qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Kirill Batuzov <batuzovk@ispras.ru>
To: qemu-devel@nongnu.org
Cc: "Richard Henderson" <rth@twiddle.net>,
	"Paolo Bonzini" <pbonzini@redhat.com>,
	"Peter Crosthwaite" <crosthwaite.peter@gmail.com>,
	"Peter Maydell" <peter.maydell@linaro.org>,
	"Andrzej Zaborowski" <balrogg@gmail.com>,
	"Alex Bennée" <alex.bennee@linaro.org>,
	"Kirill Batuzov" <batuzovk@ispras.ru>
Subject: [Qemu-devel] [PATCH v2.1 05/21] tcg: add simple alias analysis
Date: Thu,  2 Feb 2017 17:34:43 +0300	[thread overview]
Message-ID: <1486046099-17726-6-git-send-email-batuzovk@ispras.ru> (raw)
In-Reply-To: <1486046099-17726-1-git-send-email-batuzovk@ispras.ru>

Add a simple alias analysis to TCG which finds out memory loads and stores
that overlap with CPUState. This information can be used later in liveness
analysis to ensure correctness of register allocation. In particular, if load
or store overlaps with memory location of some global variable, this variable
should be spilled and reloaded at appropriate times.

Previously no such analysis was performed and for correctness reasons it was
required that no load/store operations overlap with memory locations of global
variables.

Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
---

I believe checkpatch warning here to be false-positive.

---
 tcg/optimize.c | 146 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 tcg/tcg.h      |  17 +++++++
 2 files changed, 163 insertions(+)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index adfc56c..2347ce3 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -34,6 +34,7 @@
 
 struct tcg_temp_info {
     bool is_const;
+    bool is_base;
     uint16_t prev_copy;
     uint16_t next_copy;
     tcg_target_ulong val;
@@ -61,6 +62,7 @@ static void reset_temp(TCGArg temp)
     temps[temp].next_copy = temp;
     temps[temp].prev_copy = temp;
     temps[temp].is_const = false;
+    temps[temp].is_base = false;
     temps[temp].mask = -1;
 }
 
