diff mbox

[ARM] Refactor neon-schedgen.ml and regenerate A8 neon descriptions.

Message ID 1282723669.22948.102.camel@e102325-lin.cambridge.arm.com
State New
Headers show

Commit Message

Ramana Radhakrishnan Aug. 25, 2010, 8:07 a.m. UTC 
Hi,

This patch refactors neon-schedgen a bit to allow support for multiple
cores and generate multiple scheduler descriptions out of a single
source. I am only submitting the regenerated A8 descriptions and the
changes to the ML. 

The major change to neon-schedgen.ml is the addition of a core type,
thus adding support for a new core is essentially adding the new core
type, adding the core to the list of allCores and then ending up
regenerating all the insn_reservations and the bypasses from this. The
only bits that are still hand-written are the bits that bolt these
reservations and bypasses with the interactions with the other parts of
the scheduler description.

Each scheduler description for Neon will now have reservations with the
core name prefix'd to the insn_reservation names and would also check
for tuning for the core that it needs to generate this with. I must
admit that I wasn't originally a huge fan of the ML description and
generating scheduler descriptions and considered removing it but then
having played with a Neon description for another core and then
integrating it with the other bits and pieces actually got me thinking.

I haven't gone as far as getting a scheduler description written out
into a separate file in it's own right or the ml program taking the core
name as a command line argument yet or integrating this with the build
system but that's the matter for a follow up patch. 

Ok to commit ? 

cheers
Ramana


         * config/arm/neon-schedgen.ml (core): New type.
            (allCores): List of supported cores.
            (availability_table): Add supported cores. For now
            mark support for all cores.
            (collate_bypasses): Accept core as a parameter.
            (worst_case_latencies_and_bypasses): Accept core as a
parameter.
            (emit_insn_reservations): Accept core as a parameter.
            Use tuneStr and coreStr to get tune attribute and prefix
            for functional units.
            (emit_bypasses): Accept core name and use it.
            (calculate_per_core_availability_table): New.
        (filter_core): New.
            (calculate_core_availability_table): New.
            (main): Use calculate_core_availablity_table.
         * config/arm/cortex-a8-neon.md: Update copyright year.
        Regenerated from ml file and merged in.
	(neon_mrrc, neon_mrc): Rename to cortex_a8_neon_mrrc and
cortex_a8_neon_mrc.

Comments

Julian Brown Aug. 25, 2010, 11:04 a.m. UTC | #1 
On Wed, 25 Aug 2010 09:07:49 +0100
Ramana Radhakrishnan <ramana.radhakrishnan@arm.com> wrote:

> Hi,
> 
> This patch refactors neon-schedgen a bit to allow support for multiple
> cores and generate multiple scheduler descriptions out of a single
> source. I am only submitting the regenerated A8 descriptions and the
> changes to the ML. 
> 
> The major change to neon-schedgen.ml is the addition of a core type,
> thus adding support for a new core is essentially adding the new core
> type, adding the core to the list of allCores and then ending up
> regenerating all the insn_reservations and the bypasses from this. The
> only bits that are still hand-written are the bits that bolt these
> reservations and bypasses with the interactions with the other parts
> of the scheduler description.

I might have the wrong end of the stick here but... neon.md is filled
with "neon_type" specifications which are already hairy and verbose,
e.g.:

  [(set (attr "neon_type")
      (if_then_else (ne (symbol_ref "<Is_float_mode>") (const_int 0))
                    (if_then_else (ne (symbol_ref "<Is_d_reg>") (const_int 0))
                                  (const_string "neon_fp_vadd_ddd_vabs_dd")
                                  (const_string "neon_fp_vadd_qqq_vabs_qq"))
                    (if_then_else (ne (symbol_ref "<Is_d_reg>") (const_int 0))
                                  (if_then_else
                                    (ne (symbol_ref "<Scalar_mul_8_16>") (const_int 0))
                                    (const_string "neon_mul_ddd_8_16_qdd_16_8_long_32_16_long")
                                    (const_string "neon_mul_qqq_8_16_32_ddd_32"))
                                  (if_then_else (ne (symbol_ref "<Scalar_mul_8_16>") (const_int 0))
                                    (const_string "neon_mul_qqq_8_16_32_ddd_32")
                                    (const_string "neon_mul_qqq_8_16_32_ddd_32")))))]

It seems like adding cortex_a8 to the start of each of these won't
exactly help make these clearer -- and what are we going to do about
new cores? The real problem here is that these "neon_types" are far too
strongly tied to one physical implementation of the NEON ISA. My
(admittedly vague) notion is that we should be moving towards more
"logical" units for the neon instructions, decoupling them from any
particular implementation, and hopefully cut down on these rather opaque
conditionals whilst doing it. Or is that in fact what you are
suggesting?

If you're going to hack neon-schedgen.ml (which I didn't write, and
admit I don't have any great understanding of), would it make sense to
create a new table for Cortex-A9 (etc.), rather than mixing
definitions for all cores together in the existing availability_table?

Also a minor point on the ML: you can write,

let coreStr core = 
                 match core with
                   CortexA8 -> "cortex_a8" 
                 | CortexA9 -> "cortex_a9"

as:

let coreStr = function
    CortexA8 -> "cortex_a8"
  | CortexA9 -> "cortex_a9"

I know there's no GNU style guide for ML, but please try to follow the
existing formatting (re: indentation and so forth) if possible.

Cheers,

Julian
Ramana Radhakrishnan Aug. 25, 2010, 1:07 p.m. UTC | #2 
On Wed, 2010-08-25 at 12:04 +0100, Julian Brown wrote:
> On Wed, 25 Aug 2010 09:07:49 +0100
> Ramana Radhakrishnan <ramana.radhakrishnan@arm.com> wrote:
> 
> > Hi,
> > 
> > This patch refactors neon-schedgen a bit to allow support for multiple
> > cores and generate multiple scheduler descriptions out of a single
> > source. I am only submitting the regenerated A8 descriptions and the
> > changes to the ML. 
> > 
> > The major change to neon-schedgen.ml is the addition of a core type,
> > thus adding support for a new core is essentially adding the new core
> > type, adding the core to the list of allCores and then ending up
> > regenerating all the insn_reservations and the bypasses from this. The
> > only bits that are still hand-written are the bits that bolt these
> > reservations and bypasses with the interactions with the other parts
> > of the scheduler description.
> 
> I might have the wrong end of the stick here but... neon.md is filled
> with "neon_type" specifications which are already hairy and verbose,

No that is not what I'm doing. 

> e.g.:
> 
>   [(set (attr "neon_type")
>       (if_then_else (ne (symbol_ref "<Is_float_mode>") (const_int 0))
>                     (if_then_else (ne (symbol_ref "<Is_d_reg>") (const_int 0))
>                                   (const_string "neon_fp_vadd_ddd_vabs_dd")
>                                   (const_string "neon_fp_vadd_qqq_vabs_qq"))
>                     (if_then_else (ne (symbol_ref "<Is_d_reg>") (const_int 0))
>                                   (if_then_else
>                                     (ne (symbol_ref "<Scalar_mul_8_16>") (const_int 0))
>                                     (const_string "neon_mul_ddd_8_16_qdd_16_8_long_32_16_long")
>                                     (const_string "neon_mul_qqq_8_16_32_ddd_32"))
>                                   (if_then_else (ne (symbol_ref "<Scalar_mul_8_16>") (const_int 0))
>                                     (const_string "neon_mul_qqq_8_16_32_ddd_32")
>                                     (const_string "neon_mul_qqq_8_16_32_ddd_32")))))]
> 
> It seems like adding cortex_a8 to the start of each of these won't
> exactly help make these clearer -- and what are we going to do about
> new cores? The real problem here is that these "neon_types" are far too
> strongly tied to one physical implementation of the NEON ISA. My
> (admittedly vague) notion is that we should be moving towards more
> "logical" units for the neon instructions, decoupling them from any
> particular implementation, and hopefully cut down on these rather opaque
> conditionals whilst doing it. Or is that in fact what you are
> suggesting?

