Deprecate uses_fp parameter in RTAPI and HAL APIs

All threads now unconditionally save and restore FPU/SSE state,
making the uses_fp parameter obsolete. Rather than removing it
from the API (which would break out-of-tree components), this
commit deprecates it with a grace period:

RTAPI: uses_fp parameter is accepted but ignored; FPU state is
always saved in rtapi_task_new() regardless of the value passed.

HAL: uses_fp parameter is accepted but ignored in
hal_export_funct(), hal_export_functf(), and hal_create_thread().
All functions and threads are always marked as FP-capable
internally. The addf FP compatibility check is removed since
all threads are now FP-capable.

halcompile: fp/nofp keywords in function declarations now emit
a deprecation warning and are treated as fp (always return 1).
The keywords will be removed in a future version.

Remove fp/nofp from all in-tree .comp files, conv.comp.in
template, and mkconv.sh generator. Remove fp1= from test .hal
files. Out-of-tree .comp files will still parse but emit a
deprecation warning.

Documentation: API man pages, tutorials, and guides updated with
deprecation notices. Removed references to FP thread restrictions
that no longer apply.

No API signatures changed — out-of-tree components continue to
compile unchanged but will see warnings from halcompile if they
use fp/nofp.

Based on patch by BsAtHome.

Ref: #3895

Author: Luca Toniolo

docs/src/hal/basic-hal.adoc

@@ -68,8 +68,13 @@ Owner   CodeAddr  Arg       FP   Users  Name
 
 You have to add a function from a HAL real time component to a thread to get the function to update at the rate of the thread.
 Usually there are two threads as shown in this example.
-Some components use floating point math and must be added to a thread that supports floating point math.
-The `FP` indicates if floating point math is supported in that thread.
+
+[NOTE]
+====
+The `FP` column and the `uses_fp` parameter are deprecated.
+All threads now unconditionally save and restore floating point state.
+The `fp`/`nofp` distinction will be removed in a future version.
+====
 
 ----
 $ halrun
@@ -78,14 +83,13 @@ halcmd: show thread
 Realtime Threads:
      Period  FP     Name               (     Time, Max-Time )
      995976  YES          servo-thread (        0,        0 )
-      55332  NO            base-thread (        0,        0 )
+      55332  YES           base-thread (        0,        0 )
 ----
 
 - base-thread (the high-speed thread):
   This thread handles items that need a fast response, like making step pulses, and reading and writing the parallel port.
-  Does not support floating point math.
 - servo-thread (the slow-speed thread):
-  This thread handles items that can tolerate a slower response, like the motion controller, ClassicLadder, and the motion command handler and supports floating point math.
+  This thread handles items that can tolerate a slower response, like the motion controller, ClassicLadder, and the motion command handler.
 
 .addf Syntax and Example
 [source,{hal}]
@@ -94,9 +98,6 @@ addf <function> <thread>
 addf mux4.0 servo-thread
 ----
 
-[NOTE]
-If the component requires a floating point thread that is usually the slower servo-thread.
-
 [[sub:hal-loadusr]]
 === loadusr
 

docs/src/hal/rtcomps.adoc

@@ -162,7 +162,7 @@ image::images/stepgen-type11-14.png["Step Types: Five-Phase",align="center"]
 The component exports three functions.
 Each function acts on all of the step pulse generators - running different generators in different threads is not supported.
 
-* (funct) `stepgen.make-pulses` - High speed function to generate and count pulses (no floating point).
+* (funct) `stepgen.make-pulses` - High speed function to generate and count pulses.
 * (funct) `stepgen.update-freq` - Low speed function does position to velocity conversion, scaling and limiting.
 * (funct) `stepgen.capture-position` - Low speed function for feedback, updates latches and scales position.
 
@@ -258,7 +258,7 @@ Each PWM generator will also have some of these pins, depending on the output ty
 
 The component exports two functions. Each function acts on all of the PWM generators - running different generators in different threads is not supported.
 
