[Aarch64][SVE] Dot product support

Programming / Compilers / GCC - alejandro [138bc75d-0d04-0410-961f-82ee72b054a4] - 2 May 2019 09:58 EDT

This patch does two things. For the general vectoriser, it adds support to perform fully masked reductions over expressions that don't support masking. This is achieved by using VEC_COND_EXPR where possible. At the moment this is implemented for DOT_PROD_EXPR only, but the framework is there to extend it to other expressions.

Related to that, this patch adds support to vectorize dot product using SVE. It also uses the new functionality to ensure that the resulting loop is masked.

Given this input code:

uint32_t dotprod (uint8_t *restrict x, uint8_t *restrict y, int n) { uint32_t sum = 0;

for (int i = 0; i < n; i++) { sum += x[i] * y[i]; }

return sum; }

The resulting SVE code is:

0000000000000000 : 0: 7100005f cmp w2, #0x0 4: 5400024d b.le 4c 8: d2800003 mov x3, #0x0 // #0 c: 93407c42 sxtw x2, w2 10: 2538c001 mov z1.b, #0 14: 25221fe0 whilelo p0.b, xzr, x2 18: 2538c003 mov z3.b, #0 1c: d503201f nop 20: a4034002 ld1b {z2.b}, p0/z, [x0, x3] 24: a4034020 ld1b {z0.b}, p0/z, [x1, x3] 28: 0430e3e3 incb x3 2c: 0523c000 sel z0.b, p0, z0.b, z3.b 30: 25221c60 whilelo p0.b, x3, x2 34: 44820401 udot z1.s, z0.b, z2.b 38: 54ffff41 b.ne 20 // b.any 3c: 2598e3e0 ptrue p0.s 40: 04812021 uaddv d1, p0, z1.s 44: 1e260020 fmov w0, s1 48: d65f03c0 ret 4c: 1e2703e1 fmov s1, wzr 50: 1e260020 fmov w0, s1 54: d65f03c0 ret

Notice how udot is used inside a fully masked loop.

I tested this patch in an aarch64 machine bootstrapping the compiler and running the checks. gcc/Changelog:

2019-05-02 Alejandro Martinez

- config/aarch64/aarch64-sve.md (dot_prod): Taken from SVE ACLE branch.
- config/aarch64/iterators.md: Copied Vetype_fourth, VSI2QI and vsi2qi from SVE ACLE branch.
- tree-vect-loop.c (use_mask_by_cond_expr_p): New function to check if a
VEC_COND_EXPR be inserted to emulate a conditional internal function. (build_vect_cond_expr): Emit the VEC_COND_EXPR. (vectorizable_reduction): Use the functions above to vectorize in a fully masked loop codes that don't have a conditional internal function.

gcc/testsuite/Changelog:

2019-05-02 Alejandro Martinez

- gcc.target/aarch64/sve/dot_1.c: New test for dot product.

d3a7159f226 [Aarch64][SVE] Dot product support
gcc/ChangeLog | 13 ++++++
gcc/config/aarch64/aarch64-sve.md | 16 ++++++++
gcc/config/aarch64/iterators.md | 9 +++-
gcc/testsuite/ChangeLog | 4 ++
gcc/testsuite/gcc.target/aarch64/sve/dot_1.c | 39 ++++++++++++++++++
gcc/tree-vect-loop.c | 61 +++++++++++++++++++++++++++-
6 files changed, 139 insertions(+), 3 deletions(-)

Upstream: gcc.gnu.org


  • Share