Errr.. IIUC neon-schedgen.ml is responsible only for generating the
scheduler descriptions that use *neon_types* . All that this does is
prefix the define_insn_reservation names. You need a per-core
insn_reservation to be able to describe how it interacts with the rest
of the core and the latency might well be different in different cores. 

What I have done is pretty much the following and a rather straight
forward renaming exercise.

< a small hunk from the new A8 neon scheduler description is as below> 

-(define_insn_reservation "neon_int_1" 3
+(define_insn_reservation "cortex_a8_neon_int_1" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_int_1"))
   "cortex_a8_neon_dp")


I'm not trying to change the values that "neon_types" can take. Notice
the cortex_a8 prefix is only for the insn reservations. I'm trying to
get to a point where these reservations can be made on a per-core basis.
neon-schedgen.ml generates define_insn_reservations and define_bypasses
on functional units in the A8 neon description that *use* the neon_types
to decide which instructions are mapped to which functional units. 





> 
> If you're going to hack neon-schedgen.ml (which I didn't write, and
> admit I don't have any great understanding of), would it make sense to
> create a new table for Cortex-A9 (etc.), rather than mixing
> definitions for all cores together in the existing availability_table?

There are situations where we might want to use common aspects of the engine in the table and thus a filter function appeared to be good enough
on the table rather than splitting this table into a "common across all cores" bit, and a per core bit.

> 
> Also a minor point on the ML: you can write,
> 
> let coreStr core = 
>                  match core with
>                    CortexA8 -> "cortex_a8" 
>                  | CortexA9 -> "cortex_a9"
> 
> as:
> 
> let coreStr = function
>     CortexA8 -> "cortex_a8"
>   | CortexA9 -> "cortex_a9"

Can do . I haven't written any ML in a long time.



> 
> I know there's no GNU style guide for ML, but please try to follow the
> existing formatting (re: indentation and so forth) if possible.

Thanks for the review. I thought I had done that but will go back and
double check and follow up with a patch. 

Cheers
Ramana

> 
> Cheers,
> 
> Julian
Julian Brown Aug. 25, 2010, 1:53 p.m. UTC | #3 
On Wed, 25 Aug 2010 14:07:01 +0100
Ramana Radhakrishnan <ramana.radhakrishnan@arm.com> wrote:

> On Wed, 2010-08-25 at 12:04 +0100, Julian Brown wrote:
> > [snip]
> > I might have the wrong end of the stick here but... neon.md is
> > filled with "neon_type" specifications which are already hairy and
> > verbose,
> 
> No that is not what I'm doing. 

> Errr.. IIUC neon-schedgen.ml is responsible only for generating the
> scheduler descriptions that use *neon_types* . All that this does is
> prefix the define_insn_reservation names. You need a per-core
> insn_reservation to be able to describe how it interacts with the rest
> of the core and the latency might well be different in different
> cores. 
> 
> What I have done is pretty much the following and a rather straight
> forward renaming exercise.
> 
> < a small hunk from the new A8 neon scheduler description is as
> below> 
> 
> -(define_insn_reservation "neon_int_1" 3
> +(define_insn_reservation "cortex_a8_neon_int_1" 3
>    (and (eq_attr "tune" "cortexa8")
>         (eq_attr "neon_type" "neon_int_1"))
>    "cortex_a8_neon_dp")

Oh, duh, I see, sorry. The similarity of the reservation names and the
neon_types threw me. My point still stands that the neon_types are too
physically-based, I think, but maybe that's a problem for another time.

> > If you're going to hack neon-schedgen.ml (which I didn't write, and
> > admit I don't have any great understanding of), would it make sense
> > to create a new table for Cortex-A9 (etc.), rather than mixing
> > definitions for all cores together in the existing
> > availability_table?
> 
> There are situations where we might want to use common aspects of the
> engine in the table and thus a filter function appeared to be good
> enough on the table rather than splitting this table into a "common
> across all cores" bit, and a per core bit.

OK. I'm not sure how similar the different NEON implementations are,
but I guess you have better information than me :-).

> > I know there's no GNU style guide for ML, but please try to follow
> > the existing formatting (re: indentation and so forth) if possible.
> 
> Thanks for the review. I thought I had done that but will go back and
> double check and follow up with a patch. 

I think it looks generally OK apart from the indendation on the
functions coreStr and tuneStr, looking again.

Cheers,

Julian
Ramana Radhakrishnan Aug. 25, 2010, 2:59 p.m. UTC | #4 
> Oh, duh, I see, sorry. The similarity of the reservation names and the
> neon_types threw me. My point still stands that the neon_types are too
> physically-based, I think, but maybe that's a problem for another time.

We need the distinction to model the instructions right on cores where
that distinction is necessary (e.g. A8 since that's the first one out
there), but if we don't do it with so many neon_types, we need to figure
out other abstractions that allow us to implement the same thing with
appropriate combinations.

I haven't looked at all the values of the neon_type attribute in great
detail but a possible simplication off the top of my head might be to
have an extra attribute "sets_q_reg" and then added that to the
condition in the generated reservation tables. I suspect this is what
you are alluding to here. 

You would still have the entries in the availability_table for both
neonfp_vadd_vabs_ddd and neonfp_vadd_vabs_qqq because you want different
behaviours for both. 

(define_insn_reservation "<corename>_neonfp_vadd_vabs_ddd" <latency1>
   ( and (eq_attr "neon_type" "neon_fp_vadd_vabs")
	 (eq_attr "sets_q_reg" "false"))

(define_insn_reservation "<corename>_neonfp_vadd_vabs_qqq" <latency2>
   ( and (eq_attr "neon_type" "neon_fp_vadd_vabs")
	 (eq_attr "sets_q_reg" "true"))


And then the horrible looking condition in neon.md can get much simpler
with just one test for Is_d_reg and one check for setting whether this
is an FP or int operation of the appropriate type *. 


cheers
Ramana


* Ofcourse this doesn't preclude the possibility of extending that
further to floating point values but I'll leave it there because
otherwise we then definitely need a consistency checker to make sure we
aren't accidentally using an integer resource for a floating point
resource, which could cause some more fake hazards which never exist :).
diff mbox

Patch

Index: gcc/config/arm/neon-schedgen.ml
===================================================================
--- gcc/config/arm/neon-schedgen.ml	(revision 163537)
+++ gcc/config/arm/neon-schedgen.ml	(working copy)
@@ -1,7 +1,6 @@ 
 (* Emission of the core of the Cortex-A8 NEON scheduling description.
    Copyright (C) 2007, 2010 Free Software Foundation, Inc.
    Contributed by CodeSourcery.
-
    This file is part of GCC.
 
    GCC is free software; you can redistribute it and/or modify it under
@@ -21,7 +20,14 @@ 
 
 (* This scheduling description generator works as follows.
    - Each group of instructions has source and destination requirements
-     specified.  The source requirements may be specified using
+     specified and a list of cores supported. This is then filtered
+     and per core scheduler descriptions are generated out.
+     The reservations generated are prefixed by the name of the
+     core and the check is performed on the basis of what the tuning
+     string is. Running this will generate Neon scheduler descriptions
+     for all cores supported.
+
+     The source requirements may be specified using
      Source (the stage at which all source operands not otherwise
      described are read), Source_m (the stage at which Rm operands are
      read), Source_n (likewise for Rn) and Source_d (likewise for Rd).
@@ -83,6 +89,18 @@  type reservation =
 | Ls of int
 | Fmul_then_fadd | Fmul_then_fadd_2
 
+type core = CortexA8 | CortexA9
+let allCores = [CortexA8]
+let coreStr core = 
+                 match core with
+                    CortexA8 -> "cortex_a8" 
+                   | CortexA9 -> "cortex_a9"
+let tuneStr core = 
+                 match core with 
+                    CortexA8 -> "cortexa8"
+                   | CortexA9 -> "cortexa9"
+
+
 (* This table must be kept as short as possible by conflating
    entries with the same availability behavior.
 