-* (funct) `pwmgen.make-pulses` - High speed function to generate PWM waveforms (no floating point).
+* (funct) `pwmgen.make-pulses` - High speed function to generate PWM waveforms.
   The high speed function `pwmgen.make-pulses` should be run in the base (fastest) thread, from 10 to 50 µs depending on the capabilities of the computer.
   That thread's period determines the maximum PWM carrier frequency, as well as the resolution of the PWM or PDM signals.
   If the base thread is 50,000 ns then every 50 µs the module decides if it is time to change the state of the output.
@@ -361,7 +361,7 @@ halcmd: unloadrt encoder
 The component exports two functions.
 Each function acts on all of the encoder counters - running different counters in different threads is not supported.
 
-* (funct) `encoder.update-counters` - High speed function to count pulses (no floating point).
+* (funct) `encoder.update-counters` - High speed function to count pulses.
 * (funct) `encoder.capture-position` - Low speed function to update latches and scale position.
 
 [[sec:pid]]
@@ -515,7 +515,7 @@ Most encoder counters will count four times during one complete cycle.
 
 The component exports two functions. Each function affects all simulated encoders.
 
-* (funct) `sim-encoder.make-pulses` - High speed function to generate quadrature pulses (no floating point).
+* (funct) `sim-encoder.make-pulses` - High speed function to generate quadrature pulses.
 * (funct) `sim-encoder.update-speed` - Low speed function to read `.speed`, do scaling, and set up `.make-pulses`.
 
 [[sec:debounce]]
@@ -638,7 +638,7 @@ They were changed to pins to allow control by other components.]
 (((lut5)))
 The `lut5` component is a 5 input logic component based on a look up table.
 
-* `lut5` does not require a floating point thread.
+* `lut5` does not use floating point math.
 
 .Loading `lut5`
 ----

docs/src/hal/tools.adoc

@@ -245,8 +245,7 @@ or
  2. If no pinname is specified, default is: `motion-command-handler.time`.
  3. This app may be opened for 5 pins.
  4. Pintypes float, s32, u32, bit are supported.
- 5. The pin must be associated with a thread supporting floating point.
-    For a base thread, this may require using `loadrt motmod ... base_thread_fp=1` .
+ 5. The pin must be associated with a realtime thread.
 
 .hal-histogram Window
 image::images/hal-histogram.png["hal-histogram Window"]

docs/src/hal/tutorial.adoc

@@ -213,7 +213,6 @@ Owner   CodeAddr  Arg       FP   Users  Name
 ----
 
 The siggen component exported a single function.
-It requires floating point.
 It is not currently linked to any threads, so 'users' is
 zero footnote:[CodeAddr and Arg fields were used during development and
 should probably disappear.].
@@ -241,7 +240,7 @@ Realtime Threads:
 
 It did. The period is not exactly 1,000,000 ns because of hardware
 limitations, but we have a thread that runs at approximately the
-correct rate, and which can handle floating point functions.
+correct rate.
 The next step is to connect the function to the thread:
 
 .Add Function
@@ -557,13 +556,15 @@ component.
 halrun
 halcmd: loadrt stepgen step_type=0,0 ctrl_type=v,v
 halcmd: loadrt siggen
-halcmd: loadrt threads name1=fast fp1=0 period1=50000 name2=slow period2=1000000
+halcmd: loadrt threads name1=fast period1=50000 name2=slow period2=1000000
 ----
 
 The first command loads two step generators, both configured to generate stepping type 0.
 The second command loads our old friend siggen, and the third one creates two threads,
 a fast one with a period of 50 microseconds (µs) and a slow one with a period of 1 millisecond (ms).
-The fast thread doesn't support floating point functions.
+
+NOTE: The `fp1=` parameter is deprecated and ignored.
+All threads now unconditionally support floating point.
 
 As before, we can use `halcmd show` to take a look at the HAL.
 This time we have a lot more pins and parameters than before:
@@ -731,7 +732,6 @@ is time to take a step, and if so sets the outputs accordingly.
 For smooth step pulses, it should run as frequently as possible.
 Because it needs to run so fast, 'make_pulses'
 is highly optimized and performs only a few calculations.
-Unlike the others, it does not need floating point math.
 
