tcg: Merge INDEX_op_extract_{i32,i64}

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This commit is contained in:
Richard Henderson 2025-01-11 09:01:46 -08:00
parent 5a4d034f3c
commit 07d5d502f2
7 changed files with 20 additions and 37 deletions

View File

@ -456,7 +456,7 @@ Misc
|
| *dest* = (*t1* & ~0x0f00) | ((*t2* << 8) & 0x0f00)
* - extract_i32/i64 *dest*, *t1*, *pos*, *len*
* - extract *dest*, *t1*, *pos*, *len*
sextract_i32/i64 *dest*, *t1*, *pos*, *len*
@ -467,12 +467,12 @@ Misc
to the left with zeros; for sextract_*, the result will be extended
to the left with copies of the bitfield sign bit at *pos* + *len* - 1.
|
| For example, "sextract_i32 dest, t1, 8, 4" indicates a 4-bit field
| For example, "sextract dest, t1, 8, 4" indicates a 4-bit field
at bit 8. This operation would be equivalent to
|
| *dest* = (*t1* << 20) >> 28
|
| (using an arithmetic right shift).
| (using an arithmetic right shift) on TCG_TYPE_I32.
* - extract2_i32/i64 *dest*, *t1*, *t2*, *pos*

View File

@ -54,6 +54,7 @@ DEF(divs2, 2, 3, 0, TCG_OPF_INT)
DEF(divu, 1, 2, 0, TCG_OPF_INT)
DEF(divu2, 2, 3, 0, TCG_OPF_INT)
DEF(eqv, 1, 2, 0, TCG_OPF_INT)
DEF(extract, 1, 1, 2, TCG_OPF_INT)
DEF(movcond, 1, 4, 1, TCG_OPF_INT)
DEF(mul, 1, 2, 0, TCG_OPF_INT)
DEF(muls2, 2, 2, 0, TCG_OPF_INT)
@ -89,7 +90,6 @@ DEF(st16_i32, 0, 2, 1, 0)
DEF(st_i32, 0, 2, 1, 0)
/* shifts/rotates */
DEF(deposit_i32, 1, 2, 2, 0)
DEF(extract_i32, 1, 1, 2, 0)
DEF(sextract_i32, 1, 1, 2, 0)
DEF(extract2_i32, 1, 2, 1, 0)
@ -112,7 +112,6 @@ DEF(st32_i64, 0, 2, 1, 0)
DEF(st_i64, 0, 2, 1, 0)
/* shifts/rotates */
DEF(deposit_i64, 1, 2, 2, 0)
DEF(extract_i64, 1, 1, 2, 0)
DEF(sextract_i64, 1, 1, 2, 0)
DEF(extract2_i64, 1, 2, 1, 0)

View File

@ -2317,7 +2317,7 @@ static int fold_setcond_zmask(OptContext *ctx, TCGOp *op, bool neg)
static void fold_setcond_tst_pow2(OptContext *ctx, TCGOp *op, bool neg)
{
TCGOpcode uext_opc = 0, sext_opc = 0;
TCGOpcode sext_opc = 0;
TCGCond cond = op->args[3];
TCGArg ret, src1, src2;
TCGOp *op2;
@ -2338,17 +2338,11 @@ static void fold_setcond_tst_pow2(OptContext *ctx, TCGOp *op, bool neg)
switch (ctx->type) {
case TCG_TYPE_I32:
if (TCG_TARGET_extract_valid(TCG_TYPE_I32, sh, 1)) {
uext_opc = INDEX_op_extract_i32;
}
if (TCG_TARGET_sextract_valid(TCG_TYPE_I32, sh, 1)) {
sext_opc = INDEX_op_sextract_i32;
}
break;
case TCG_TYPE_I64:
if (TCG_TARGET_extract_valid(TCG_TYPE_I64, sh, 1)) {
uext_opc = INDEX_op_extract_i64;
}
if (TCG_TARGET_sextract_valid(TCG_TYPE_I64, sh, 1)) {
sext_opc = INDEX_op_sextract_i64;
}
@ -2367,8 +2361,8 @@ static void fold_setcond_tst_pow2(OptContext *ctx, TCGOp *op, bool neg)
op->args[2] = sh;
op->args[3] = 1;
return;
} else if (sh && uext_opc) {
op->opc = uext_opc;
} else if (sh && TCG_TARGET_extract_valid(ctx->type, sh, 1)) {
op->opc = INDEX_op_extract;
op->args[1] = src1;
op->args[2] = sh;
op->args[3] = 1;
@ -2897,7 +2891,7 @@ void tcg_optimize(TCGContext *s)
case INDEX_op_eqv_vec:
done = fold_eqv(&ctx, op);
break;
CASE_OP_32_64(extract):
case INDEX_op_extract:
done = fold_extract(&ctx, op);
break;
CASE_OP_32_64(extract2):