@@ -90,129 +108,136 @@  type reservation =
    Second components: availability requirements, in the order in which
    they should appear in the comments in the .md file.
    Third components: reservation info
+   Fourth components: List of supported cores.
 *)
 let availability_table = [
   (* NEON integer ALU instructions.  *)
   (* vbit vbif vbsl vorr vbic vnot vcls vclz vcnt vadd vand vorr
      veor vbic vorn ddd qqq *)
-  "neon_int_1", [Source n2; Dest n3], ALU;
+  "neon_int_1", [Source n2; Dest n3], ALU, allCores;
   (* vadd vsub qqd vsub ddd qqq *)
-  "neon_int_2", [Source_m n1; Source_n n2; Dest n3], ALU;
+  "neon_int_2", [Source_m n1; Source_n n2; Dest n3], ALU, allCores;
   (* vsum vneg dd qq vadd vsub qdd *)
-  "neon_int_3", [Source n1; Dest n3], ALU;
+  "neon_int_3", [Source n1; Dest n3], ALU, allCores;
   (* vabs vceqz vcgez vcbtz vclez vcltz vadh vradh vsbh vrsbh dqq *)
   (* vhadd vrhadd vqadd vtst ddd qqq *)
-  "neon_int_4", [Source n2; Dest n4], ALU;
+  "neon_int_4", [Source n2; Dest n4], ALU, allCores;
   (* vabd qdd vhsub vqsub vabd vceq vcge vcgt vmax vmin vfmx vfmn ddd ddd *)
-  "neon_int_5", [Source_m n1; Source_n n2; Dest n4], ALU;
+  "neon_int_5", [Source_m n1; Source_n n2; Dest n4], ALU, allCores;
   (* vqneg vqabs dd qq *)
-  "neon_vqneg_vqabs", [Source n1; Dest n4], ALU;
+  "neon_vqneg_vqabs", [Source n1; Dest n4], ALU, allCores;
   (* vmov vmvn *)
-  "neon_vmov", [Dest n3], ALU;
+  "neon_vmov", [Dest n3], ALU, allCores;
   (* vaba *)
-  "neon_vaba", [Source_n n2; Source_m n1; Source_d n3; Dest n6], ALU;
+  "neon_vaba", [Source_n n2; Source_m n1; Source_d n3; Dest n6], ALU, allCores;
   "neon_vaba_qqq",
-    [Source_n n2; Source_m n1; Source_d n3; Dest_n_after (1, n6)], ALU_2cycle;
+    [Source_n n2; Source_m n1; Source_d n3; Dest_n_after (1, n6)], 
+   ALU_2cycle, allCores;
   (* vsma *)
-  "neon_vsma", [Source_m n1; Source_d n3; Dest n6], ALU;
+  "neon_vsma", [Source_m n1; Source_d n3; Dest n6], ALU, allCores;
 
   (* NEON integer multiply instructions.  *)
   (* vmul, vqdmlh, vqrdmlh *)
   (* vmul, vqdmul, qdd 16/8 long 32/16 long *)
-  "neon_mul_ddd_8_16_qdd_16_8_long_32_16_long", [Source n2; Dest n6], Mul;
-  "neon_mul_qqq_8_16_32_ddd_32", [Source n2; Dest_n_after (1, n6)], Mul_2cycle;
+  "neon_mul_ddd_8_16_qdd_16_8_long_32_16_long", [Source n2; Dest n6], 
+   Mul, allCores;
+  "neon_mul_qqq_8_16_32_ddd_32", [Source n2; Dest_n_after (1, n6)], 
+   Mul_2cycle, allCores;
   (* vmul, vqdmul again *)
   "neon_mul_qdd_64_32_long_qqd_16_ddd_32_scalar_64_32_long_scalar",
-    [Source_n n2; Source_m n1; Dest_n_after (1, n6)], Mul_2cycle;
+    [Source_n n2; Source_m n1; Dest_n_after (1, n6)], Mul_2cycle, allCores;
   (* vmla, vmls *)
   "neon_mla_ddd_8_16_qdd_16_8_long_32_16_long",
-    [Source_n n2; Source_m n2; Source_d n3; Dest n6], Mul;
+    [Source_n n2; Source_m n2; Source_d n3; Dest n6], Mul, allCores;
   "neon_mla_qqq_8_16",
-    [Source_n n2; Source_m n2; Source_d n3; Dest_n_after (1, n6)], Mul_2cycle;
+    [Source_n n2; Source_m n2; Source_d n3; Dest_n_after (1, n6)], 
+   Mul_2cycle, allCores;
   "neon_mla_ddd_32_qqd_16_ddd_32_scalar_qdd_64_32_long_scalar_qdd_64_32_long",
-    [Source_n n2; Source_m n1; Source_d n3; Dest_n_after (1, n6)], Mul_2cycle;
+    [Source_n n2; Source_m n1; Source_d n3; Dest_n_after (1, n6)], 
+   Mul_2cycle, allCores;
   "neon_mla_qqq_32_qqd_32_scalar",
-    [Source_n n2; Source_m n1; Source_d n3; Dest_n_after (3, n6)], Mul_4cycle;
+    [Source_n n2; Source_m n1; Source_d n3; Dest_n_after (3, n6)], 
+   Mul_4cycle, allCores;
   (* vmul, vqdmulh, vqrdmulh *)
   (* vmul, vqdmul *)
   "neon_mul_ddd_16_scalar_32_16_long_scalar",
-    [Source_n n2; Source_m n1; Dest n6], Mul;
+    [Source_n n2; Source_m n1; Dest n6], Mul, allCores;
   "neon_mul_qqd_32_scalar",
-    [Source_n n2; Source_m n1; Dest_n_after (3, n6)], Mul_4cycle;
+    [Source_n n2; Source_m n1; Dest_n_after (3, n6)], Mul_4cycle, allCores;
   (* vmla, vmls *)
   (* vmla, vmla, vqdmla, vqdmls *)
   "neon_mla_ddd_16_scalar_qdd_32_16_long_scalar",
-    [Source_n n2; Source_m n1; Source_d n3; Dest n6], Mul;
+    [Source_n n2; Source_m n1; Source_d n3; Dest n6], Mul, allCores;
 
   (* NEON integer shift instructions.  *)
   (* vshr/vshl immediate, vshr_narrow, vshl_vmvh, vsli_vsri_ddd *)
-  "neon_shift_1", [Source n1; Dest n3], Shift;
-  (* vqshl, vrshr immediate; vqshr, vqmov, vrshr, vqrshr narrow;
+  "neon_shift_1", [Source n1; Dest n3], Shift, allCores;
+  (* vqshl, vrshr immediate; vqshr, vqmov, vrshr, vqrshr narrow, allCores;
      vqshl_vrshl_vqrshl_ddd *)
-  "neon_shift_2", [Source n1; Dest n4], Shift;
+  "neon_shift_2", [Source n1; Dest n4], Shift, allCores;
   (* vsli, vsri and vshl for qqq *)
-  "neon_shift_3", [Source n1; Dest_n_after (1, n3)], Shift_2cycle;
-  "neon_vshl_ddd", [Source n1; Dest n1], Shift;
+  "neon_shift_3", [Source n1; Dest_n_after (1, n3)], Shift_2cycle, allCores;
+  "neon_vshl_ddd", [Source n1; Dest n1], Shift, allCores;
   "neon_vqshl_vrshl_vqrshl_qqq", [Source n1; Dest_n_after (1, n4)],
-    Shift_2cycle;
-  "neon_vsra_vrsra", [Source_m n1; Source_d n3; Dest n6], Shift;
+    Shift_2cycle, allCores;
+  "neon_vsra_vrsra", [Source_m n1; Source_d n3; Dest n6], Shift, allCores;
 
