bb32
bb32: ; preds = %bb31, %bb12
%tmp33 = icmp eq i64 %tmp13, %tmp9
%tmp34 = add i64 %tmp13, 1
br i1 %tmp33, label %bb35, label %bb12
bb35: ; preds = %bb32
br label %bb36
bb36: ; preds = %bb35, %bb5
ret void
}
attributes #0 = { nounwind uwtable }