On Tue, Oct 4, 2011 at 6:06 PM, H.J. Lu <hjl.to...@gmail.com> wrote:
>> OTOH, x86_64 and i686 targets can also benefit from this change. If
>> combine can't create more complex address (covered by lea), then it
>> will simply propagate memory operand back into the add insn. It looks
>> to me that we can't loose here, so:
>>
>> /* Improve address combine. */
>> if (code == PLUS && MEM_P (src2))
>> src2 = force_reg (mode, src2);
>>
>> Any opinions?
>>
>
> It doesn't work with 64bit libstdc++:
Yeah, yeah. ix86_output_mi_thunk has some ... issues.
Please try attached patch that introduces ix86_emit_binop and uses it
in a bunch of places.
Uros.
Index: i386-protos.h
===================================================================
--- i386-protos.h (revision 179506)
+++ i386-protos.h (working copy)
@@ -94,6 +94,7 @@ extern bool ix86_lea_outperforms (rtx, unsigned in
unsigned int, unsigned int);
extern bool ix86_avoid_lea_for_add (rtx, rtx[]);
extern bool ix86_avoid_lea_for_addr (rtx, rtx[]);
+extern void ix86_emit_binop (enum rtx_code, enum machine_mode, rtx, rtx);
extern void ix86_split_lea_for_addr (rtx[], enum machine_mode);
extern bool ix86_lea_for_add_ok (rtx, rtx[]);
extern bool ix86_vec_interleave_v2df_operator_ok (rtx operands[3], bool high);
Index: i386.c
===================================================================
--- i386.c (revision 179506)
+++ i386.c (working copy)
@@ -15727,6 +15727,10 @@ ix86_fixup_binary_operands (enum rtx_code code, en
if (MEM_P (src1) && !rtx_equal_p (dst, src1))
src1 = force_reg (mode, src1);
+ /* Improve address combine. */
+ if (code == PLUS && MEM_P (src2))
+ src2 = force_reg (mode, src2);
+
operands[1] = src1;
operands[2] = src2;
return dst;
@@ -16470,6 +16474,20 @@ ix86_avoid_lea_for_addr (rtx insn, rtx operands[])
return !ix86_lea_outperforms (insn, regno0, regno1, regno2, split_cost);
}
+/* Emit x86 binary operand CODE in mode MODE, where the first operand
+ matches destination. RTX includes clobber of FLAGS_REG. */
+
+extern void ix86_emit_binop (enum rtx_code code, enum machine_mode mode,
+ rtx dst, rtx src)
+{
+ rtx op, clob;
+
+ op = gen_rtx_SET (VOIDmode, dst, gen_rtx_fmt_ee (code, mode, dst, src));
+ clob = gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (CCmode, FLAGS_REG));
+
+ emit_insn (gen_rtx_PARALLEL (VOIDmode, gen_rtvec (2, op, clob)));
+}
+
/* Split lea instructions into a sequence of instructions
which are executed on ALU to avoid AGU stalls.
It is assumed that it is allowed to clobber flags register
@@ -16482,8 +16500,7 @@ ix86_split_lea_for_addr (rtx operands[], enum mach
unsigned int regno1 = INVALID_REGNUM;
unsigned int regno2 = INVALID_REGNUM;
struct ix86_address parts;
- rtx tmp, clob;
- rtvec par;
+ rtx tmp;
int ok, adds;
ok = ix86_decompose_address (operands[1], &parts);
@@ -16515,14 +16532,7 @@ ix86_split_lea_for_addr (rtx operands[], enum mach
gcc_assert (regno2 != regno0);
for (adds = parts.scale; adds > 0; adds--)
- {
- tmp = gen_rtx_PLUS (mode, operands[0], parts.index);
- tmp = gen_rtx_SET (VOIDmode, operands[0], tmp);
- clob = gen_rtx_CLOBBER (VOIDmode,
- gen_rtx_REG (CCmode, FLAGS_REG));
- par = gen_rtvec (2, tmp, clob);
- emit_insn (gen_rtx_PARALLEL (VOIDmode, par));
- }
+ ix86_emit_binop (PLUS, mode, operands[0], parts.index);
}
else
{
@@ -16531,30 +16541,14 @@ ix86_split_lea_for_addr (rtx operands[], enum mach
