RISC-V: Optimize codegen of VLA SLP

RISC-V: Optimize codegen of VLA SLP

Add comments for Robin:
We want to create a pattern where value[ix] = floor (ix / NPATTERNS).
As NPATTERNS is always a power of two we can rewrite this as
= ix & -NPATTERNS.
`
Recently, I figure out a better approach in case of codegen for VLA stepped vector.

Here is the detail descriptions:

Case 1:
void
f (uint8_t *restrict a, uint8_t *restrict b)
{
  for (int i = 0; i < 100; ++i)
    {
      a[i * 8] = b[i * 8 + 37] + 1;
      a[i * 8 + 1] = b[i * 8 + 37] + 2;
      a[i * 8 + 2] = b[i * 8 + 37] + 3;
      a[i * 8 + 3] = b[i * 8 + 37] + 4;
      a[i * 8 + 4] = b[i * 8 + 37] + 5;
      a[i * 8 + 5] = b[i * 8 + 37] + 6;
      a[i * 8 + 6] = b[i * 8 + 37] + 7;
      a[i * 8 + 7] = b[i * 8 + 37] + 8;
    }
}

We need to generate the stepped vector:
NPATTERNS = 8.
{ 0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 8, 8, 8, 8 }

Before this patch:
vid.v    v4         ;; {0,1,2,3,4,5,6,7,...}
vsrl.vi  v4,v4,3    ;; {0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,...}
li       a3,8       ;; {8}
vmul.vx  v4,v4,a3   ;; {0,0,0,0,0,0,0,8,8,8,8,8,8,8,8,...}

After this patch:
vid.v    v4                    ;; {0,1,2,3,4,5,6,7,...}
vand.vi  v4,v4,-8(-NPATTERNS)  ;; {0,0,0,0,0,0,0,8,8,8,8,8,8,8,8,...}

Case 2:
void
f (uint8_t *restrict a, uint8_t *restrict b)
{
  for (int i = 0; i < 100; ++i)
    {
      a[i * 8] = b[i * 8 + 3] + 1;
      a[i * 8 + 1] = b[i * 8 + 2] + 2;
      a[i * 8 + 2] = b[i * 8 + 1] + 3;
      a[i * 8 + 3] = b[i * 8 + 0] + 4;
      a[i * 8 + 4] = b[i * 8 + 7] + 5;
      a[i * 8 + 5] = b[i * 8 + 6] + 6;
      a[i * 8 + 6] = b[i * 8 + 5] + 7;
      a[i * 8 + 7] = b[i * 8 + 4] + 8;
    }
}

We need to generate the stepped vector:
NPATTERNS = 4.
{ 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12, ... }

Before this patch:
li       a6,134221824
slli     a6,a6,5
addi     a6,a6,3        ;; 64-bit: 0x0003000200010000
vmv.v.x  v6,a6          ;; {3, 2, 1, 0, ... }
vid.v    v4             ;; {0, 1, 2, 3, 4, 5, 6, 7, ... }
vsrl.vi  v4,v4,2        ;; {0, 0, 0, 0, 1, 1, 1, 1, ... }
li       a3,4           ;; {4}
vmul.vx  v4,v4,a3       ;; {0, 0, 0, 0, 4, 4, 4, 4, ... }
vadd.vv  v4,v4,v6       ;; {3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12, ... }

After this patch:
li	a3,-536875008
slli	a3,a3,4
addi	a3,a3,1
slli	a3,a3,16
vmv.v.x	v2,a3           ;; {3, 1, -1, -3, ... }
vid.v	v4              ;; {0, 1, 2, 3, 4, 5, 6, 7, ... }
vadd.vv	v4,v4,v2        ;; {3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12, ... }

gcc/ChangeLog:

	* config/riscv/riscv-v.cc (expand_const_vector): Optimize codegen.

gcc/testsuite/ChangeLog:

	* gcc.target/riscv/rvv/autovec/partial/slp-1.c: Adapt testcase.
	* gcc.target/riscv/rvv/autovec/partial/slp-16.c: New test.
	* gcc.target/riscv/rvv/autovec/partial/slp_run-16.c: New test.