From: Richard Henderson <rth@twiddle.net>
To: qemu-devel@nongnu.org
Cc: peter.maydell@linaro.org
Subject: [Qemu-devel] [PULL 16/24] tcg-mips: Improve add2/sub2
Date: Sat, 24 May 2014 08:53:53 -0700 [thread overview]
Message-ID: <1400946841-21079-17-git-send-email-rth@twiddle.net> (raw)
In-Reply-To: <1400946841-21079-1-git-send-email-rth@twiddle.net>
Reduce insn count from 5 to either 3 or 4.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
---
tcg/mips/tcg-target.c | 87 +++++++++++++++++++++++++++++++++------------------
1 file changed, 56 insertions(+), 31 deletions(-)
diff --git a/tcg/mips/tcg-target.c b/tcg/mips/tcg-target.c
index 516c2b7..76a7852 100644
--- a/tcg/mips/tcg-target.c
+++ b/tcg/mips/tcg-target.c
@@ -1297,6 +1297,52 @@ static void tcg_out_qemu_st_direct(TCGContext *s, TCGReg datalo, TCGReg datahi,
}
}
+static void tcg_out_addsub2(TCGContext *s, TCGReg rl, TCGReg rh, TCGReg al,
+ TCGReg ah, TCGArg bl, TCGArg bh, bool cbl,
+ bool cbh, bool is_sub)
+{
+ TCGReg th = TCG_TMP1;
+
+ /* If we have a negative constant such that negating it would
+ make the high part zero, we can (usually) eliminate one insn. */
+ if (cbl && cbh && bh == -1 && bl != 0) {
+ bl = -bl;
+ bh = 0;
+ is_sub = !is_sub;
+ }
+
+ /* By operating on the high part first, we get to use the final
+ carry operation to move back from the temporary. */
+ if (!cbh) {
+ tcg_out_opc_reg(s, (is_sub ? OPC_SUBU : OPC_ADDU), th, ah, bh);
+ } else if (bh != 0 || ah == rl) {
+ tcg_out_opc_imm(s, OPC_ADDIU, th, ah, (is_sub ? -bh : bh));
+ } else {
+ th = ah;
+ }
+
+ /* Note that tcg optimization should eliminate the bl == 0 case. */
+ if (is_sub) {
+ if (cbl) {
+ tcg_out_opc_imm(s, OPC_SLTIU, TCG_TMP0, al, bl);
+ tcg_out_opc_imm(s, OPC_ADDIU, rl, al, -bl);
+ } else {
+ tcg_out_opc_reg(s, OPC_SLTU, TCG_TMP0, al, bl);
+ tcg_out_opc_reg(s, OPC_SUBU, rl, al, bl);
+ }
+ tcg_out_opc_reg(s, OPC_SUBU, rh, th, TCG_TMP0);
+ } else {
+ if (cbl) {
+ tcg_out_opc_imm(s, OPC_ADDIU, rl, al, bl);
+ tcg_out_opc_imm(s, OPC_SLTIU, TCG_TMP0, rl, bl);
+ } else {
+ tcg_out_opc_reg(s, OPC_ADDU, rl, al, bl);
+ tcg_out_opc_reg(s, OPC_SLTU, TCG_TMP0, rl, (rl == bl ? al : bl));
+ }
+ tcg_out_opc_reg(s, OPC_ADDU, rh, th, TCG_TMP0);
+ }
+}
+
static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
{
TCGReg addr_regl, addr_regh __attribute__((unused));
@@ -1419,21 +1465,6 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
tcg_out_opc_reg(s, OPC_ADDU, a0, a1, a2);
}
break;
- case INDEX_op_add2_i32:
- if (const_args[4]) {
- tcg_out_opc_imm(s, OPC_ADDIU, TCG_TMP0, a2, args[4]);
- } else {
- tcg_out_opc_reg(s, OPC_ADDU, TCG_TMP0, a2, args[4]);
- }
- tcg_out_opc_reg(s, OPC_SLTU, TCG_TMP1, TCG_TMP0, a2);
- if (const_args[5]) {
- tcg_out_opc_imm(s, OPC_ADDIU, a1, args[3], args[5]);
- } else {
- tcg_out_opc_reg(s, OPC_ADDU, a1, args[3], args[5]);
- }
- tcg_out_opc_reg(s, OPC_ADDU, a1, a1, TCG_TMP1);
- tcg_out_mov(s, TCG_TYPE_I32, a0, TCG_TMP0);
- break;
case INDEX_op_sub_i32:
if (c2) {
tcg_out_opc_imm(s, OPC_ADDIU, a0, a1, -a2);
@@ -1441,21 +1472,6 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
tcg_out_opc_reg(s, OPC_SUBU, a0, a1, a2);
}
break;
- case INDEX_op_sub2_i32:
- if (const_args[4]) {
- tcg_out_opc_imm(s, OPC_ADDIU, TCG_TMP0, a2, -args[4]);
- } else {
- tcg_out_opc_reg(s, OPC_SUBU, TCG_TMP0, a2, args[4]);
- }
- tcg_out_opc_reg(s, OPC_SLTU, TCG_TMP1, a2, TCG_TMP0);
- if (const_args[5]) {
- tcg_out_opc_imm(s, OPC_ADDIU, a1, args[3], -args[5]);
- } else {
- tcg_out_opc_reg(s, OPC_SUBU, a1, args[3], args[5]);
- }
- tcg_out_opc_reg(s, OPC_SUBU, a1, a1, TCG_TMP1);
- tcg_out_mov(s, TCG_TYPE_I32, a0, TCG_TMP0);
- break;
case INDEX_op_mul_i32:
if (use_mips32_instructions) {
tcg_out_opc_reg(s, OPC_MUL, a0, a1, a2);
