Release 0.6.0 "Christmas": Introducing variable precision

This release was a bit longer in coming than earlier ones, because we are about
to reach the limits of what can fit on a DM42. This release uses 711228 bytes
out of the 716800 (99.2%).

Without the Intel Decimal Library code, we use only 282980 bytes. This means
that the Intel Decimal Library code uses 60.2% of the total code space. Being
able to move further requires a rather radical rethinking of the project, where
we replace the Intel Decimal Library with size-optimized decimal code.

As a result, release 0.6.0 introduces a new table-free and variable-precision
implementation of decimal computations. In this release, most operations are
implemented, but some features are still missing (e.g. Gamma function). This
release will be simultaneous with 0.5.2, which is functionally equivalent but
still uses the Intel Decimal library. The new implementation is much more
compact, allowing us to return to normal optimizations for the DM42 and regain
some of the lost performance. On the other hand, having to switch to a table
free implementation means that it's significantly slower than the Intel Decimal
Library. The upside of course is that you can compute with decimal numbers that
have up to 9999 digits, and a decimal exponent that can be up to 2^60
(1 152 921 504 606 846 976).

**New features**

Variable precision decimal floating point implementation for arithmetic,
trigonometrics, logs, exponential and integer factorial. Other functions may
still return "unimplemented error".

**Bug fixes**

None. If anything, this release introduces bugs in computations and performance
regressions. However, it frees *a lot* of space for further DM42 development.

**Improvements**

The `Precision` setting now sets the number of digits with a granularity of one,
between 3 and 9999. Ideal use of memory is with multiples of 12 digits, e.g. 12,
24 or 36 digits, where decimal packing does not cause lost bits.

Performance on the DM42 is somewhat improved, since it is now possible to return
to a higher level of optimization.

**Regressions**

In addition to lower performance and unimplemented functions, this version no
longer builds a compatible QSPI. This means that returning to the DM42 requires
flashing *both* the QSPI and the PGM file.

Signed-off-by: Christophe de Dinechin <christophe@dinechin.org>

doc/4-ReleaseNotes.md

@@ -26,10 +26,29 @@ have up to 9999 digits, and a decimal exponent that can be up to 2^60
 
 ### New features
 
+Variable precision decimal floating point implementation for arithmetic,
+trigonometrics, logs, exponential and integer factorial. Other functions may
+still return "unimplemented error".
+
 ### Bug fixes
 
+None. If anything, this release introduces bugs in computations and performance
+regressions. However, it frees *a lot* of space for further DM42 development.
+
 ### Improvements
 
+The `Precision` setting now sets the number of digits with a granularity of one,
+between 3 and 9999. Ideal use of memory is with multiples of 12 digits, e.g. 12,
+24 or 36 digits, where decimal packing does not cause lost bits.
+
+Performance on the DM42 is somewhat improved, since it is now possible to return
+to a higher level of optimization.
+
+### Regressions
+
+In addition to lower performance and unimplemented functions, this version no
+longer builds a compatible QSPI. This means that returning to the DM42 requires
+flashing *both* the QSPI and the PGM file.
 
 
 ## Release 0.5.2 "Christmas Eve": Reaching hard limits on the DM42

help/db48x.md

@@ -1820,10 +1820,29 @@ have up to 9999 digits, and a decimal exponent that can be up to 2^60
 
 ### New features
 
+Variable precision decimal floating point implementation for arithmetic,
+trigonometrics, logs, exponential and integer factorial. Other functions may
+still return "unimplemented error".
+
 ### Bug fixes
 
+None. If anything, this release introduces bugs in computations and performance
+regressions. However, it frees *a lot* of space for further DM42 development.
+
 ### Improvements
 
+The `Precision` setting now sets the number of digits with a granularity of one,
+between 3 and 9999. Ideal use of memory is with multiples of 12 digits, e.g. 12,
+24 or 36 digits, where decimal packing does not cause lost bits.
+
+Performance on the DM42 is somewhat improved, since it is now possible to return
+to a higher level of optimization.
+
+### Regressions
+
+In addition to lower performance and unimplemented functions, this version no
+longer builds a compatible QSPI. This means that returning to the DM42 requires
+flashing *both* the QSPI and the PGM file.
 
 
 ## Release 0.5.2 "Christmas Eve": Reaching hard limits on the DM42

help/db50x.md

@@ -1820,10 +1820,29 @@ have up to 9999 digits, and a decimal exponent that can be up to 2^60
 
 ### New features
 
+Variable precision decimal floating point implementation for arithmetic,
+trigonometrics, logs, exponential and integer factorial. Other functions may
+still return "unimplemented error".
+
 ### Bug fixes
 
+None. If anything, this release introduces bugs in computations and performance
+regressions. However, it frees *a lot* of space for further DM32 development.
+
 ### Improvements
 
+The `Precision` setting now sets the number of digits with a granularity of one,
+between 3 and 9999. Ideal use of memory is with multiples of 12 digits, e.g. 12,
+24 or 36 digits, where decimal packing does not cause lost bits.
+
+Performance on the DM32 is somewhat improved, since it is now possible to return
+to a higher level of optimization.
+
+### Regressions
+
+In addition to lower performance and unimplemented functions, this version no
+longer builds a compatible QSPI. This means that returning to the DM32 requires
+flashing *both* the QSPI and the PGM file.
 
 
 ## Release 0.5.2 "Christmas Eve": Reaching hard limits on the DM32

src/dm32/qspi_crc.h

@@ -1,4 +1,4 @@
 
-#define QSPI_DATA_SIZE  1550436
+#define QSPI_DATA_SIZE   187948
 #define QSPI_DATA_CRC  0x000cfed6
 

src/dm42/qspi_crc.h

@@ -1,4 +1,4 @@
 
-#define QSPI_DATA_SIZE  1548076
+#define QSPI_DATA_SIZE   187516
 #define QSPI_DATA_CRC  0x000cfed6