From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: Song Gao <gaosong@loongson.cn>
Subject: [PATCH 26/35] tcg: Don't free vector results
Date: Mon, 6 Nov 2023 18:48:33 -0800 [thread overview]
Message-ID: <20231107024842.7650-27-richard.henderson@linaro.org> (raw)
In-Reply-To: <20231107024842.7650-1-richard.henderson@linaro.org>
Avoid reusing vector temporaries so that we may re-use them
when propagating stores to loads.
Reviewed-by: Song Gao <gaosong@loongson.cn>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
tcg/tcg-op-gvec.c | 112 ++++++++++++++++------------------------------
1 file changed, 38 insertions(+), 74 deletions(-)
diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c
index feb2d3686b..bb88943f79 100644
--- a/tcg/tcg-op-gvec.c
+++ b/tcg/tcg-op-gvec.c
@@ -561,7 +561,6 @@ static void do_dup(unsigned vece, uint32_t dofs, uint32_t oprsz,
tcg_gen_dupi_vec(vece, t_vec, in_c);
}
do_dup_store(type, dofs, oprsz, maxsz, t_vec);
- tcg_temp_free_vec(t_vec);
return;
}
@@ -1024,11 +1023,10 @@ static void expand_2_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
bool load_dest,
void (*fni)(unsigned, TCGv_vec, TCGv_vec))
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- TCGv_vec t1 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t0, tcg_env, aofs + i);
if (load_dest) {
tcg_gen_ld_vec(t1, tcg_env, dofs + i);
@@ -1036,8 +1034,6 @@ static void expand_2_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
fni(vece, t1, t0);
tcg_gen_st_vec(t1, tcg_env, dofs + i);
}
- tcg_temp_free_vec(t0);
- tcg_temp_free_vec(t1);
}
/* Expand OPSZ bytes worth of two-vector operands and an immediate operand
@@ -1047,11 +1043,10 @@ static void expand_2i_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
int64_t c, bool load_dest,
void (*fni)(unsigned, TCGv_vec, TCGv_vec, int64_t))
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- TCGv_vec t1 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t0, tcg_env, aofs + i);
if (load_dest) {
tcg_gen_ld_vec(t1, tcg_env, dofs + i);
@@ -1059,8 +1054,6 @@ static void expand_2i_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
fni(vece, t1, t0, c);
tcg_gen_st_vec(t1, tcg_env, dofs + i);
}
- tcg_temp_free_vec(t0);
- tcg_temp_free_vec(t1);
}
static void expand_2s_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
@@ -1068,11 +1061,10 @@ static void expand_2s_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
TCGv_vec c, bool scalar_first,
void (*fni)(unsigned, TCGv_vec, TCGv_vec, TCGv_vec))
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- TCGv_vec t1 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t0, tcg_env, aofs + i);
if (scalar_first) {
fni(vece, t1, c, t0);
@@ -1081,8 +1073,6 @@ static void expand_2s_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
}
tcg_gen_st_vec(t1, tcg_env, dofs + i);
}
- tcg_temp_free_vec(t0);
- tcg_temp_free_vec(t1);
}
/* Expand OPSZ bytes worth of three-operand operations using host vectors. */
@@ -1091,12 +1081,11 @@ static void expand_3_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t tysz, TCGType type, bool load_dest,