@@ -1621,6 +1637,15 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
tcg_out_qemu_st(s, args, true);
break;
+ case INDEX_op_add2_i32:
+ tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
+ const_args[4], const_args[5], false);
+ break;
+ case INDEX_op_sub2_i32:
+ tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
+ const_args[4], const_args[5], true);
+ break;
+
case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */
case INDEX_op_call: /* Always emitted via tcg_out_call. */
@@ -1680,7 +1705,7 @@ static const TCGTargetOpDef mips_op_defs[] = {
{ INDEX_op_setcond_i32, { "r", "rZ", "rZ" } },
{ INDEX_op_setcond2_i32, { "r", "rZ", "rZ", "rZ", "rZ" } },
- { INDEX_op_add2_i32, { "r", "r", "rZ", "rZ", "rJ", "rJ" } },
+ { INDEX_op_add2_i32, { "r", "r", "rZ", "rZ", "rN", "rN" } },
{ INDEX_op_sub2_i32, { "r", "r", "rZ", "rZ", "rN", "rN" } },
{ INDEX_op_brcond2_i32, { "rZ", "rZ", "rZ", "rZ" } },
--
1.9.0
next prev parent reply other threads:[~2014-05-24 15:54 UTC|newest]
Thread overview: 25+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-05-24 15:53 [Qemu-devel] [PULL 00/24] tcg mips updates Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 01/24] tcg-mips: Layout executable and code_gen_buffer Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 02/24] tcg-mips: Constrain the code_gen_buffer to be within one 256mb segment Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 03/24] tcg-mips: Use J and JAL opcodes Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 04/24] tcg-mips: Fill the exit_tb delay slot Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 05/24] tcg-mips: Split large ldst offsets Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 06/24] tcg-mips: Move softmmu slow path out of line Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 07/24] tcg-mips: Convert to new qemu_l/st helpers Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 08/24] tcg-mips: Convert to new_ldst Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 09/24] tcg-mips: Rearrange register allocation Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 10/24] tcg-mips: Introduce TCG_TMP0, TCG_TMP1 Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 11/24] tcg-mips: Use T9 for TCG_TMP1 Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 12/24] tcg-mips: Use EXT for AND on mips32r2 Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 13/24] tcg-mips: Name the opcode enumeration Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 14/24] tcg-mips: Fix subtract immediate range Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 15/24] tcg-mips: Hoist args loads Richard Henderson
2014-05-24 15:53 ` Richard Henderson [this message]
2014-05-24 15:53 ` [Qemu-devel] [PULL 17/24] tcg-mips: Commonize opcode implementations Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 18/24] tcg-mips: Simplify setcond Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 19/24] tcg-mips: Simplify brcond Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 20/24] tcg-mips: Simplify setcond2 Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 21/24] tcg-mips: Improve setcond eq/ne vs zeros Richard Henderson
2014-05-24 15:53 ` [Qemu-devel] [PULL 22/24] tcg-mips: Simplify brcond2 Richard Henderson
2014-05-24 15:54 ` [Qemu-devel] [PULL 23/24] tcg-mips: Simplify movcond Richard Henderson
2014-05-24 15:54 ` [Qemu-devel] [PULL 24/24] tcg-mips: Enable direct chaining of TBs Richard Henderson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1400946841-21079-17-git-send-email-rth@twiddle.net \
--to=rth@twiddle.net \
--cc=peter.maydell@linaro.org \
--cc=qemu-devel@nongnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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 NNTP newsgroup(s).