AArch64: Remove AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS

Remove the tune AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS since it is only
used by an old core and doesn't properly support -Os.  SPECINT_2017
shows that removing it has no performance difference, while codesize
is reduced by 0.07%.

gcc/ChangeLog:
	* config/aarch64/aarch64.cc (aarch64_mode_valid_for_sched_fusion_p):
	Remove check for AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS.
	(aarch64_advsimd_ldp_stp_p): Likewise.
	(aarch64_stp_sequence_cost): Likewise.
	(aarch64_expand_cpymem): Likewise.
	(aarch64_expand_setmem): Likewise.
	* config/aarch64/aarch64-ldp-fusion.cc (ldp_operand_mode_ok_p):
	Likewise.
	* config/aarch64/aarch64-ldpstp.md: Likewise.
	* config/aarch64/aarch64-tuning-flags.def: Remove NO_LDP_STP_QREGS.
	* config/aarch64/tuning_models/emag.h: Likewise.
	* config/aarch64/tuning_models/xgene1.h: Likewise.

gcc/testsuite:
	* gcc.target/aarch64/ldp_stp_q_disable.c: Remove test.
This commit is contained in:
Wilco Dijkstra 2024-04-26 15:09:31 +01:00
parent 5716f8daf3
commit 768fbb56b3
7 changed files with 8 additions and 63 deletions

View file

@ -315,17 +315,9 @@ any_post_modify_p (rtx x)
static bool
ldp_operand_mode_ok_p (machine_mode mode)
{
const bool allow_qregs
= !(aarch64_tune_params.extra_tuning_flags
& AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS);
if (!aarch64_ldpstp_operand_mode_p (mode))
return false;
const auto size = GET_MODE_SIZE (mode).to_constant ();
if (size == 16 && !allow_qregs)
return false;
// We don't pair up TImode accesses before RA because TImode is
// special in that it can be allocated to a pair of GPRs or a single
// FPR, and the RA is best placed to make that decision.

View file

@ -96,9 +96,7 @@
(set (match_operand:VQ2 2 "register_operand" "")
(match_operand:VQ2 3 "memory_operand" ""))]
"TARGET_FLOAT
&& aarch64_operands_ok_for_ldpstp (operands, true)
&& (aarch64_tune_params.extra_tuning_flags
& AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS) == 0"
&& aarch64_operands_ok_for_ldpstp (operands, true)"
[(const_int 0)]
{
aarch64_finish_ldpstp_peephole (operands, true);
@ -111,9 +109,7 @@
(set (match_operand:VQ2 2 "memory_operand" "")
(match_operand:VQ2 3 "register_operand" ""))]
"TARGET_FLOAT
&& aarch64_operands_ok_for_ldpstp (operands, false)
&& (aarch64_tune_params.extra_tuning_flags
& AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS) == 0"
&& aarch64_operands_ok_for_ldpstp (operands, false)"
[(const_int 0)]
{
aarch64_finish_ldpstp_peephole (operands, false);

View file

@ -36,9 +36,6 @@ AARCH64_EXTRA_TUNING_OPTION ("rename_fma_regs", RENAME_FMA_REGS)
are not considered cheap. */
AARCH64_EXTRA_TUNING_OPTION ("cheap_shift_extend", CHEAP_SHIFT_EXTEND)
/* Disallow load/store pair instructions on Q-registers. */
AARCH64_EXTRA_TUNING_OPTION ("no_ldp_stp_qregs", NO_LDP_STP_QREGS)
AARCH64_EXTRA_TUNING_OPTION ("rename_load_regs", RENAME_LOAD_REGS)
AARCH64_EXTRA_TUNING_OPTION ("cse_sve_vl_constants", CSE_SVE_VL_CONSTANTS)

