[PATCH v2 2/2] RISC-V: Disable by pieces for vector setmem length > UNITS_PER_WORD

For fast unaligned access targets, by pieces uses up to UNITS_PER_WORD
size pieces resulting in more store instructions than needed.  For
example gcc.target/riscv/rvv/base/setmem-2.c:f1 built with
`-O3 -march=rv64gcv -mtune=thead-c906`:
```
f1:
        vsetivli        zero,8,e8,mf2,ta,ma
        vmv.v.x v1,a1
        vsetivli        zero,0,e32,mf2,ta,ma
        sb      a1,14(a0)
        vmv.x.s a4,v1
        vsetivli        zero,8,e16,m1,ta,ma
        vmv.x.s a5,v1
        vse8.v  v1,0(a0)
        sw      a4,8(a0)
        sh      a5,12(a0)
        ret
```

The slow unaligned access version built with `-O3 -march=rv64gcv` used
15 sb instructions:
```
f1:
        sb      a1,0(a0)
        sb      a1,1(a0)
        sb      a1,2(a0)
        sb      a1,3(a0)
        sb      a1,4(a0)
        sb      a1,5(a0)
        sb      a1,6(a0)
        sb      a1,7(a0)
        sb      a1,8(a0)
        sb      a1,9(a0)
        sb      a1,10(a0)
        sb      a1,11(a0)
        sb      a1,12(a0)
        sb      a1,13(a0)
        sb      a1,14(a0)
        ret
```

After this patch, the following is generated in both cases:
```
f1:
        vsetivli        zero,15,e8,m1,ta,ma
        vmv.v.x v1,a1
        vse8.v  v1,0(a0)
        ret
```

gcc/ChangeLog:

	* config/riscv/riscv.cc (riscv_use_by_pieces_infrastructure_p):
	New function.
	(TARGET_USE_BY_PIECES_INFRASTRUCTURE_P): Define.

gcc/testsuite/ChangeLog:

	* gcc.target/riscv/rvv/autovec/pr113469.c: Expect mf2 setmem.
	* gcc.target/riscv/rvv/base/setmem-2.c: Update f1 to expect
	straight-line vector memset.
	* gcc.target/riscv/rvv/base/setmem-3.c: Likewise.

gcc/config/riscv/riscv.cc

@@ -12695,6 +12695,22 @@ riscv_stack_clash_protection_alloca_probe_range (void)
   return STACK_CLASH_CALLER_GUARD;
 }
 
+static bool
+riscv_use_by_pieces_infrastructure_p (unsigned HOST_WIDE_INT size,
+				      unsigned alignment,
+				      enum by_pieces_operation op, bool speed_p)
+{
+  /* For set/clear with size > UNITS_PER_WORD, by pieces uses vector broadcasts
+     with UNITS_PER_WORD size pieces.  Use setmem<mode> instead which can use
+     bigger chunks.  */
+  if (TARGET_VECTOR && stringop_strategy & STRATEGY_VECTOR
+      && (op == CLEAR_BY_PIECES || op == SET_BY_PIECES)
+      && speed_p && size > UNITS_PER_WORD)
+    return false;
+
+  return default_use_by_pieces_infrastructure_p (size, alignment, op, speed_p);
+}
+
 /* Initialize the GCC target structure.  */
 #undef TARGET_ASM_ALIGNED_HI_OP
 #define TARGET_ASM_ALIGNED_HI_OP "\t.half\t"
@@ -13060,6 +13076,9 @@ riscv_stack_clash_protection_alloca_probe_range (void)
 #undef TARGET_C_MODE_FOR_FLOATING_TYPE
 #define TARGET_C_MODE_FOR_FLOATING_TYPE riscv_c_mode_for_floating_type
 
+#undef TARGET_USE_BY_PIECES_INFRASTRUCTURE_P
+#define TARGET_USE_BY_PIECES_INFRASTRUCTURE_P riscv_use_by_pieces_infrastructure_p
+
 struct gcc_target targetm = TARGET_INITIALIZER;
 
 #include "gt-riscv.h"

gcc/testsuite/gcc.target/riscv/rvv/autovec/pr113469.c

@@ -51,4 +51,5 @@ void p(int buf, __builtin_va_list ab, int q) {
  } while (k);
 }
 
-/* { dg-final { scan-assembler-times {vsetivli\tzero,\s*4,\s*e8,\s*mf4,\s*t[au],\s*m[au]} 2 } } */
+/* { dg-final { scan-assembler-times {vsetivli\tzero,\s*4,\s*e8,\s*mf4,\s*t[au],\s*m[au]} 1 } } */
+/* { dg-final { scan-assembler-times {vsetivli\tzero,\s*8,\s*e8,\s*mf2,\s*t[au],\s*m[au]} 1 } } */

gcc/testsuite/gcc.target/riscv/rvv/base/setmem-2.c

@@ -5,15 +5,17 @@
 
 #define MIN_VECTOR_BYTES (__riscv_v_min_vlen / 8)
 
-/* Small memsets shouldn't be vectorised.
+/* Vectorise with no loop.
 ** f1:
 **  (
-**  sb\s+a1,0\(a0\)
-**  ...
+**  vsetivli\s+zero,\d+,e8,m1,ta,ma
 **  |
-**  li\s+a2,\d+
-**  tail\s+memset
+**  li\s+a\d+,\d+
+**  vsetvli\s+zero,a\d+,e8,m1,ta,ma
 **  )
+**  vmv\.v\.x\s+v\d+,a1
+**  vse8\.v\s+v\d+,0\(a0\)
+**  ret
 */
 void *
 f1 (void *a, int const b)

gcc/testsuite/gcc.target/riscv/rvv/base/setmem-3.c

@@ -5,15 +5,17 @@
 
 #define MIN_VECTOR_BYTES (__riscv_v_min_vlen / 8)
 
-/* Small memsets shouldn't be vectorised.
+/* Vectorise with no loop.
 ** f1:
 **  (
-**  sb\s+a1,0\(a0\)
-**  ...
+**  vsetivli\s+zero,\d+,e8,m1,ta,ma
 **  |
-**  li\s+a2,\d+
-**  tail\s+memset
+**  li\s+a\d+,\d+
+**  vsetvli\s+zero,a\d+,e8,m1,ta,ma
 **  )
+**  vmv\.v\.x\s+v\d+,a1
+**  vse8\.v\s+v\d+,0\(a0\)
+**  ret
 */
 void *
 f1 (void *a, int const b)