Keep the original data type of integer do-variables
for structured loops. When do-variable's data type
is an integer type shorter than IndexType, processing
the do-variable separately from the DoLoop's iteration index
allows getting rid of type casts, which can make backend
optimizations easier.
For example,
```
do i = 2, n-1
do j = 2, n-1
... = a(j-1, i)
end do
end do
```
If value of 'j' is computed by casting the DoLoop's iteration
index to 'i32', then Flang will produce the following LLVM IR:
```
%1 = trunc i64 %iter_index to i32
%2 = sub i32 %1, 1
%3 = sext i32 %2 to i64
```
LLVM's InstCombine may try to get rid of the sign extension,
and may transform this into:
```
%1 = shl i64 %iter_index, 32
%2 = add i64 %1, -4294967296
%3 = ashr exact i64 %2, 32
```
The extra computations for the element address applied on top
of this awkward pattern confuse LLVM vectorizer so that
it does not recognize the unit-strided access of 'a'.
Measured performance improvements on `SPEC CPU2000@IceLake`:
```
168.wupwise: 11.96%
171.swim: 11.22%
172.mrgid: 56.38%
178.galgel: 7.29%
301.apsi: 8.32%
```
Differential Revision: https://reviews.llvm.org/D132176
The FIR `do_loop` is designed as a structured operation with a single
block inside it. Presence of unstructured constructs like jumps, exits
inside the loop will cause the loop to be marked as unstructured. These
loops are lowered using the `control-flow` dialect branch operations.
Fortran semantics do not allow the loop variable to be modified inside
the loop. To prevent accidental modification, the iteration of the
loop is modeled by two variables, trip-count and loop-variable.
-> The trip-count and loop-variable are initialized in the pre-header.
The trip-count is set as (end-start+step)/step where end, start and
step have the usual meanings. The loop-variable is initialized to start.
-> The header block contains a conditional branch instruction which
selects between branching to the body of the loop or the exit block
depending on the value of the trip-count.
-> Inside the body, the trip-count is decremented and the loop-variable
incremented by the step value. Finally it branches to the header of the
loop.
Part of the upstreaming effort to move LLVM Flang from fir-dev branch of
https://github.com/flang-compiler/f18-llvm-project to the LLVM Project.
Reviewed By: awarzynski
Differential Revision: https://reviews.llvm.org/D124837
Co-authored-by: Val Donaldson <vdonaldson@nvidia.com>
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: Peter Klausler <pklausler@nvidia.com>