View file

@ -10400,9 +10400,7 @@ aarch64_mode_valid_for_sched_fusion_p (machine_mode mode)
|| mode == SDmode || mode == DDmode
|| (aarch64_vector_mode_supported_p (mode)
&& (known_eq (GET_MODE_SIZE (mode), 8)
|| (known_eq (GET_MODE_SIZE (mode), 16)
&& (aarch64_tune_params.extra_tuning_flags
& AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS) == 0)));
|| known_eq (GET_MODE_SIZE (mode), 16)));
}
/* Return true if REGNO is a virtual pointer register, or an eliminable
@ -16519,10 +16517,6 @@ aarch64_advsimd_ldp_stp_p (enum vect_cost_for_stmt kind,
return false;
}
if (aarch64_tune_params.extra_tuning_flags
& AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS)
return false;
return is_gimple_assign (stmt_info->stmt);
}
@ -17170,9 +17164,6 @@ aarch64_stp_sequence_cost (unsigned int count, vect_cost_for_stmt kind,
/* Count 1 insn per vector if we can't form STP Q pairs. */
if (aarch64_sve_mode_p (TYPE_MODE (vectype)))
return count * 2;
if (aarch64_tune_params.extra_tuning_flags
& AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS)
return count * 2;
if (stmt_info)
{
@ -26625,11 +26616,9 @@ aarch64_expand_cpymem (rtx *operands, bool is_memmove)
return aarch64_expand_cpymem_mops (operands, is_memmove);
unsigned HOST_WIDE_INT size = UINTVAL (operands[2]);
bool use_ldpq = TARGET_SIMD && !(aarch64_tune_params.extra_tuning_flags
& AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS);
/* Set inline limits for memmove/memcpy. MOPS has a separate threshold. */
unsigned max_copy_size = use_ldpq ? 256 : 128;
unsigned max_copy_size = TARGET_SIMD ? 256 : 128;
unsigned mops_threshold = is_memmove ? aarch64_mops_memmove_size_threshold
: aarch64_mops_memcpy_size_threshold;
@ -26834,11 +26823,8 @@ aarch64_expand_setmem (rtx *operands)
/* Convert len to bits to make the rest of the code simpler. */
n = len * BITS_PER_UNIT;
/* Maximum amount to copy in one go. We allow 256-bit chunks based on the
AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS tuning parameter. */
const int copy_limit = (aarch64_tune_params.extra_tuning_flags
& AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS)
? GET_MODE_BITSIZE (TImode) : 256;
/* Maximum amount to copy in one go. We allow 256-bit chunks. */
const int copy_limit = 256;
while (n > 0)
{

View file

@ -51,7 +51,7 @@ static const struct tune_params emag_tunings =
2, /* min_div_recip_mul_df. */
17, /* max_case_values. */
tune_params::AUTOPREFETCHER_OFF, /* autoprefetcher_model. */
(AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS), /* tune_flags. */
(AARCH64_EXTRA_TUNE_NONE), /* tune_flags. */
&xgene1_prefetch_tune,
AARCH64_LDP_STP_POLICY_ALWAYS, /* ldp_policy_model. */
AARCH64_LDP_STP_POLICY_ALWAYS /* stp_policy_model. */

View file

@ -136,7 +136,7 @@ static const struct tune_params xgene1_tunings =
2, /* min_div_recip_mul_df. */
17, /* max_case_values. */
tune_params::AUTOPREFETCHER_OFF, /* autoprefetcher_model. */
(AARCH64_EXTRA_TUNE_NO_LDP_STP_QREGS), /* tune_flags. */
(AARCH64_EXTRA_TUNE_NONE), /* tune_flags. */
&xgene1_prefetch_tune,
AARCH64_LDP_STP_POLICY_ALWAYS, /* ldp_policy_model. */
AARCH64_LDP_STP_POLICY_ALWAYS /* stp_policy_model. */

View file

@ -1,26 +0,0 @@
/* { dg-options "-O2 -moverride=tune=no_ldp_stp_qregs" } */
typedef float float32x4_t __attribute__ ((__vector_size__ ((16))));
float32x4_t arr[4][4];
void
foo (float32x4_t x, float32x4_t y)
{
arr[0][1] = x;
arr[1][0] = y;
arr[2][0] = x;
arr[1][1] = y;
arr[0][2] = x;
arr[0][3] = y;
arr[1][2] = x;
arr[2][1] = y;
arr[3][0] = x;
arr[3][1] = y;
arr[2][2] = x;
arr[1][3] = y;
arr[2][3] = x;
arr[3][2] = y;
}
/* { dg-final { scan-assembler-not "stp\tq\[0-9\]+, q\[0-9\]" } } */