   (* NEON floating-point instructions.  *)
   (* vadd, vsub, vabd, vmul, vceq, vcge, vcgt, vcage, vcagt, vmax, vmin *)
   (* vabs, vneg, vceqz, vcgez, vcgtz, vclez, vcltz, vrecpe, vrsqrte, vcvt *)
-  "neon_fp_vadd_ddd_vabs_dd", [Source n2; Dest n5], Fadd;
+  "neon_fp_vadd_ddd_vabs_dd", [Source n2; Dest n5], Fadd, allCores;
   "neon_fp_vadd_qqq_vabs_qq", [Source n2; Dest_n_after (1, n5)],
-    Fadd_2cycle;
+    Fadd_2cycle, allCores;
   (* vsum, fvmx, vfmn *)
-  "neon_fp_vsum", [Source n1; Dest n5], Fadd;
-  "neon_fp_vmul_ddd", [Source_n n2; Source_m n1; Dest n5], Fmul;
+  "neon_fp_vsum", [Source n1; Dest n5], Fadd, allCores;
+  "neon_fp_vmul_ddd", [Source_n n2; Source_m n1; Dest n5], Fmul, allCores;
   "neon_fp_vmul_qqd", [Source_n n2; Source_m n1; Dest_n_after (1, n5)],
-    Fmul_2cycle;
+    Fmul_2cycle, allCores;
   (* vmla, vmls *)
   "neon_fp_vmla_ddd",
-    [Source_n n2; Source_m n2; Source_d n3; Dest n9], Fmul_then_fadd;
+    [Source_n n2; Source_m n2; Source_d n3; Dest n9], Fmul_then_fadd, allCores;
   "neon_fp_vmla_qqq",
     [Source_n n2; Source_m n2; Source_d n3; Dest_n_after (1, n9)],
-    Fmul_then_fadd_2;
+    Fmul_then_fadd_2, allCores;
   "neon_fp_vmla_ddd_scalar",
-    [Source_n n2; Source_m n1; Source_d n3; Dest n9], Fmul_then_fadd;
+    [Source_n n2; Source_m n1; Source_d n3; Dest n9], Fmul_then_fadd, allCores;
   "neon_fp_vmla_qqq_scalar",
     [Source_n n2; Source_m n1; Source_d n3; Dest_n_after (1, n9)],
-    Fmul_then_fadd_2;
-  "neon_fp_vrecps_vrsqrts_ddd", [Source n2; Dest n9], Fmul_then_fadd;
+    Fmul_then_fadd_2, allCores;
+  "neon_fp_vrecps_vrsqrts_ddd", [Source n2; Dest n9], Fmul_then_fadd, allCores;
   "neon_fp_vrecps_vrsqrts_qqq", [Source n2; Dest_n_after (1, n9)],
-    Fmul_then_fadd_2;
+    Fmul_then_fadd_2, allCores;
 
   (* NEON byte permute instructions.  *)
   (* vmov; vtrn and vswp for dd; vzip for dd; vuzp for dd; vrev; vext for dd *)
-  "neon_bp_simple", [Source n1; Dest n2], Permute 1;
-  (* vswp for qq; vext for qqq; vtbl with {Dn} or {Dn, Dn1};
+  "neon_bp_simple", [Source n1; Dest n2], Permute 1, allCores;
+  (* vswp for qq; vext for qqq; vtbl with {Dn} or {Dn, Dn1}, allCores;
      similarly for vtbx *)
-  "neon_bp_2cycle", [Source n1; Dest_n_after (1, n2)], Permute 2;
+  "neon_bp_2cycle", [Source n1; Dest_n_after (1, n2)], Permute 2, allCores;
   (* all the rest *)
-  "neon_bp_3cycle", [Source n1; Dest_n_after (2, n2)], Permute 3;
+  "neon_bp_3cycle", [Source n1; Dest_n_after (2, n2)], Permute 3, allCores;
 
   (* NEON load/store instructions.  *)
-  "neon_ldr", [Dest n1], Ls 1;
-  "neon_str", [Source n1], Ls 1;
-  "neon_vld1_1_2_regs", [Dest_n_after (1, n1)], Ls 2;
-  "neon_vld1_3_4_regs", [Dest_n_after (2, n1)], Ls 3;
-  "neon_vld2_2_regs_vld1_vld2_all_lanes", [Dest_n_after (1, n2)], Ls 2;
-  "neon_vld2_4_regs", [Dest_n_after (2, n2)], Ls 3;
-  "neon_vld3_vld4", [Dest_n_after (3, n2)], Ls 4;
-  "neon_vst1_1_2_regs_vst2_2_regs", [Source n1], Ls 2;
-  "neon_vst1_3_4_regs", [Source n1], Ls 3;
-  "neon_vst2_4_regs_vst3_vst4", [Source n1], Ls 4;
-  "neon_vst3_vst4", [Source n1], Ls 4;
-  "neon_vld1_vld2_lane", [Source n1; Dest_n_after (2, n2)], Ls 3;
-  "neon_vld3_vld4_lane", [Source n1; Dest_n_after (4, n2)], Ls 5;
-  "neon_vst1_vst2_lane", [Source n1], Ls 2;
-  "neon_vst3_vst4_lane", [Source n1], Ls 3;
-  "neon_vld3_vld4_all_lanes", [Dest_n_after (1, n2)], Ls 3;
+  "neon_ldr", [Dest n1], Ls 1, allCores;
+  "neon_str", [Source n1], Ls 1, allCores;
+  "neon_vld1_1_2_regs", [Dest_n_after (1, n1)], Ls 2, allCores;
+  "neon_vld1_3_4_regs", [Dest_n_after (2, n1)], Ls 3, allCores;
+  "neon_vld2_2_regs_vld1_vld2_all_lanes", [Dest_n_after (1, n2)], Ls 2, allCores;
+  "neon_vld2_4_regs", [Dest_n_after (2, n2)], Ls 3, allCores;
+  "neon_vld3_vld4", [Dest_n_after (3, n2)], Ls 4, allCores;
+  "neon_vst1_1_2_regs_vst2_2_regs", [Source n1], Ls 2, allCores;
+  "neon_vst1_3_4_regs", [Source n1], Ls 3, allCores;
+  "neon_vst2_4_regs_vst3_vst4", [Source n1], Ls 4, allCores;
+  "neon_vst3_vst4", [Source n1], Ls 4, allCores;
+  "neon_vld1_vld2_lane", [Source n1; Dest_n_after (2, n2)], Ls 3, allCores;
+  "neon_vld3_vld4_lane", [Source n1; Dest_n_after (4, n2)], Ls 5, allCores;
+  "neon_vst1_vst2_lane", [Source n1], Ls 2, allCores;
+  "neon_vst3_vst4_lane", [Source n1], Ls 3, allCores;
+  "neon_vld3_vld4_all_lanes", [Dest_n_after (1, n2)], Ls 3, allCores;
 
