public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc(refs/vendors/riscv/heads/gcc-13-with-riscv-opts)] RISC-V: Fold all the cond-move variants together
@ 2023-11-22 5:10 Jeff Law
0 siblings, 0 replies; only message in thread
From: Jeff Law @ 2023-11-22 5:10 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:136b1f355092bff70396ecfe07cdd1386d2fd800
commit 136b1f355092bff70396ecfe07cdd1386d2fd800
Author: Maciej W. Rozycki <macro@embecosm.com>
Date: Wed Nov 22 01:18:27 2023 +0000
RISC-V: Fold all the cond-move variants together
Code in `riscv_expand_conditional_move' for Ventana and Zicond targets
seems like bolted on as an afterthought rather than properly merged so
as to handle all the cases together.
Fold the existing code pieces together then (observing that for short
forward branch targets no integer comparisons need to be canonicalized),
letting T-Head targets produce branchless sequences for all the integer
comparisons rather than for equality ones only, and preparing for the
handling of floating-point comparisons here across all conditional-move
targets.
gcc/
* config/riscv/riscv.cc (riscv_expand_conditional_move): Unify
conditional-move handling across all the relevant targets.
(cherry picked from commit 413ebfd660793ce16f0e6173b38ad91cd2f0cf64)
Diff:
---
gcc/config/riscv/riscv.cc | 60 ++++++++++++++++++++---------------------------
1 file changed, 26 insertions(+), 34 deletions(-)
diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
index cdd1a4f9e25..4424882ec3a 100644
--- a/gcc/config/riscv/riscv.cc
+++ b/gcc/config/riscv/riscv.cc
@@ -4119,35 +4119,9 @@ riscv_expand_conditional_move (rtx dest, rtx op, rtx cons, rtx alt)
rtx_code code = GET_CODE (op);
rtx op0 = XEXP (op, 0);
rtx op1 = XEXP (op, 1);
- bool need_eq_ne_p = false;
-
- if (TARGET_XTHEADCONDMOV
- && GET_MODE_CLASS (mode) == MODE_INT
- && (GET_MODE (op) == mode || GET_MODE (op) == E_VOIDmode)
- && (GET_MODE (op0) == mode || CONST_INT_P (op0))
- && (GET_MODE (op1) == mode || CONST_INT_P (op1))
- && (code == EQ || code == NE))
- need_eq_ne_p = true;
-
- if (need_eq_ne_p
- || (TARGET_SFB_ALU && GET_MODE (op0) == word_mode))
- {
- riscv_emit_int_compare (&code, &op0, &op1, need_eq_ne_p);
- rtx cond = gen_rtx_fmt_ee (code, GET_MODE (op0), op0, op1);
-
- /* The expander is a bit loose in its specification of the true
- arm of the conditional move. That allows us to support more
- cases for extensions which are more general than SFB. But
- does mean we need to force CONS into a register at this point. */
- cons = force_reg (mode, cons);
- /* With XTheadCondMov we need to force ALT into a register too. */
- alt = force_reg (mode, alt);
- emit_insn (gen_rtx_SET (dest, gen_rtx_IF_THEN_ELSE (mode,
- cond, cons, alt)));
- return true;
- }
- else if (TARGET_ZICOND_LIKE
- && GET_MODE_CLASS (mode) == MODE_INT)
+
+ if ((TARGET_ZICOND_LIKE && GET_MODE_CLASS (mode) == MODE_INT)
+ || TARGET_SFB_ALU || TARGET_XTHEADCONDMOV)
{
machine_mode mode0 = GET_MODE (op0);
machine_mode mode1 = GET_MODE (op1);
@@ -4161,9 +4135,11 @@ riscv_expand_conditional_move (rtx dest, rtx op, rtx cons, rtx alt)
return false;
/* Canonicalize the comparison. It must be an equality comparison
- of integer operands. If it isn't, then emit an SCC instruction
+ of integer operands, or with SFB it can be any comparison of
+ integer operands. If it isn't, then emit an SCC instruction
so that we can then use an equality comparison against zero. */
- if (!equality_operator (op, VOIDmode) || !INTEGRAL_MODE_P (mode0))
+ if ((!TARGET_SFB_ALU && !equality_operator (op, VOIDmode))
+ || !INTEGRAL_MODE_P (mode0))
{
bool *invert_ptr = nullptr;
bool invert = false;
@@ -4195,10 +4171,26 @@ riscv_expand_conditional_move (rtx dest, rtx op, rtx cons, rtx alt)
op1 = XEXP (op, 1);
}
+ if (TARGET_SFB_ALU || TARGET_XTHEADCONDMOV)
+ {
+ riscv_emit_int_compare (&code, &op0, &op1, !TARGET_SFB_ALU);
+ rtx cond = gen_rtx_fmt_ee (code, GET_MODE (op0), op0, op1);
+
+ /* The expander is a bit loose in its specification of the true
+ arm of the conditional move. That allows us to support more
+ cases for extensions which are more general than SFB. But
+ does mean we need to force CONS into a register at this point. */
+ cons = force_reg (mode, cons);
+ /* With XTheadCondMov we need to force ALT into a register too. */
+ alt = force_reg (mode, alt);
+ emit_insn (gen_rtx_SET (dest, gen_rtx_IF_THEN_ELSE (mode, cond,
+ cons, alt)));
+ return true;
+ }
/* 0, reg or 0, imm */
- if (cons == CONST0_RTX (mode)
- && (REG_P (alt)
- || (CONST_INT_P (alt) && alt != CONST0_RTX (mode))))
+ else if (cons == CONST0_RTX (mode)
+ && (REG_P (alt)
+ || (CONST_INT_P (alt) && alt != CONST0_RTX (mode))))
{
riscv_emit_int_compare (&code, &op0, &op1, true);
rtx cond = gen_rtx_fmt_ee (code, GET_MODE (op0), op0, op1);
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2023-11-22 5:10 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-11-22 5:10 [gcc(refs/vendors/riscv/heads/gcc-13-with-riscv-opts)] RISC-V: Fold all the cond-move variants together Jeff Law
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).