[Bug tree-optimization/29777] missed optimization: model missing widen_mult* idioms for SSE

[ubizjak at gmail dot com]

------- Comment #1 from ubizjak at gmail dot com  2006-11-09 15:36 -------
> The testcases that failed (on assembler error) are two of tests that require
> "vect_widen_mult_hi_to_si":
> testsuite/gcc.dg/vect/vect-reduc-dot-s16a.c
> testsuite/gcc.dg/vect/vect-widen-mult-s16.c
> testsuite/gcc.dg/vect/vect-widen-mult-sum.c

But these files can be succesfully vectorized using current (gcc version 4.3.0
20061109) version on i686:

gcc -O2 -msse2 -ftree-vectorize -fdump-tree-vect-all vect-widen-mult-sum.c

vect-widen-mult-sum.c:16: note: LOOP VECTORIZED.
vect-widen-mult-sum.c:12: note: vectorized 1 loops in function.

.L8:
        movdqu  (%eax), %xmm4
        movdqu  (%ecx,%eax), %xmm2
        movdqa  %xmm4, %xmm1
        movdqa  %xmm2, %xmm0
        pxor    %xmm6, %xmm6
        pxor    %xmm5, %xmm5
        pcmpgtw %xmm2, %xmm6
        pcmpgtw %xmm4, %xmm5
        punpcklwd       %xmm6, %xmm0
        punpcklwd       %xmm5, %xmm1
        movdqa  %xmm0, %xmm3
        psrldq  $4, %xmm0
        pmuludq %xmm1, %xmm3
        psrldq  $4, %xmm1
        punpckhwd       %xmm6, %xmm2
        pmuludq %xmm1, %xmm0
        punpckhwd       %xmm5, %xmm4
        pshufd  $8, %xmm3, %xmm3
        pshufd  $8, %xmm0, %xmm0
        punpckldq       %xmm0, %xmm3
        movd    -20(%ebp), %xmm0
        psrad   %xmm0, %xmm3
        movdqa  %xmm2, %xmm0
        psrldq  $4, %xmm2
        pmuludq %xmm4, %xmm0
        psrldq  $4, %xmm4
        pmuludq %xmm4, %xmm2
        addl    $1, %edx
        paddd   %xmm7, %xmm3
        pshufd  $8, %xmm2, %xmm2
        pshufd  $8, %xmm0, %xmm7
        addl    $16, %eax
        punpckldq       %xmm2, %xmm7
        movd    -20(%ebp), %xmm0
        cmpl    %edx, %ebx
        psrad   %xmm0, %xmm7
        paddd   %xmm3, %xmm7
        ja      .L8
        movdqa  %xmm7, %xmm1
        movl    -16(%ebp), %esi
        psrldq  $8, %xmm1
        paddd   %xmm7, %xmm1
        cmpl    20(%ebp), %esi
        movdqa  %xmm1, %xmm0
        psrldq  $4, %xmm0
        paddd   %xmm1, %xmm0
        movd    %xmm0, -24(%ebp)
        movd    %xmm0, %edi
        je      .L4


> The missing insns (that should be merged from autovect-branch and debugged):
> vec_widen_umult_hi_v8hi
> vec_widen_umult_lo_v8hi

These patterns _are_ present in gcc version 4.3.0 20061109 (experimental) in
sse.md.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29777