   (* NEON register transfer instructions.  *)
-  "neon_mcr", [Dest n2], Permute 1;
-  "neon_mcr_2_mcrr", [Dest n2], Permute 2;
+  "neon_mcr", [Dest n2], Permute 1, allCores;
+  "neon_mcr_2_mcrr", [Dest n2], Permute 2, allCores;
   (* MRC instructions are in the .tpl file.  *)
 ]
 
@@ -221,7 +246,7 @@  let availability_table = [
    required.  (It is also possible that an entry in the table has no
    source requirements.)  *)
 let calculate_sources =
-  List.map (fun (name, avail, res) ->
+  List.map (fun (name, avail, res, cores) ->
               let earliest_stage =
                 List.fold_left
                   (fun cur -> fun info ->
@@ -331,7 +356,7 @@  let pick_latency largest worst guards =
    of one bypass from this producer to any particular consumer listed
    in LATENCIES.)  Use a hash table to collate bypasses with the
    same latency and guard.  *)
-let collate_bypasses (producer_name, _, _, _) largest latencies =
+let collate_bypasses (producer_name, _, _, _) largest latencies core =
   let ht = Hashtbl.create 42 in
   let keys = ref [] in
     List.iter (
@@ -350,7 +375,7 @@  let collate_bypasses (producer_name, _, 
               (if (try ignore (Hashtbl.find ht (guard, latency)); false
                    with Not_found -> true) then
                  keys := (guard, latency) :: !keys);
-              Hashtbl.add ht (guard, latency) consumer
+              Hashtbl.add ht (guard, latency) ((coreStr core) ^ "_" ^ consumer)
             end
     ) latencies;
     (* The hash table now has bypasses collated so that ones with the
@@ -372,7 +397,7 @@  let collate_bypasses (producer_name, _, 
    the output in such a way that all bypasses with the same producer
    and latency are together, and so that bypasses with the worst-case
    latency are ignored.  *)
-let worst_case_latencies_and_bypasses =
+let worst_case_latencies_and_bypasses core =
   let rec f (worst_acc, bypasses_acc) prev xs =
     match xs with
       [] -> (worst_acc, bypasses_acc)
@@ -400,7 +425,7 @@  let worst_case_latencies_and_bypasses =
           (* Having got the largest latency, collect all bypasses for
              this producer and filter out those with that larger
              latency.  Record the others for later emission.  *)
-          let bypasses = collate_bypasses producer largest latencies in
+          let bypasses = collate_bypasses producer largest latencies core in
             (* Go on to process remaining producers, having noted
                the result for this one.  *)
             f ((producer_name, producer_avail, largest,
@@ -444,14 +469,18 @@  let write_comment producer avail =
     in
       f avail 0
 
+
 (* Emit a define_insn_reservation for each producer.  The latency
    written in will be its worst-case latency.  *)
-let emit_insn_reservations =
-  List.iter (
+let emit_insn_reservations core =
+  let corestring = coreStr core in
+  let tunestring = tuneStr core
+  in  List.iter (
      fun (producer, avail, latency, reservation) ->
         write_comment producer avail;
-        Printf.printf "(define_insn_reservation \"%s\" %d\n" producer latency;
-        Printf.printf "  (and (eq_attr \"tune\" \"cortexa8\")\n";
+        Printf.printf "(define_insn_reservation \"%s_%s\" %d\n" 
+            corestring producer latency;
+            Printf.printf "  (and (eq_attr \"tune\" \"%s\")\n" tunestring;
         Printf.printf "       (eq_attr \"neon_type\" \"%s\"))\n" producer;
         let str =
           match reservation with
@@ -467,7 +496,7 @@  let emit_insn_reservations =
 	  | Fmul_then_fadd -> "fmul_then_fadd"
 	  | Fmul_then_fadd_2 -> "fmul_then_fadd_2"
         in
-          Printf.printf "  \"cortex_a8_neon_%s\")\n\n" str
+          Printf.printf "  \"%s_neon_%s\")\n\n" corestring str
     )
 
 (* Given a guard description, return the name of the C function to
@@ -480,10 +509,12 @@  let guard_fn g =
   | Guard_none -> assert false
 
 (* Emit a define_bypass for each bypass.  *)
-let emit_bypasses =
+let emit_bypasses core =
   List.iter (
       fun (producer, consumers, latency, guard) ->
-        Printf.printf "(define_bypass %d \"%s\"\n" latency producer;
+        Printf.printf "(define_bypass %d \"%s_%s\"\n" 
+	latency (coreStr core) producer;
+
         if guard = Guard_none then
           Printf.printf "               \"%s\")\n\n" consumers
         else
@@ -493,11 +524,21 @@  let emit_bypasses =
           end
     )
 
-(* Program entry point.  *)
-let main =
+
+let calculate_per_core_availability_table core availability_table =
   let table = calculate_sources availability_table in
-  let worst_cases, bypasses = worst_case_latencies_and_bypasses table in
-    emit_insn_reservations (List.rev worst_cases);
+  let worst_cases, bypasses = worst_case_latencies_and_bypasses core table in
+    emit_insn_reservations core (List.rev worst_cases);
     Printf.printf ";; Exceptions to the default latencies.\n\n";
-    emit_bypasses bypasses
+    emit_bypasses core bypasses
+
+let calculate_core_availability_table core availability_table =
+let filter_core = List.filter (fun (_, _, _, cores) 
+				   -> List.exists ((==) core) cores)
+in calculate_per_core_availability_table core (filter_core availability_table)
 
+
+(* Program entry point.  *)
+let main =
+  List.map (fun core -> calculate_core_availability_table 
+		core availability_table) allCores
Index: gcc/config/arm/cortex-a8-neon.md
===================================================================
--- gcc/config/arm/cortex-a8-neon.md	(revision 163537)
+++ gcc/config/arm/cortex-a8-neon.md	(working copy)
@@ -1,5 +1,5 @@ 
 ;; ARM Cortex-A8 NEON scheduling description.
-;; Copyright (C) 2007, 2008 Free Software Foundation, Inc.
+;; Copyright (C) 2007, 2008, 2010 Free Software Foundation, Inc.
 ;; Contributed by CodeSourcery.
 
 ;; This file is part of GCC.
@@ -182,12 +182,12 @@ 
 
 ;; NEON -> core transfers.
 
-(define_insn_reservation "neon_mrc" 20
+(define_insn_reservation "cortex_a8_neon_mrc" 20
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mrc"))
   "cortex_a8_neon_ls")
 
-(define_insn_reservation "neon_mrrc" 21
+(define_insn_reservation "cortex_a8_neon_mrrc" 21
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mrrc"))
   "cortex_a8_neon_ls_2")
@@ -196,48 +196,48 @@ 
 
 ;; Instructions using this reservation read their source operands at N2, and
 ;; produce a result at N3.
-(define_insn_reservation "neon_int_1" 3
+(define_insn_reservation "cortex_a8_neon_int_1" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_int_1"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their (D|Q)m operands at N1,
 ;; their (D|Q)n operands at N2, and produce a result at N3.
-(define_insn_reservation "neon_int_2" 3
+(define_insn_reservation "cortex_a8_neon_int_2" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_int_2"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N3.
-(define_insn_reservation "neon_int_3" 3
+(define_insn_reservation "cortex_a8_neon_int_3" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_int_3"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N2, and
 ;; produce a result at N4.
-(define_insn_reservation "neon_int_4" 4
+(define_insn_reservation "cortex_a8_neon_int_4" 4
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_int_4"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their (D|Q)m operands at N1,
 ;; their (D|Q)n operands at N2, and produce a result at N4.
-(define_insn_reservation "neon_int_5" 4
+(define_insn_reservation "cortex_a8_neon_int_5" 4
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_int_5"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N4.
-(define_insn_reservation "neon_vqneg_vqabs" 4
+(define_insn_reservation "cortex_a8_neon_vqneg_vqabs" 4
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vqneg_vqabs"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation produce a result at N3.
-(define_insn_reservation "neon_vmov" 3
+(define_insn_reservation "cortex_a8_neon_vmov" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vmov"))
   "cortex_a8_neon_dp")
@@ -245,7 +245,7 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, their (D|Q)d operands at N3, and
 ;; produce a result at N6.
-(define_insn_reservation "neon_vaba" 6
+(define_insn_reservation "cortex_a8_neon_vaba" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vaba"))
   "cortex_a8_neon_dp")
@@ -253,35 +253,35 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, their (D|Q)d operands at N3, and
 ;; produce a result at N6 on cycle 2.
-(define_insn_reservation "neon_vaba_qqq" 7
+(define_insn_reservation "cortex_a8_neon_vaba_qqq" 7
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vaba_qqq"))
   "cortex_a8_neon_dp_2")
 