emit_insn (gen_rtx_SET (VOIDmode, operands[0], parts.index));
/* Use shift for scaling. */
- tmp = gen_rtx_ASHIFT (mode, operands[0],
- GEN_INT (exact_log2 (parts.scale)));
- tmp = gen_rtx_SET (VOIDmode, operands[0], tmp);
- clob = gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (CCmode, FLAGS_REG));
- par = gen_rtvec (2, tmp, clob);
- emit_insn (gen_rtx_PARALLEL (VOIDmode, par));
+ ix86_emit_binop (ASHIFT, mode, operands[0],
+ GEN_INT (exact_log2 (parts.scale)));
if (parts.base)
- {
- tmp = gen_rtx_PLUS (mode, operands[0], parts.base);
- tmp = gen_rtx_SET (VOIDmode, operands[0], tmp);
- clob = gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (CCmode,
FLAGS_REG));
- par = gen_rtvec (2, tmp, clob);
- emit_insn (gen_rtx_PARALLEL (VOIDmode, par));
- }
+ ix86_emit_binop (PLUS, mode, operands[0], parts.base);
if (parts.disp && parts.disp != const0_rtx)
- {
- tmp = gen_rtx_PLUS (mode, operands[0], parts.disp);
- tmp = gen_rtx_SET (VOIDmode, operands[0], tmp);
- clob = gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (CCmode,
FLAGS_REG));
- par = gen_rtvec (2, tmp, clob);
- emit_insn (gen_rtx_PARALLEL (VOIDmode, par));
- }
+ ix86_emit_binop (PLUS, mode, operands[0], parts.disp);
}
}
else if (!parts.base && !parts.index)
@@ -16565,41 +16559,32 @@ ix86_split_lea_for_addr (rtx operands[], enum mach
else
{
if (!parts.base)
- {
- if (regno0 != regno2)
- emit_insn (gen_rtx_SET (VOIDmode, operands[0], parts.index));
- }
+ {
+ if (regno0 != regno2)
+ emit_insn (gen_rtx_SET (VOIDmode, operands[0], parts.index));
+ }
else if (!parts.index)
- {
- if (regno0 != regno1)
- emit_insn (gen_rtx_SET (VOIDmode, operands[0], parts.base));
- }
- else
- {
- if (regno0 == regno1)
- tmp = gen_rtx_PLUS (mode, operands[0], parts.index);
- else if (regno0 == regno2)
- tmp = gen_rtx_PLUS (mode, operands[0], parts.base);
- else
- {
+ {
+ if (regno0 != regno1)
emit_insn (gen_rtx_SET (VOIDmode, operands[0], parts.base));
- tmp = gen_rtx_PLUS (mode, operands[0], parts.index);
- }
+ }
+ else
+ {
+ if (regno0 == regno1)
+ tmp = parts.index;
+ else if (regno0 == regno2)
+ tmp = parts.base;
+ else
+ {
+ emit_insn (gen_rtx_SET (VOIDmode, operands[0], parts.base));
+ tmp = parts.index;
+ }
- tmp = gen_rtx_SET (VOIDmode, operands[0], tmp);
- clob = gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (CCmode, FLAGS_REG));
- par = gen_rtvec (2, tmp, clob);
- emit_insn (gen_rtx_PARALLEL (VOIDmode, par));
- }
+ ix86_emit_binop (PLUS, mode, operands[0], tmp);
+ }
if (parts.disp && parts.disp != const0_rtx)
- {
- tmp = gen_rtx_PLUS (mode, operands[0], parts.disp);
- tmp = gen_rtx_SET (VOIDmode, operands[0], tmp);
- clob = gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (CCmode, FLAGS_REG));
- par = gen_rtvec (2, tmp, clob);
- emit_insn (gen_rtx_PARALLEL (VOIDmode, par));
- }
+ ix86_emit_binop (PLUS, mode, operands[0], parts.disp);
}
}
@@ -30940,7 +30925,7 @@ x86_output_mi_thunk (FILE *file,
}
}
- emit_insn (ix86_gen_add3 (delta_dst, delta_dst, delta_rtx));
+ ix86_emit_binop (PLUS, Pmode, delta_dst, delta_rtx);
}
/* Adjust the this parameter by a value stored in the vtable. */
@@ -30983,7 +30968,7 @@ x86_output_mi_thunk (FILE *file,
REGNO (this_reg)),
vcall_mem));
else
- emit_insn (ix86_gen_add3 (this_reg, this_reg, vcall_mem));
+ ix86_emit_binop (PLUS, Pmode, this_reg, vcall_mem);
}
/* If necessary, drop THIS back to its stack slot. */