void (*fni)(unsigned, TCGv_vec, TCGv_vec, TCGv_vec))
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- TCGv_vec t1 = tcg_temp_new_vec(type);
- TCGv_vec t2 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
+ TCGv_vec t2 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t0, tcg_env, aofs + i);
tcg_gen_ld_vec(t1, tcg_env, bofs + i);
if (load_dest) {
@@ -1105,9 +1094,6 @@ static void expand_3_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
fni(vece, t2, t0, t1);
tcg_gen_st_vec(t2, tcg_env, dofs + i);
}
- tcg_temp_free_vec(t2);
- tcg_temp_free_vec(t1);
- tcg_temp_free_vec(t0);
}
/*
@@ -1120,12 +1106,11 @@ static void expand_3i_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
void (*fni)(unsigned, TCGv_vec, TCGv_vec, TCGv_vec,
int64_t))
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- TCGv_vec t1 = tcg_temp_new_vec(type);
- TCGv_vec t2 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
+ TCGv_vec t2 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t0, tcg_env, aofs + i);
tcg_gen_ld_vec(t1, tcg_env, bofs + i);
if (load_dest) {
@@ -1134,9 +1119,6 @@ static void expand_3i_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
fni(vece, t2, t0, t1, c);
tcg_gen_st_vec(t2, tcg_env, dofs + i);
}
- tcg_temp_free_vec(t0);
- tcg_temp_free_vec(t1);
- tcg_temp_free_vec(t2);
}
/* Expand OPSZ bytes worth of four-operand operations using host vectors. */
@@ -1146,13 +1128,12 @@ static void expand_4_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
void (*fni)(unsigned, TCGv_vec, TCGv_vec,
TCGv_vec, TCGv_vec))
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- TCGv_vec t1 = tcg_temp_new_vec(type);
- TCGv_vec t2 = tcg_temp_new_vec(type);
- TCGv_vec t3 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
+ TCGv_vec t2 = tcg_temp_new_vec(type);
+ TCGv_vec t3 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t1, tcg_env, aofs + i);
tcg_gen_ld_vec(t2, tcg_env, bofs + i);
tcg_gen_ld_vec(t3, tcg_env, cofs + i);
@@ -1162,10 +1143,6 @@ static void expand_4_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
tcg_gen_st_vec(t1, tcg_env, aofs + i);
}
}
- tcg_temp_free_vec(t3);
- tcg_temp_free_vec(t2);
- tcg_temp_free_vec(t1);
- tcg_temp_free_vec(t0);
}
/*
@@ -1178,23 +1155,18 @@ static void expand_4i_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
void (*fni)(unsigned, TCGv_vec, TCGv_vec,
TCGv_vec, TCGv_vec, int64_t))
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- TCGv_vec t1 = tcg_temp_new_vec(type);
- TCGv_vec t2 = tcg_temp_new_vec(type);
- TCGv_vec t3 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
+ TCGv_vec t2 = tcg_temp_new_vec(type);
+ TCGv_vec t3 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t1, tcg_env, aofs + i);
tcg_gen_ld_vec(t2, tcg_env, bofs + i);
tcg_gen_ld_vec(t3, tcg_env, cofs + i);
fni(vece, t0, t1, t2, t3, c);
tcg_gen_st_vec(t0, tcg_env, dofs + i);
}
- tcg_temp_free_vec(t3);
- tcg_temp_free_vec(t2);
- tcg_temp_free_vec(t1);
- tcg_temp_free_vec(t0);
}
/* Expand a vector two-operand operation. */