@@ -1429,3 +1431,147 @@ void tcg_optimize(TCGContext *s)
         }
     }
 }
+
+/* Simple alias analysis. It finds out which load/store operations overlap
+   with CPUArchState. The result is stored in TCGContext and can be used
+   during liveness analysis and register allocation. */
+void tcg_alias_analysis(TCGContext *s)
+{
+    int oi, oi_next;
+
+    reset_all_temps(s->nb_temps);
+    temps[GET_TCGV_PTR(s->tcg_env)].is_base = true;
+    temps[GET_TCGV_PTR(s->tcg_env)].val = 0;
+
+    for (oi = s->gen_op_buf[0].next; oi != 0; oi = oi_next) {
+        int nb_oargs, i;
+        int size;
+        TCGAliasType tp;
+
+        TCGOp * const op = &s->gen_op_buf[oi];
+        TCGArg * const args = &s->gen_opparam_buf[op->args];
+        TCGOpcode opc = op->opc;
+        const TCGOpDef *def = &tcg_op_defs[opc];
+
+        oi_next = op->next;
+
+        if (opc == INDEX_op_call) {
+            nb_oargs = op->callo;
+        } else {
+            nb_oargs = def->nb_oargs;
+        }
+
+        s->alias_info[oi] = (TCGAliasInfo){
+                TCG_NOT_ALIAS,
+                false,
+                0,
+                0
+            };
+
+        switch (opc) {
+        CASE_OP_32_64(movi):
+            temps[args[0]].is_const = 1;
+            temps[args[0]].val = args[1];
+            break;
+        CASE_OP_32_64(mov):
+            temps[args[0]].is_const = temps[args[1]].is_const;
+            temps[args[0]].is_base = temps[args[1]].is_base;
+            temps[args[0]].val = temps[args[1]].val;
+            break;
+        CASE_OP_32_64(add):
+        CASE_OP_32_64(sub):
+            if (temps[args[1]].is_base && temps[args[2]].is_const) {
+                temps[args[0]].is_base = true;
+                temps[args[0]].is_const = false;
+                temps[args[0]].val =
+                    do_constant_folding(opc, temps[args[1]].val,
+                                        temps[args[2]].val);
+            } else {
+                reset_temp(args[0]);
+            }
+        CASE_OP_32_64(ld8s):
+        CASE_OP_32_64(ld8u):
+            size = 1;
+            tp = TCG_ALIAS_READ;
+            goto do_ldst;
+        CASE_OP_32_64(ld16s):
+        CASE_OP_32_64(ld16u):
+            size = 2;
+            tp = TCG_ALIAS_READ;
+            goto do_ldst;
+        case INDEX_op_ld_i32:
+        case INDEX_op_ld32s_i64:
+        case INDEX_op_ld32u_i64:
+            size = 4;
+            tp = TCG_ALIAS_READ;
+            goto do_ldst;
+        case INDEX_op_ld_i64:
+            size = 8;
+            tp = TCG_ALIAS_READ;
+            goto do_ldst;
+        case INDEX_op_ld_v128:
+            size = 16;
+            tp = TCG_ALIAS_READ;
+            goto do_ldst;
+        CASE_OP_32_64(st8):
+            size = 1;
+            tp = TCG_ALIAS_WRITE;
+            goto do_ldst;
+        CASE_OP_32_64(st16):
+            size = 2;
+            tp = TCG_ALIAS_WRITE;
+            goto do_ldst;
+        case INDEX_op_st_i32:
+        case INDEX_op_st32_i64:
+            size = 4;
+            tp = TCG_ALIAS_WRITE;
+            goto do_ldst;
+        case INDEX_op_st_i64:
+            size = 8;
+            tp = TCG_ALIAS_WRITE;
+            goto do_ldst;
+        case INDEX_op_st_v128:
+            size = 16;
+            tp = TCG_ALIAS_WRITE;
+            goto do_ldst;
+        do_ldst:
+            if (temps[args[1]].is_base) {
+                TCGArg val;
+#if TCG_TARGET_REG_BITS == 32
+                val = do_constant_folding(INDEX_op_add_i32,
+                                          temps[args[1]].val,
+                                          args[2]);
+#else
+                val = do_constant_folding(INDEX_op_add_i64,
+                                          temps[args[1]].val,
+                                          args[2]);
+#endif
+                if ((tcg_target_long)val < sizeof(CPUArchState) &&
+                    (tcg_target_long)val + size > 0) {
+                    s->alias_info[oi].alias_type = tp;
+                    s->alias_info[oi].fixed_offset = true;
+                    s->alias_info[oi].offset = val;
+                    s->alias_info[oi].size = size;
+                } else {
+                    s->alias_info[oi].alias_type = TCG_NOT_ALIAS;
+                }
+            } else {
+                s->alias_info[oi].alias_type = tp;
+                s->alias_info[oi].fixed_offset = false;
+            }
+            goto do_reset_output;
+        default:
+            if (def->flags & TCG_OPF_BB_END) {
+                reset_all_temps(s->nb_temps);
+                temps[GET_TCGV_PTR(s->tcg_env)].is_base = true;
+                temps[GET_TCGV_PTR(s->tcg_env)].val = 0;
+            } else {
+        do_reset_output:
+                for (i = 0; i < nb_oargs; i++) {
+                    reset_temp(args[i]);
+                }
+            }
+            break;
+        }
+    }
+}
diff --git a/tcg/tcg.h b/tcg/tcg.h
index fa455ae..0e1fbe9 100644
--- a/tcg/tcg.h
+++ b/tcg/tcg.h
@@ -678,6 +678,20 @@ QEMU_BUILD_BUG_ON(OPPARAM_BUF_SIZE > (1 << 14));
 /* Make sure that we don't overflow 64 bits without noticing.  */
 QEMU_BUILD_BUG_ON(sizeof(TCGOp) > 8);
 
+typedef enum TCGAliasType {
+    TCG_NOT_ALIAS = 0,
+    TCG_ALIAS_READ = 1,
+    TCG_ALIAS_WRITE = 2,
+    TCG_ALIAS_RW = TCG_ALIAS_READ | TCG_ALIAS_WRITE
+} TCGAliasType;
+
+typedef struct TCGAliasInfo {
+    TCGAliasType alias_type;
+    bool fixed_offset;
+    tcg_target_long offset;
+    tcg_target_long size;
+} TCGAliasInfo;
+
 struct TCGContext {
     uint8_t *pool_cur, *pool_end;
     TCGPool *pool_first, *pool_current, *pool_first_large;
@@ -762,6 +776,8 @@ struct TCGContext {
     TCGOp gen_op_buf[OPC_BUF_SIZE];
     TCGArg gen_opparam_buf[OPPARAM_BUF_SIZE];
 