View File

@ -998,7 +998,7 @@ void tcg_gen_extract_i32(TCGv_i32 ret, TCGv_i32 arg,
}
if (TCG_TARGET_extract_valid(TCG_TYPE_I32, ofs, len)) {
tcg_gen_op4ii_i32(INDEX_op_extract_i32, ret, arg, ofs, len);
tcg_gen_op4ii_i32(INDEX_op_extract, ret, arg, ofs, len);
return;
}
if (ofs == 0) {
@ -1008,7 +1008,7 @@ void tcg_gen_extract_i32(TCGv_i32 ret, TCGv_i32 arg,
/* Assume that zero-extension, if available, is cheaper than a shift. */
if (TCG_TARGET_extract_valid(TCG_TYPE_I32, 0, ofs + len)) {
tcg_gen_op4ii_i32(INDEX_op_extract_i32, ret, arg, 0, ofs + len);
tcg_gen_op4ii_i32(INDEX_op_extract, ret, arg, 0, ofs + len);
tcg_gen_shri_i32(ret, ret, ofs);
return;
}
@ -2670,7 +2670,7 @@ void tcg_gen_extract_i64(TCGv_i64 ret, TCGv_i64 arg,
}
if (TCG_TARGET_extract_valid(TCG_TYPE_I64, ofs, len)) {
tcg_gen_op4ii_i64(INDEX_op_extract_i64, ret, arg, ofs, len);
tcg_gen_op4ii_i64(INDEX_op_extract, ret, arg, ofs, len);
return;
}
if (ofs == 0) {
@ -2680,7 +2680,7 @@ void tcg_gen_extract_i64(TCGv_i64 ret, TCGv_i64 arg,
/* Assume that zero-extension, if available, is cheaper than a shift. */
if (TCG_TARGET_extract_valid(TCG_TYPE_I64, 0, ofs + len)) {
tcg_gen_op4ii_i64(INDEX_op_extract_i64, ret, arg, 0, ofs + len);
tcg_gen_op4ii_i64(INDEX_op_extract, ret, arg, 0, ofs + len);
tcg_gen_shri_i64(ret, ret, ofs);
return;
}