 ;; Instructions using this reservation read their (D|Q)m operands at N1,
 ;; their (D|Q)d operands at N3, and produce a result at N6.
-(define_insn_reservation "neon_vsma" 6
+(define_insn_reservation "cortex_a8_neon_vsma" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vsma"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N2, and
 ;; produce a result at N6.
-(define_insn_reservation "neon_mul_ddd_8_16_qdd_16_8_long_32_16_long" 6
+(define_insn_reservation "cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mul_ddd_8_16_qdd_16_8_long_32_16_long"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N2, and
 ;; produce a result at N6 on cycle 2.
-(define_insn_reservation "neon_mul_qqq_8_16_32_ddd_32" 7
+(define_insn_reservation "cortex_a8_neon_mul_qqq_8_16_32_ddd_32" 7
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mul_qqq_8_16_32_ddd_32"))
   "cortex_a8_neon_dp_2")
 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, and produce a result at N6 on cycle 2.
-(define_insn_reservation "neon_mul_qdd_64_32_long_qqd_16_ddd_32_scalar_64_32_long_scalar" 7
+(define_insn_reservation "cortex_a8_neon_mul_qdd_64_32_long_qqd_16_ddd_32_scalar_64_32_long_scalar" 7
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mul_qdd_64_32_long_qqd_16_ddd_32_scalar_64_32_long_scalar"))
   "cortex_a8_neon_dp_2")
@@ -289,7 +289,7 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N2, their (D|Q)d operands at N3, and
 ;; produce a result at N6.
-(define_insn_reservation "neon_mla_ddd_8_16_qdd_16_8_long_32_16_long" 6
+(define_insn_reservation "cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mla_ddd_8_16_qdd_16_8_long_32_16_long"))
   "cortex_a8_neon_dp")
@@ -297,7 +297,7 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N2, their (D|Q)d operands at N3, and
 ;; produce a result at N6 on cycle 2.
-(define_insn_reservation "neon_mla_qqq_8_16" 7
+(define_insn_reservation "cortex_a8_neon_mla_qqq_8_16" 7
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mla_qqq_8_16"))
   "cortex_a8_neon_dp_2")
@@ -305,7 +305,7 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, their (D|Q)d operands at N3, and
 ;; produce a result at N6 on cycle 2.
-(define_insn_reservation "neon_mla_ddd_32_qqd_16_ddd_32_scalar_qdd_64_32_long_scalar_qdd_64_32_long" 7
+(define_insn_reservation "cortex_a8_neon_mla_ddd_32_qqd_16_ddd_32_scalar_qdd_64_32_long_scalar_qdd_64_32_long" 7
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mla_ddd_32_qqd_16_ddd_32_scalar_qdd_64_32_long_scalar_qdd_64_32_long"))
   "cortex_a8_neon_dp_2")
@@ -313,21 +313,21 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, their (D|Q)d operands at N3, and
 ;; produce a result at N6 on cycle 4.
-(define_insn_reservation "neon_mla_qqq_32_qqd_32_scalar" 9
+(define_insn_reservation "cortex_a8_neon_mla_qqq_32_qqd_32_scalar" 9
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mla_qqq_32_qqd_32_scalar"))
   "cortex_a8_neon_dp_4")
 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, and produce a result at N6.
-(define_insn_reservation "neon_mul_ddd_16_scalar_32_16_long_scalar" 6
+(define_insn_reservation "cortex_a8_neon_mul_ddd_16_scalar_32_16_long_scalar" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mul_ddd_16_scalar_32_16_long_scalar"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, and produce a result at N6 on cycle 4.
-(define_insn_reservation "neon_mul_qqd_32_scalar" 9
+(define_insn_reservation "cortex_a8_neon_mul_qqd_32_scalar" 9
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mul_qqd_32_scalar"))
   "cortex_a8_neon_dp_4")
@@ -335,84 +335,84 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, their (D|Q)d operands at N3, and
 ;; produce a result at N6.
-(define_insn_reservation "neon_mla_ddd_16_scalar_qdd_32_16_long_scalar" 6
+(define_insn_reservation "cortex_a8_neon_mla_ddd_16_scalar_qdd_32_16_long_scalar" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mla_ddd_16_scalar_qdd_32_16_long_scalar"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N3.
-(define_insn_reservation "neon_shift_1" 3
+(define_insn_reservation "cortex_a8_neon_shift_1" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_shift_1"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N4.
-(define_insn_reservation "neon_shift_2" 4
+(define_insn_reservation "cortex_a8_neon_shift_2" 4
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_shift_2"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N3 on cycle 2.
-(define_insn_reservation "neon_shift_3" 4
+(define_insn_reservation "cortex_a8_neon_shift_3" 4
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_shift_3"))
   "cortex_a8_neon_dp_2")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N1.
-(define_insn_reservation "neon_vshl_ddd" 1
+(define_insn_reservation "cortex_a8_neon_vshl_ddd" 1
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vshl_ddd"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N4 on cycle 2.
-(define_insn_reservation "neon_vqshl_vrshl_vqrshl_qqq" 5
+(define_insn_reservation "cortex_a8_neon_vqshl_vrshl_vqrshl_qqq" 5
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vqshl_vrshl_vqrshl_qqq"))
   "cortex_a8_neon_dp_2")
 
 ;; Instructions using this reservation read their (D|Q)m operands at N1,
 ;; their (D|Q)d operands at N3, and produce a result at N6.
-(define_insn_reservation "neon_vsra_vrsra" 6
+(define_insn_reservation "cortex_a8_neon_vsra_vrsra" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vsra_vrsra"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their source operands at N2, and
 ;; produce a result at N5.
-(define_insn_reservation "neon_fp_vadd_ddd_vabs_dd" 5
+(define_insn_reservation "cortex_a8_neon_fp_vadd_ddd_vabs_dd" 5
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vadd_ddd_vabs_dd"))
   "cortex_a8_neon_fadd")
 
