diff options
author | Juzhe-Zhong <juzhe.zhong@rivai.ai> | 2024-01-22 10:49:05 +0800 |
---|---|---|
committer | Pan Li <pan2.li@intel.com> | 2024-01-22 15:01:15 +0800 |
commit | a618b3c65745a6992ced4df60c186241c2da6eae (patch) | |
tree | 226dadb87b330c3f74aa427e076d435962622397 /gcc/ccmp.h | |
parent | 46a664dd59ec6c1121914d6857d6edb94501c517 (diff) | |
download | gcc-a618b3c65745a6992ced4df60c186241c2da6eae.zip gcc-a618b3c65745a6992ced4df60c186241c2da6eae.tar.gz gcc-a618b3c65745a6992ced4df60c186241c2da6eae.tar.bz2 |
RISC-V: Fix vfirst/vmsbf/vmsif/vmsof ratio attributes
vfirst/vmsbf/vmsif/vmsof instructions are supposed to demand ratio instead of demanding sew_lmul.
But my previous typo makes VSETVL PASS miss honor the risc-v v spec.
Consider this following simple case:
int foo4 (void * in, void * out)
{
vint32m1_t v = __riscv_vle32_v_i32m1 (in, 4);
v = __riscv_vadd_vv_i32m1 (v, v, 4);
vbool32_t mask = __riscv_vreinterpret_v_i32m1_b32(v);
mask = __riscv_vmsof_m_b32(mask, 4);
return __riscv_vfirst_m_b32(mask, 4);
}
Before this patch:
foo4:
vsetivli zero,4,e32,m1,ta,ma
vle32.v v1,0(a0)
vadd.vv v1,v1,v1
vsetvli zero,zero,e8,mf4,ta,ma ----> redundant.
vmsof.m v2,v1
vfirst.m a0,v2
ret
After this patch:
foo4:
vsetivli zero,4,e32,m1,ta,ma
vle32.v v1,0(a0)
vadd.vv v1,v1,v1
vmsof.m v2,v1
vfirst.m a0,v2
ret
Confirm RVV spec and Clang, this patch makes VSETVL PASS match the correct behavior.
Tested on both RV32/RV64, no regression.
gcc/ChangeLog:
* config/riscv/vector.md: Fix vfirst/vmsbf/vmsof ratio attributes.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/vsetvl/attribute-1.c: New test.
Diffstat (limited to 'gcc/ccmp.h')
0 files changed, 0 insertions, 0 deletions