diff options
author | Pan Li <pan2.li@intel.com> | 2023-09-26 16:28:54 +0800 |
---|---|---|
committer | Pan Li <pan2.li@intel.com> | 2023-09-26 19:37:51 +0800 |
commit | d324984f9fa50e0055e3818f29d8c019df08cecf (patch) | |
tree | d90f53949fd5f49717da02f16f9c5ee2ebb60845 /gcc/tree-ssa-phiopt.cc | |
parent | 31ef3fe9e49b5446d465628830a9b6591c2168ce (diff) | |
download | gcc-d324984f9fa50e0055e3818f29d8c019df08cecf.zip gcc-d324984f9fa50e0055e3818f29d8c019df08cecf.tar.gz gcc-d324984f9fa50e0055e3818f29d8c019df08cecf.tar.bz2 |
RISC-V: Support FP round auto-vectorization
This patch would like to support auto-vectorization for the
round API in math.h. It depends on the -ffast-math option.
When we would like to call round/roundf like v2 = round (v1),
we will convert it into below insns (reference the implementation of llvm).
* vfcvt.x.f v3, v1, RMM
* vfcvt.f.x v2, v3
However, the floating point value may not need the cvt as above if
its mantissa is zero. Take single precision floating point as example:
+------------+---------------+-----------------+
| raw float | binary layout | after round |
+------------+---------------+-----------------+
| -8388607.5 | 0xcaffffff | -8388608.0 |
| 8388607.5 | 0x4affffff | 8388608.0 |
| 8388608.0 | 0x4b000000 | 8388608.0 |
| 8388609.0 | 0x4b000001 | 8388609.0 |
+------------+---------------+-----------------+
All single floating point >= 8388608.0 will have all zero mantisaa.
We leverage vmflt and mask to filter them out in vector and only do
the cvt on mask.
Befor this patch:
math-round-1.c:21:1: missed: couldn't vectorize loop
...
.L3:
flw fa0,0(s0)
addi s0,s0,4
addi s1,s1,4
call round
fsw fa0,-4(s1)
bne s0,s2,.L3
After this patch:
...
fsrmi 4 // RMM, rounding to nearest, ties to max magnitude
.L4:
vfabs.v v2,v1
vmflt.vf v0,v2,fa5
vfcvt.x.f.v v4,v1,v0.t
vfcvt.f.x.v v2,v4,v0.t
vfsgnj.vv v2,v2,v1
bne .L4
.L14:
fsrm a6
ret
Please note VLS mode is also involved in this patch and covered by the
test cases.
gcc/ChangeLog:
* config/riscv/autovec.md (round<mode>2): New pattern.
* config/riscv/riscv-protos.h (enum insn_flags): New enum type.
(enum insn_type): Ditto.
(expand_vec_round): New function decl.
* config/riscv/riscv-v.cc (expand_vec_round): New function impl.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/unop/math-round-0.c: New test.
* gcc.target/riscv/rvv/autovec/unop/math-round-1.c: New test.
* gcc.target/riscv/rvv/autovec/unop/math-round-2.c: New test.
* gcc.target/riscv/rvv/autovec/unop/math-round-3.c: New test.
* gcc.target/riscv/rvv/autovec/unop/math-round-run-1.c: New test.
* gcc.target/riscv/rvv/autovec/unop/math-round-run-2.c: New test.
* gcc.target/riscv/rvv/autovec/vls/math-round-1.c: New test.
Signed-off-by: Pan Li <pan2.li@intel.com>
Diffstat (limited to 'gcc/tree-ssa-phiopt.cc')
0 files changed, 0 insertions, 0 deletions