 ;; Instructions using this reservation read their source operands at N2, and
 ;; produce a result at N5 on cycle 2.
-(define_insn_reservation "neon_fp_vadd_qqq_vabs_qq" 6
+(define_insn_reservation "cortex_a8_neon_fp_vadd_qqq_vabs_qq" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vadd_qqq_vabs_qq"))
   "cortex_a8_neon_fadd_2")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N5.
-(define_insn_reservation "neon_fp_vsum" 5
+(define_insn_reservation "cortex_a8_neon_fp_vsum" 5
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vsum"))
   "cortex_a8_neon_fadd")
 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, and produce a result at N5.
-(define_insn_reservation "neon_fp_vmul_ddd" 5
+(define_insn_reservation "cortex_a8_neon_fp_vmul_ddd" 5
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vmul_ddd"))
   "cortex_a8_neon_dp")
 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, and produce a result at N5 on cycle 2.
-(define_insn_reservation "neon_fp_vmul_qqd" 6
+(define_insn_reservation "cortex_a8_neon_fp_vmul_qqd" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vmul_qqd"))
   "cortex_a8_neon_dp_2")
@@ -420,7 +420,7 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N2, their (D|Q)d operands at N3, and
 ;; produce a result at N9.
-(define_insn_reservation "neon_fp_vmla_ddd" 9
+(define_insn_reservation "cortex_a8_neon_fp_vmla_ddd" 9
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vmla_ddd"))
   "cortex_a8_neon_fmul_then_fadd")
@@ -428,7 +428,7 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N2, their (D|Q)d operands at N3, and
 ;; produce a result at N9 on cycle 2.
-(define_insn_reservation "neon_fp_vmla_qqq" 10
+(define_insn_reservation "cortex_a8_neon_fp_vmla_qqq" 10
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vmla_qqq"))
   "cortex_a8_neon_fmul_then_fadd_2")
@@ -436,7 +436,7 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, their (D|Q)d operands at N3, and
 ;; produce a result at N9.
-(define_insn_reservation "neon_fp_vmla_ddd_scalar" 9
+(define_insn_reservation "cortex_a8_neon_fp_vmla_ddd_scalar" 9
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vmla_ddd_scalar"))
   "cortex_a8_neon_fmul_then_fadd")
@@ -444,869 +444,869 @@ 
 ;; Instructions using this reservation read their (D|Q)n operands at N2,
 ;; their (D|Q)m operands at N1, their (D|Q)d operands at N3, and
 ;; produce a result at N9 on cycle 2.
-(define_insn_reservation "neon_fp_vmla_qqq_scalar" 10
+(define_insn_reservation "cortex_a8_neon_fp_vmla_qqq_scalar" 10
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vmla_qqq_scalar"))
   "cortex_a8_neon_fmul_then_fadd_2")
 
 ;; Instructions using this reservation read their source operands at N2, and
 ;; produce a result at N9.
-(define_insn_reservation "neon_fp_vrecps_vrsqrts_ddd" 9
+(define_insn_reservation "cortex_a8_neon_fp_vrecps_vrsqrts_ddd" 9
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vrecps_vrsqrts_ddd"))
   "cortex_a8_neon_fmul_then_fadd")
 
 ;; Instructions using this reservation read their source operands at N2, and
 ;; produce a result at N9 on cycle 2.
-(define_insn_reservation "neon_fp_vrecps_vrsqrts_qqq" 10
+(define_insn_reservation "cortex_a8_neon_fp_vrecps_vrsqrts_qqq" 10
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_fp_vrecps_vrsqrts_qqq"))
   "cortex_a8_neon_fmul_then_fadd_2")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N2.
-(define_insn_reservation "neon_bp_simple" 2
+(define_insn_reservation "cortex_a8_neon_bp_simple" 2
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_bp_simple"))
   "cortex_a8_neon_perm")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N2 on cycle 2.
-(define_insn_reservation "neon_bp_2cycle" 3
+(define_insn_reservation "cortex_a8_neon_bp_2cycle" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_bp_2cycle"))
   "cortex_a8_neon_perm_2")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N2 on cycle 3.
-(define_insn_reservation "neon_bp_3cycle" 4
+(define_insn_reservation "cortex_a8_neon_bp_3cycle" 4
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_bp_3cycle"))
   "cortex_a8_neon_perm_3")
 
 ;; Instructions using this reservation produce a result at N1.
-(define_insn_reservation "neon_ldr" 1
+(define_insn_reservation "cortex_a8_neon_ldr" 1
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_ldr"))
   "cortex_a8_neon_ls")
 
 ;; Instructions using this reservation read their source operands at N1.
-(define_insn_reservation "neon_str" 0
+(define_insn_reservation "cortex_a8_neon_str" 0
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_str"))
   "cortex_a8_neon_ls")
 
 ;; Instructions using this reservation produce a result at N1 on cycle 2.
-(define_insn_reservation "neon_vld1_1_2_regs" 2
+(define_insn_reservation "cortex_a8_neon_vld1_1_2_regs" 2
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vld1_1_2_regs"))
   "cortex_a8_neon_ls_2")
 
 ;; Instructions using this reservation produce a result at N1 on cycle 3.
-(define_insn_reservation "neon_vld1_3_4_regs" 3
+(define_insn_reservation "cortex_a8_neon_vld1_3_4_regs" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vld1_3_4_regs"))
   "cortex_a8_neon_ls_3")
 
 ;; Instructions using this reservation produce a result at N2 on cycle 2.
-(define_insn_reservation "neon_vld2_2_regs_vld1_vld2_all_lanes" 3
+(define_insn_reservation "cortex_a8_neon_vld2_2_regs_vld1_vld2_all_lanes" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vld2_2_regs_vld1_vld2_all_lanes"))
   "cortex_a8_neon_ls_2")
 
 ;; Instructions using this reservation produce a result at N2 on cycle 3.
-(define_insn_reservation "neon_vld2_4_regs" 4
+(define_insn_reservation "cortex_a8_neon_vld2_4_regs" 4
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vld2_4_regs"))
   "cortex_a8_neon_ls_3")
 
 ;; Instructions using this reservation produce a result at N2 on cycle 4.
-(define_insn_reservation "neon_vld3_vld4" 5
+(define_insn_reservation "cortex_a8_neon_vld3_vld4" 5
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vld3_vld4"))
   "cortex_a8_neon_ls_4")
 
 ;; Instructions using this reservation read their source operands at N1.
-(define_insn_reservation "neon_vst1_1_2_regs_vst2_2_regs" 0
+(define_insn_reservation "cortex_a8_neon_vst1_1_2_regs_vst2_2_regs" 0
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vst1_1_2_regs_vst2_2_regs"))
   "cortex_a8_neon_ls_2")
 
 ;; Instructions using this reservation read their source operands at N1.
-(define_insn_reservation "neon_vst1_3_4_regs" 0
+(define_insn_reservation "cortex_a8_neon_vst1_3_4_regs" 0
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vst1_3_4_regs"))
   "cortex_a8_neon_ls_3")
 
 ;; Instructions using this reservation read their source operands at N1.
-(define_insn_reservation "neon_vst2_4_regs_vst3_vst4" 0
+(define_insn_reservation "cortex_a8_neon_vst2_4_regs_vst3_vst4" 0
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vst2_4_regs_vst3_vst4"))
   "cortex_a8_neon_ls_4")
 
 ;; Instructions using this reservation read their source operands at N1.
-(define_insn_reservation "neon_vst3_vst4" 0
+(define_insn_reservation "cortex_a8_neon_vst3_vst4" 0
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vst3_vst4"))
   "cortex_a8_neon_ls_4")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N2 on cycle 3.
-(define_insn_reservation "neon_vld1_vld2_lane" 4
+(define_insn_reservation "cortex_a8_neon_vld1_vld2_lane" 4
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vld1_vld2_lane"))
   "cortex_a8_neon_ls_3")
 
 ;; Instructions using this reservation read their source operands at N1, and
 ;; produce a result at N2 on cycle 5.
-(define_insn_reservation "neon_vld3_vld4_lane" 6
+(define_insn_reservation "cortex_a8_neon_vld3_vld4_lane" 6
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vld3_vld4_lane"))
   "cortex_a8_neon_ls_5")
 
 ;; Instructions using this reservation read their source operands at N1.
-(define_insn_reservation "neon_vst1_vst2_lane" 0
+(define_insn_reservation "cortex_a8_neon_vst1_vst2_lane" 0
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vst1_vst2_lane"))
   "cortex_a8_neon_ls_2")
 