 The last function, `stepgen.update-freq`, is responsible for doing
 scaling and some other calculations that need to be performed
@@ -754,13 +754,13 @@ halcmd: show thread
 Realtime Threads:
      Period  FP     Name               (     Time, Max-Time )
      996980  YES                  slow (        0,        0 )
-      49849  NO                   fast (        0,        0 )
+      49849  YES                  fast (        0,        0 )
 ----
 
 The two threads were created when we loaded `threads`.
-The first one, 'slow', runs every millisecond, and is capable of running floating point functions.
+The first one, 'slow', runs every millisecond.
 We will use it for `siggen.0.update` and `stepgen.update_freq`.
-The second thread is 'fast', which runs every 50 microseconds (µs), and does not support floating point.
+The second thread is 'fast', which runs every 50 microseconds (µs).
 We will use it for `stepgen.make_pulses`.
 To connect the functions to the proper thread, we use the `addf` command.
 We specify the function first, followed by the thread.

docs/src/man/man3/hal_create_thread.3.adoc

@@ -17,7 +17,8 @@ name::
 period::
   The interval, in nanoseconds, between iterations of the thread.
 uses_fp::
-  Must be nonzero if a function which uses floating-point will be attached to this thread.
+  Deprecated and ignored. All threads now unconditionally save and restore
+  floating point state. This parameter will be removed in a future version.
 
 == DESCRIPTION
 

docs/src/man/man3/hal_export_funct.3.adoc

@@ -21,8 +21,8 @@ funct::
 arg::
   The argument to be passed as the first parameter of _funct_.
 uses_fp::
-  Nonzero if the function uses floating-point operations, including
-  assignment of floating point values with "=".
+  Deprecated and ignored. All threads now unconditionally save and restore
+  floating point state. This parameter will be removed in a future version.
 reentrant::
   If reentrant is non-zero, the function may be preempted and called again before the first call completes.
   Otherwise, it may only be added to one thread.

docs/src/man/man3/rtapi_task_new.3.adoc

@@ -24,7 +24,8 @@ arg::
 prio::
   A task priority value returned by *rtapi_prio_xxxx*
 uses_fp::
-  A flag that tells the OS whether the task uses floating point or not.
+  Deprecated and ignored. All tasks now unconditionally save and restore
+  floating point state. This parameter will be removed in a future version.
 task_id::
   A task ID returned by a previous call to *rtapi_task_new*
 

docs/src/man/man9/motion.9.adoc

@@ -60,10 +60,9 @@ the motion-controller function.
 
 == DESCRIPTION
 
-By default, the base thread does not support floating point. Software
-stepping, software encoder counting, and software pwm do not use floating point.
-*base_thread_fp* can be used to enable floating point in
-the base thread (for example for brushless DC motor control).
+NOTE: The *base_thread_fp* parameter is deprecated and ignored.
+All threads now unconditionally save and restore floating point state.
+This parameter will be removed in a future version.
 
 These pins and parameters are created by the realtime *motmod* module.
 This module provides a HAL interface for LinuxCNC's motion planner.

src/hal/components/abs_s32.comp

@@ -22,7 +22,7 @@ pin out bit is_positive "TRUE if input is positive, FALSE if input is 0 or negat
 pin out bit is_negative "TRUE if input is negative, FALSE if input is 0 or positive";
 
 option period no;
-function _ nofp;
+function _;
 license "GPL";
 author "Sebastian Kuzminsky";
 ;;

src/hal/components/abs_s64.comp

@@ -24,7 +24,7 @@ pin out bit is_positive "true if input is positive, false if input is 0 or negat
 pin out bit is_negative "true if input is negative, false if input is 0 or positive";
 
 option period no;
-function _ nofp;
+function _;
 license "GPL";
 author "ArcEye based on code from Sebastian Kuzminsky";
 ;;