@@ -1732,7 +1704,6 @@ void tcg_gen_gvec_dup_mem(unsigned vece, uint32_t dofs, uint32_t aofs,
TCGv_vec t_vec = tcg_temp_new_vec(type);
tcg_gen_dup_mem_vec(vece, t_vec, tcg_env, aofs);
do_dup_store(type, dofs, oprsz, maxsz, t_vec);
- tcg_temp_free_vec(t_vec);
} else if (vece <= MO_32) {
TCGv_i32 in = tcg_temp_ebb_new_i32();
switch (vece) {
@@ -1766,7 +1737,6 @@ void tcg_gen_gvec_dup_mem(unsigned vece, uint32_t dofs, uint32_t aofs,
for (i = (aofs == dofs) * 16; i < oprsz; i += 16) {
tcg_gen_st_vec(in, tcg_env, dofs + i);
}
- tcg_temp_free_vec(in);
} else {
TCGv_i64 in0 = tcg_temp_ebb_new_i64();
TCGv_i64 in1 = tcg_temp_ebb_new_i64();
@@ -1796,7 +1766,6 @@ void tcg_gen_gvec_dup_mem(unsigned vece, uint32_t dofs, uint32_t aofs,
for (i = (aofs == dofs) * 32; i < oprsz; i += 32) {
tcg_gen_st_vec(in, tcg_env, dofs + i);
}
- tcg_temp_free_vec(in);
} else if (TCG_TARGET_HAS_v128) {
TCGv_vec in0 = tcg_temp_new_vec(TCG_TYPE_V128);
TCGv_vec in1 = tcg_temp_new_vec(TCG_TYPE_V128);
@@ -1807,8 +1776,6 @@ void tcg_gen_gvec_dup_mem(unsigned vece, uint32_t dofs, uint32_t aofs,
tcg_gen_st_vec(in0, tcg_env, dofs + i);
tcg_gen_st_vec(in1, tcg_env, dofs + i + 16);
}
- tcg_temp_free_vec(in0);
- tcg_temp_free_vec(in1);
} else {
TCGv_i64 in[4];
int j;
@@ -3136,15 +3103,14 @@ static void expand_2sh_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
TCGv_i32 shift,
void (*fni)(unsigned, TCGv_vec, TCGv_vec, TCGv_i32))
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t0, tcg_env, aofs + i);
- fni(vece, t0, t0, shift);
- tcg_gen_st_vec(t0, tcg_env, dofs + i);
+ fni(vece, t1, t0, shift);
+ tcg_gen_st_vec(t1, tcg_env, dofs + i);
}
- tcg_temp_free_vec(t0);
}
static void
@@ -3720,18 +3686,16 @@ static void expand_cmp_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t tysz,
TCGType type, TCGCond cond)
{
- TCGv_vec t0 = tcg_temp_new_vec(type);
- TCGv_vec t1 = tcg_temp_new_vec(type);
- uint32_t i;
+ for (uint32_t i = 0; i < oprsz; i += tysz) {
+ TCGv_vec t0 = tcg_temp_new_vec(type);
+ TCGv_vec t1 = tcg_temp_new_vec(type);
+ TCGv_vec t2 = tcg_temp_new_vec(type);
- for (i = 0; i < oprsz; i += tysz) {
tcg_gen_ld_vec(t0, tcg_env, aofs + i);
tcg_gen_ld_vec(t1, tcg_env, bofs + i);
- tcg_gen_cmp_vec(cond, vece, t0, t0, t1);
- tcg_gen_st_vec(t0, tcg_env, dofs + i);
+ tcg_gen_cmp_vec(cond, vece, t2, t0, t1);
+ tcg_gen_st_vec(t2, tcg_env, dofs + i);
}
- tcg_temp_free_vec(t1);
- tcg_temp_free_vec(t0);
}
void tcg_gen_gvec_cmp(TCGCond cond, unsigned vece, uint32_t dofs,
--
2.34.1
next prev parent reply other threads:[~2023-11-07 2:54 UTC|newest]
Thread overview: 39+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-11-07 2:48 [PATCH 00/35] tcg patch queue Richard Henderson
2023-11-07 2:48 ` [PATCH 01/35] accel/tcg: Move HMP info jit and info opcount code Richard Henderson
2023-11-07 2:48 ` [PATCH 02/35] tcg: Add C_N2_I1 Richard Henderson
2023-11-07 2:48 ` [PATCH 03/35] tcg/loongarch64: Use C_N2_I1 for INDEX_op_qemu_ld_a*_i128 Richard Henderson
2023-11-07 2:48 ` [PATCH 04/35] util: Add cpuinfo for loongarch64 Richard Henderson
2023-11-07 2:48 ` [PATCH 05/35] tcg/loongarch64: Use cpuinfo.h Richard Henderson