 ;; Instructions using this reservation read their source operands at N1.
-(define_insn_reservation "neon_vst3_vst4_lane" 0
+(define_insn_reservation "cortex_a8_neon_vst3_vst4_lane" 0
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vst3_vst4_lane"))
   "cortex_a8_neon_ls_3")
 
 ;; Instructions using this reservation produce a result at N2 on cycle 2.
-(define_insn_reservation "neon_vld3_vld4_all_lanes" 3
+(define_insn_reservation "cortex_a8_neon_vld3_vld4_all_lanes" 3
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_vld3_vld4_all_lanes"))
   "cortex_a8_neon_ls_3")
 
 ;; Instructions using this reservation produce a result at N2.
-(define_insn_reservation "neon_mcr" 2
+(define_insn_reservation "cortex_a8_neon_mcr" 2
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mcr"))
   "cortex_a8_neon_perm")
 
 ;; Instructions using this reservation produce a result at N2.
-(define_insn_reservation "neon_mcr_2_mcrr" 2
+(define_insn_reservation "cortex_a8_neon_mcr_2_mcrr" 2
   (and (eq_attr "tune" "cortexa8")
        (eq_attr "neon_type" "neon_mcr_2_mcrr"))
   "cortex_a8_neon_perm_2")
 
 ;; Exceptions to the default latencies.
 
-(define_bypass 1 "neon_mcr_2_mcrr"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 1 "neon_mcr"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_vld3_vld4_all_lanes"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_vld3_vld4_lane"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 3 "neon_vld1_vld2_lane"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 4 "neon_vld3_vld4"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 3 "neon_vld2_4_regs"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_vld2_2_regs_vld1_vld2_all_lanes"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_vld1_3_4_regs"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 1 "neon_vld1_1_2_regs"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 0 "neon_ldr"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 3 "neon_bp_3cycle"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_bp_2cycle"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 1 "neon_bp_simple"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 9 "neon_fp_vrecps_vrsqrts_qqq"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 8 "neon_fp_vrecps_vrsqrts_ddd"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 9 "neon_fp_vmla_qqq_scalar"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 8 "neon_fp_vmla_ddd_scalar"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 9 "neon_fp_vmla_qqq"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 8 "neon_fp_vmla_ddd"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_fp_vmul_qqd"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 4 "neon_fp_vmul_ddd"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 4 "neon_fp_vsum"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_fp_vadd_qqq_vabs_qq"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 4 "neon_fp_vadd_ddd_vabs_dd"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_vsra_vrsra"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 4 "neon_vqshl_vrshl_vqrshl_qqq"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 0 "neon_vshl_ddd"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 3 "neon_shift_3"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 3 "neon_shift_2"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_shift_1"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_mla_ddd_16_scalar_qdd_32_16_long_scalar"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 8 "neon_mul_qqd_32_scalar"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_mul_ddd_16_scalar_32_16_long_scalar"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 8 "neon_mla_qqq_32_qqd_32_scalar"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 6 "neon_mla_ddd_32_qqd_16_ddd_32_scalar_qdd_64_32_long_scalar_qdd_64_32_long"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 6 "neon_mla_qqq_8_16"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_mla_ddd_8_16_qdd_16_8_long_32_16_long"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 6 "neon_mul_qdd_64_32_long_qqd_16_ddd_32_scalar_64_32_long_scalar"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 6 "neon_mul_qqq_8_16_32_ddd_32"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_mul_ddd_8_16_qdd_16_8_long_32_16_long"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_vsma"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 6 "neon_vaba_qqq"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 5 "neon_vaba"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_vmov"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 3 "neon_vqneg_vqabs"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 3 "neon_int_5"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 3 "neon_int_4"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_int_3"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_int_2"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
-
-(define_bypass 2 "neon_int_1"
-               "neon_int_1,\
-               neon_int_4,\
-               neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mul_qqq_8_16_32_ddd_32,\
-               neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
-               neon_mla_qqq_8_16,\
-               neon_fp_vadd_ddd_vabs_dd,\
-               neon_fp_vadd_qqq_vabs_qq,\
-               neon_fp_vmla_ddd,\
-               neon_fp_vmla_qqq,\
-               neon_fp_vrecps_vrsqrts_ddd,\
-               neon_fp_vrecps_vrsqrts_qqq")
+(define_bypass 1 "cortex_a8_neon_mcr_2_mcrr"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 1 "cortex_a8_neon_mcr"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_vld3_vld4_all_lanes"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_vld3_vld4_lane"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 3 "cortex_a8_neon_vld1_vld2_lane"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 4 "cortex_a8_neon_vld3_vld4"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 3 "cortex_a8_neon_vld2_4_regs"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_vld2_2_regs_vld1_vld2_all_lanes"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_vld1_3_4_regs"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 1 "cortex_a8_neon_vld1_1_2_regs"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 0 "cortex_a8_neon_ldr"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 3 "cortex_a8_neon_bp_3cycle"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_bp_2cycle"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 1 "cortex_a8_neon_bp_simple"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 9 "cortex_a8_neon_fp_vrecps_vrsqrts_qqq"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 8 "cortex_a8_neon_fp_vrecps_vrsqrts_ddd"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 9 "cortex_a8_neon_fp_vmla_qqq_scalar"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 8 "cortex_a8_neon_fp_vmla_ddd_scalar"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 9 "cortex_a8_neon_fp_vmla_qqq"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 8 "cortex_a8_neon_fp_vmla_ddd"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_fp_vmul_qqd"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 4 "cortex_a8_neon_fp_vmul_ddd"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 4 "cortex_a8_neon_fp_vsum"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_fp_vadd_qqq_vabs_qq"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 4 "cortex_a8_neon_fp_vadd_ddd_vabs_dd"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_vsra_vrsra"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 4 "cortex_a8_neon_vqshl_vrshl_vqrshl_qqq"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 0 "cortex_a8_neon_vshl_ddd"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 3 "cortex_a8_neon_shift_3"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 3 "cortex_a8_neon_shift_2"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_shift_1"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_mla_ddd_16_scalar_qdd_32_16_long_scalar"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 8 "cortex_a8_neon_mul_qqd_32_scalar"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_mul_ddd_16_scalar_32_16_long_scalar"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 8 "cortex_a8_neon_mla_qqq_32_qqd_32_scalar"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 6 "cortex_a8_neon_mla_ddd_32_qqd_16_ddd_32_scalar_qdd_64_32_long_scalar_qdd_64_32_long"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 6 "cortex_a8_neon_mla_qqq_8_16"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 6 "cortex_a8_neon_mul_qdd_64_32_long_qqd_16_ddd_32_scalar_64_32_long_scalar"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 6 "cortex_a8_neon_mul_qqq_8_16_32_ddd_32"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_vsma"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 6 "cortex_a8_neon_vaba_qqq"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 5 "cortex_a8_neon_vaba"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_vmov"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 3 "cortex_a8_neon_vqneg_vqabs"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 3 "cortex_a8_neon_int_5"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 3 "cortex_a8_neon_int_4"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_int_3"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_int_2"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")
+
+(define_bypass 2 "cortex_a8_neon_int_1"
+               "cortex_a8_neon_int_1,\
+               cortex_a8_neon_int_4,\
+               cortex_a8_neon_mul_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mul_qqq_8_16_32_ddd_32,\
+               cortex_a8_neon_mla_ddd_8_16_qdd_16_8_long_32_16_long,\
+               cortex_a8_neon_mla_qqq_8_16,\
+               cortex_a8_neon_fp_vadd_ddd_vabs_dd,\
+               cortex_a8_neon_fp_vadd_qqq_vabs_qq,\
+               cortex_a8_neon_fp_vmla_ddd,\
+               cortex_a8_neon_fp_vmla_qqq,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_ddd,\
+               cortex_a8_neon_fp_vrecps_vrsqrts_qqq")