+    TCGAliasInfo alias_info[OPC_BUF_SIZE];
+
     uint16_t gen_insn_end_off[TCG_MAX_INSNS];
     target_ulong gen_insn_data[TCG_MAX_INSNS][TARGET_INSN_START_WORDS];
 };
@@ -1009,6 +1025,7 @@ TCGOp *tcg_op_insert_before(TCGContext *s, TCGOp *op, TCGOpcode opc, int narg);
 TCGOp *tcg_op_insert_after(TCGContext *s, TCGOp *op, TCGOpcode opc, int narg);
 
 void tcg_optimize(TCGContext *s);
+void tcg_alias_analysis(TCGContext *s);
 
 /* only used for debugging purposes */
 void tcg_dump_ops(TCGContext *s);
-- 
2.1.4

  parent reply	other threads:[~2017-02-02 14:35 UTC|newest]

Thread overview: 28+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-02-02 14:34 [Qemu-devel] [PATCH v2.1 00/20] Emulate guest vector operations with host vector operations Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 01/21] tcg: add support for 128bit vector type Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 02/21] tcg: add support for 64bit " Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 03/21] tcg: support representing vector type with smaller vector or scalar types Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 04/21] tcg: add ld_v128, ld_v64, st_v128 and st_v64 opcodes Kirill Batuzov
2017-02-02 14:34 ` Kirill Batuzov [this message]
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 06/21] tcg: use results of alias analysis in liveness analysis Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 07/21] tcg: allow globals to overlap Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 08/21] tcg: add vector addition operations Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 09/21] target/arm: support access to vector guest registers as globals Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 10/21] target/arm: use vector opcode to handle vadd.<size> instruction Kirill Batuzov
2017-02-09 13:19   ` Philippe Mathieu-Daudé
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 11/21] tcg/i386: add support for vector opcodes Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 12/21] tcg/i386: support 64-bit vector operations Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 13/21] tcg/i386: support remaining vector addition operations Kirill Batuzov
     [not found]   ` <2089cbe3-0e9b-fae2-0e35-224f2765dc28@amsat.org>
     [not found]     ` <32a902a1-e8c7-c2f7-ac66-148e02ee0b2d@amsat.org>
2017-02-21 13:29       ` Kirill Batuzov
2017-02-21 16:21         ` Alex Bennée
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 14/21] tcg: do not rely on exact values of MO_BSWAP or MO_SIGN in backend Kirill Batuzov
2017-05-05 13:59   ` Alex Bennée
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 15/21] target/aarch64: do not check for non-existent TCGMemOp Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 16/21] tcg: introduce new TCGMemOp - MO_128 Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 17/21] tcg: introduce qemu_ld_v128 and qemu_st_v128 opcodes Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 18/21] softmmu: create helpers for vector loads Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 19/21] tcg/i386: add support for qemu_ld_v128/qemu_st_v128 ops Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 20/21] target/arm: load two consecutive 64-bits vector regs as a 128-bit vector reg Kirill Batuzov
2017-02-02 14:34 ` [Qemu-devel] [PATCH v2.1 21/21] tcg/README: update README to include information about vector opcodes Kirill Batuzov
2017-02-02 15:25 ` [Qemu-devel] [PATCH v2.1 00/20] Emulate guest vector operations with host vector operations no-reply
2017-02-21 12:19 ` Kirill Batuzov

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=1486046099-17726-6-git-send-email-batuzovk@ispras.ru \
    --to=batuzovk@ispras.ru \
    --cc=alex.bennee@linaro.org \
    --cc=balrogg@gmail.com \
    --cc=crosthwaite.peter@gmail.com \
    --cc=pbonzini@redhat.com \
    --cc=peter.maydell@linaro.org \
    --cc=qemu-devel@nongnu.org \
    --cc=rth@twiddle.net \
    /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).