2023-11-07 2:48 ` [PATCH 06/35] host/include/loongarch64: Add atomic16 load and store Richard Henderson
2023-11-07 2:48 ` [PATCH 07/35] accel/tcg: Remove redundant case in store_atom_16 Richard Henderson
2023-11-07 2:48 ` [PATCH 08/35] accel/tcg: Fix condition for store_atom_insert_al16 Richard Henderson
2023-11-07 2:48 ` [PATCH 09/35] tcg: Mark tcg_gen_op* as noinline Richard Henderson
2023-11-07 2:48 ` [PATCH 10/35] tcg: Move tcg_gen_op* out of line Richard Henderson
2023-11-07 2:48 ` [PATCH 11/35] tcg: Move generic expanders " Richard Henderson
2023-11-07 2:48 ` [PATCH 12/35] tcg: Move 32-bit " Richard Henderson
2023-11-07 2:48 ` [PATCH 13/35] tcg: Move 64-bit " Richard Henderson
2023-11-07 2:48 ` [PATCH 14/35] tcg: Move vec_gen_* declarations to tcg-internal.h Richard Henderson
2023-11-07 2:48 ` [PATCH 15/35] tcg: Move tcg_gen_opN " Richard Henderson
2023-11-07 2:48 ` [PATCH 16/35] tcg: Unexport tcg_gen_op*_{i32,i64} Richard Henderson
2023-11-07 2:48 ` [PATCH 17/35] tcg: Move tcg_constant_* out of line Richard Henderson
2023-11-07 2:48 ` [PATCH 18/35] tcg: Move tcg_temp_new_*, tcg_global_mem_new_* " Richard Henderson
2023-11-07 2:48 ` [PATCH 19/35] tcg: Move tcg_temp_free_* " Richard Henderson
2023-11-07 2:48 ` [PATCH 20/35] tcg/mips: Split out tcg_out_setcond_int Richard Henderson
2023-11-07 2:48 ` [PATCH 21/35] tcg/mips: Always implement movcond Richard Henderson
2023-11-07 2:48 ` [PATCH 22/35] tcg: Remove TCG_TARGET_HAS_movcond_{i32,i64} Richard Henderson
2023-11-07 2:48 ` [PATCH 23/35] tcg/mips: Implement neg opcodes Richard Henderson
2023-11-07 2:48 ` [PATCH 24/35] tcg/loongarch64: " Richard Henderson
2023-11-07 2:48 ` [PATCH 25/35] tcg: Remove TCG_TARGET_HAS_neg_{i32,i64} Richard Henderson
2023-11-07 2:48 ` Richard Henderson [this message]
2023-11-07 2:48 ` [PATCH 27/35] tcg/optimize: Pipe OptContext into reset_ts Richard Henderson
2023-11-07 2:48 ` [PATCH 28/35] tcg/optimize: Split out cmp_better_copy Richard Henderson
2023-11-07 2:48 ` [PATCH 29/35] tcg/optimize: Optimize env memory operations Richard Henderson
2023-11-07 2:48 ` [PATCH 30/35] tcg: Eliminate duplicate env store operations Richard Henderson
2023-11-07 2:48 ` [PATCH 31/35] tcg/optimize: Split out arg_new_constant Richard Henderson
2023-11-07 2:48 ` [PATCH 32/35] tcg: Canonicalize subi to addi during opcode generation Richard Henderson
2023-11-07 2:48 ` [PATCH 33/35] tcg/optimize: Canonicalize subi to addi during optimization Richard Henderson
2023-11-07 2:48 ` [PATCH 34/35] tcg/optimize: Canonicalize sub2 with constants to add2 Richard Henderson
2023-11-07 2:48 ` [PATCH 35/35] tcg/sparc64: Implement tcg_out_extrl_i64_i32 Richard Henderson
2023-11-07 2:55 ` [PULL 00/35] tcg patch queue Richard Henderson
2023-11-07 3:06 ` Stefan Hajnoczi
2023-11-07 4:59 ` [PATCH " Stefan Hajnoczi
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=20231107024842.7650-27-richard.henderson@linaro.org \
--to=richard.henderson@linaro.org \
--cc=gaosong@loongson.cn \
--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).