https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111232
Bug ID: 111232
Summary: RISC-V: Failed to combine vwmul + vadd into vwmacc
Product: gcc
Version: 14.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: juzhe.zhong at rivai dot ai
Target Milestone: ---
#include <stdlib.h>
#define N 64
#define DOT -21856
signed char X[N] __attribute__ ((__aligned__(__BIGGEST_ALIGNMENT__))) =
{0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63};
signed char Y[N] __attribute__ ((__aligned__(__BIGGEST_ALIGNMENT__))) =
{64,63,62,61,60,59,58,57,56,55,54,53,52,51,50,49,48,47,46,45,44,43,42,41,40,39,38,37,36,35,34,33,32,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1};
/* char->short->short dot product.
The dot-product pattern should be detected.
Should be vectorized on vect_sdot_qi targets (targets that support
dot-product of signed char).
This test currently fails to vectorize on targets that support
dot-product of chars into and int accumulator.
Can also be vectorized as widening-mult + summation,
or with type-conversion support.
*/
__attribute__ ((noinline)) short
foo(int len) {
int i;
short result = 0;
for (i=0; i<len; i++) {
result += (X[i] * Y[i]);
}
return result;
}
ASM:
vsetvli a5,a0,e8,mf2,ta,ma
vle8.v v4,0(a4)
vle8.v v3,0(a3)
sub a0,a0,a5
vsetvli a2,zero,e8,mf2,ta,ma
add a4,a4,a5
vwmul.vv v2,v4,v3
add a3,a3,a5
vsetvli zero,a5,e16,m1,tu,ma
vadd.vv v1,v2,v1
bne a0,zero,.L8
vsetvli a2,zero,e16,m1,ta,ma
vmv.s.x v2,zero
vredsum.vs v1,v1,v2
vmv.x.s a0,v1
ret
It should be vwmacc instead of vwmul + vadd.
