tcg: Add support for vector bitwise select

This operation performs d = (b & a) | (c & ~a), and is present
on a majority of host vector units.  Include gvec expanders.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This commit is contained in:
Richard Henderson 2019-04-30 11:02:23 -07:00
parent 532ba368a1
commit 38dc12947e
12 changed files with 86 additions and 0 deletions

View File

@ -1444,3 +1444,17 @@ void HELPER(gvec_umax64)(void *d, void *a, void *b, uint32_t desc)
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_bitsel)(void *d, void *a, void *b, void *c, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(vec64)) {
vec64 aa = *(vec64 *)(a + i);
vec64 bb = *(vec64 *)(b + i);
vec64 cc = *(vec64 *)(c + i);
*(vec64 *)(d + i) = (bb & aa) | (cc & ~aa);
}
clear_high(d, oprsz, desc);
}

View File

@ -303,3 +303,5 @@ DEF_HELPER_FLAGS_4(gvec_leu8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_leu16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_leu32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_leu64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_5(gvec_bitsel, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)

View File

@ -627,6 +627,10 @@ E.g. VECL=1 -> 64 << 1 -> v128, and VECE=2 -> 1 << 2 -> i32.
Compare vectors by element, storing -1 for true and 0 for false.
* bitsel_vec v0, v1, v2, v3
Bitwise select, v0 = (v2 & v1) | (v3 & ~v1), across the entire vector.
*********
Note 1: Some shortcuts are defined when the last operand is known to be

View File

@ -140,6 +140,7 @@ typedef enum {
#define TCG_TARGET_HAS_mul_vec 1
#define TCG_TARGET_HAS_sat_vec 1
#define TCG_TARGET_HAS_minmax_vec 1
#define TCG_TARGET_HAS_bitsel_vec 0
#define TCG_TARGET_DEFAULT_MO (0)
#define TCG_TARGET_HAS_MEMORY_BSWAP 1

View File

@ -190,6 +190,7 @@ extern bool have_avx2;
#define TCG_TARGET_HAS_mul_vec 1
#define TCG_TARGET_HAS_sat_vec 1
#define TCG_TARGET_HAS_minmax_vec 1
#define TCG_TARGET_HAS_bitsel_vec 0
#define TCG_TARGET_deposit_i32_valid(ofs, len) \
(((ofs) == 0 && (len) == 8) || ((ofs) == 8 && (len) == 8) || \

View File

@ -3195,3 +3195,26 @@ void tcg_gen_gvec_cmp(TCGCond cond, unsigned vece, uint32_t dofs,
expand_clr(dofs + oprsz, maxsz - oprsz);
}
}
static void tcg_gen_bitsel_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b, TCGv_i64 c)
{
TCGv_i64 t = tcg_temp_new_i64();
tcg_gen_and_i64(t, b, a);
tcg_gen_andc_i64(d, c, a);
tcg_gen_or_i64(d, d, t);
tcg_temp_free_i64(t);
}
void tcg_gen_gvec_bitsel(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t cofs,
uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen4 g = {
.fni8 = tcg_gen_bitsel_i64,
.fniv = tcg_gen_bitsel_vec,
.fno = gen_helper_gvec_bitsel,
};
tcg_gen_gvec_4(dofs, aofs, bofs, cofs, oprsz, maxsz, &g);
}

View File

@ -342,6 +342,13 @@ void tcg_gen_gvec_cmp(TCGCond cond, unsigned vece, uint32_t dofs,
uint32_t aofs, uint32_t bofs,
uint32_t oprsz, uint32_t maxsz);
/*
* Perform vector bit select: d = (b & a) | (c & ~a).
*/
void tcg_gen_gvec_bitsel(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t cofs,
uint32_t oprsz, uint32_t maxsz);
/*
* 64-bit vector operations. Use these when the register has been allocated
* with tcg_global_mem_new_i64, and so we cannot also address it via pointer.

View File

@ -88,6 +88,7 @@ bool tcg_can_emit_vecop_list(const TCGOpcode *list,
case INDEX_op_dup2_vec:
case INDEX_op_ld_vec:
case INDEX_op_st_vec:
case INDEX_op_bitsel_vec:
/* These opcodes are mandatory and should not be listed. */
g_assert_not_reached();
default:
@ -691,3 +692,28 @@ void tcg_gen_sars_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 b)
{
do_shifts(vece, r, a, b, INDEX_op_sars_vec, INDEX_op_sarv_vec);
}
void tcg_gen_bitsel_vec(unsigned vece, TCGv_vec r, TCGv_vec a,
TCGv_vec b, TCGv_vec c)
{
TCGTemp *rt = tcgv_vec_temp(r);
TCGTemp *at = tcgv_vec_temp(a);
TCGTemp *bt = tcgv_vec_temp(b);
TCGTemp *ct = tcgv_vec_temp(c);
TCGType type = rt->base_type;
tcg_debug_assert(at->base_type >= type);
tcg_debug_assert(bt->base_type >= type);
tcg_debug_assert(ct->base_type >= type);
if (TCG_TARGET_HAS_bitsel_vec) {
vec_gen_4(INDEX_op_bitsel_vec, type, MO_8,
temp_arg(rt), temp_arg(at), temp_arg(bt), temp_arg(ct));
} else {
TCGv_vec t = tcg_temp_new_vec(type);
tcg_gen_and_vec(MO_8, t, a, b);
tcg_gen_andc_vec(MO_8, r, c, a);
tcg_gen_or_vec(MO_8, r, r, t);
tcg_temp_free_vec(t);
}
}

View File

@ -1000,6 +1000,9 @@ void tcg_gen_sarv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec s);
void tcg_gen_cmp_vec(TCGCond cond, unsigned vece, TCGv_vec r,
TCGv_vec a, TCGv_vec b);
void tcg_gen_bitsel_vec(unsigned vece, TCGv_vec r, TCGv_vec a,
TCGv_vec b, TCGv_vec c);
void tcg_gen_ld_vec(TCGv_vec r, TCGv_ptr base, TCGArg offset);
void tcg_gen_st_vec(TCGv_vec r, TCGv_ptr base, TCGArg offset);
void tcg_gen_stl_vec(TCGv_vec r, TCGv_ptr base, TCGArg offset, TCGType t);

View File

@ -256,6 +256,8 @@ DEF(sarv_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_shv_vec))
DEF(cmp_vec, 1, 2, 1, IMPLVEC)
DEF(bitsel_vec, 1, 3, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_bitsel_vec))
DEF(last_generic, 0, 0, 0, TCG_OPF_NOT_PRESENT)
#if TCG_TARGET_MAYBE_vec

View File

@ -1646,6 +1646,8 @@ bool tcg_op_supported(TCGOpcode op)
case INDEX_op_smax_vec:
case INDEX_op_umax_vec:
return have_vec && TCG_TARGET_HAS_minmax_vec;
case INDEX_op_bitsel_vec:
return have_vec && TCG_TARGET_HAS_bitsel_vec;
default:
tcg_debug_assert(op > INDEX_op_last_generic && op < NB_OPS);

View File

@ -187,6 +187,7 @@ typedef uint64_t TCGRegSet;
#define TCG_TARGET_HAS_mul_vec 0
#define TCG_TARGET_HAS_sat_vec 0
#define TCG_TARGET_HAS_minmax_vec 0
#define TCG_TARGET_HAS_bitsel_vec 0
#else
#define TCG_TARGET_MAYBE_vec 1
#endif