View File

@ -1091,8 +1091,7 @@ static const TCGOutOp * const all_outop[NB_OPS] = {
OUTOP(INDEX_op_divs2, TCGOutOpDivRem, outop_divs2),
OUTOP(INDEX_op_divu2, TCGOutOpDivRem, outop_divu2),
OUTOP(INDEX_op_eqv, TCGOutOpBinary, outop_eqv),
OUTOP(INDEX_op_extract_i32, TCGOutOpExtract, outop_extract),
OUTOP(INDEX_op_extract_i64, TCGOutOpExtract, outop_extract),
OUTOP(INDEX_op_extract, TCGOutOpExtract, outop_extract),
OUTOP(INDEX_op_movcond, TCGOutOpMovcond, outop_movcond),
OUTOP(INDEX_op_mul, TCGOutOpBinary, outop_mul),
OUTOP(INDEX_op_muls2, TCGOutOpMul2, outop_muls2),
@ -2326,6 +2325,7 @@ bool tcg_op_supported(TCGOpcode op, TCGType type, unsigned flags)
case INDEX_op_add:
case INDEX_op_and:
case INDEX_op_brcond:
case INDEX_op_extract:
case INDEX_op_mov:
case INDEX_op_movcond:
case INDEX_op_negsetcond:
@ -2342,7 +2342,6 @@ bool tcg_op_supported(TCGOpcode op, TCGType type, unsigned flags)
case INDEX_op_st8_i32:
case INDEX_op_st16_i32:
case INDEX_op_st_i32:
case INDEX_op_extract_i32:
case INDEX_op_sextract_i32:
case INDEX_op_deposit_i32:
return true;
@ -2371,7 +2370,6 @@ bool tcg_op_supported(TCGOpcode op, TCGType type, unsigned flags)
case INDEX_op_st_i64:
case INDEX_op_ext_i32_i64:
case INDEX_op_extu_i32_i64:
case INDEX_op_extract_i64:
case INDEX_op_sextract_i64:
case INDEX_op_deposit_i64:
return TCG_TARGET_REG_BITS == 64;
@ -5519,8 +5517,7 @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op)
}
break;
case INDEX_op_extract_i32:
case INDEX_op_extract_i64:
case INDEX_op_extract:
{
const TCGOutOpExtract *out =
container_of(all_outop[op->opc], TCGOutOpExtract, base);

View File

@ -27,6 +27,7 @@
#define ctpop_tr glue(ctpop, TCG_TARGET_REG_BITS)
#define extract_tr glue(extract, TCG_TARGET_REG_BITS)
/*
* Enable TCI assertions only when debugging TCG (and without NDEBUG defined).
@ -656,9 +657,9 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
tci_args_rrrbb(insn, &r0, &r1, &r2, &pos, &len);
regs[r0] = deposit32(regs[r1], pos, len, regs[r2]);
break;
case INDEX_op_extract_i32:
case INDEX_op_extract:
tci_args_rrbb(insn, &r0, &r1, &pos, &len);
regs[r0] = extract32(regs[r1], pos, len);
regs[r0] = extract_tr(regs[r1], pos, len);
break;
case INDEX_op_sextract_i32:
tci_args_rrbb(insn, &r0, &r1, &pos, &len);
@ -772,10 +773,6 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
tci_args_rrrbb(insn, &r0, &r1, &r2, &pos, &len);
regs[r0] = deposit64(regs[r1], pos, len, regs[r2]);
break;
case INDEX_op_extract_i64:
tci_args_rrbb(insn, &r0, &r1, &pos, &len);
regs[r0] = extract64(regs[r1], pos, len);
break;
case INDEX_op_sextract_i64:
tci_args_rrbb(insn, &r0, &r1, &pos, &len);
regs[r0] = sextract64(regs[r1], pos, len);
@ -1057,8 +1054,7 @@ int print_insn_tci(bfd_vma addr, disassemble_info *info)
op_name, str_r(r0), str_r(r1), str_r(r2), pos, len);
break;
case INDEX_op_extract_i32:
case INDEX_op_extract_i64:
case INDEX_op_extract:
case INDEX_op_sextract_i32:
case INDEX_op_sextract_i64:
tci_args_rrbb(insn, &r0, &r1, &pos, &len);

View File

@ -436,10 +436,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type,
static void tcg_out_extract(TCGContext *s, TCGType type, TCGReg rd,
TCGReg rs, unsigned pos, unsigned len)
{
TCGOpcode opc = type == TCG_TYPE_I32 ?
INDEX_op_extract_i32 :
INDEX_op_extract_i64;
tcg_out_op_rrbb(s, opc, rd, rs, pos, len);
tcg_out_op_rrbb(s, INDEX_op_extract, rd, rs, pos, len);
}
static const TCGOutOpExtract outop_extract = {