Changeset 9079

Timestamp:

02/21/08 00:57:51 (5 years ago)

Author:

BrainSlayer

Message:

fastfpe support

Location:

src/linux/xscale/linux-2.6.23/arch/arm

Files:

• Kconfig (modified) (1 diff)
• fastfpe (added)
• fastfpe/CPDO.S (added)
• fastfpe/CPDT.S (added)
• fastfpe/CPRT.S (added)
• fastfpe/Makefile (added)
• fastfpe/entry.S (added)
• fastfpe/module.c (added)
• fastfpe/round.S (added)

src/linux/xscale/linux-2.6.23/arch/arm/Kconfig

@@ -943 +943 @@
 config FPE_FASTFPE
         bool "FastFPE math emulation (EXPERIMENTAL)"
-        depends on (!AEABI || OABI_COMPAT) && !CPU_32v3 && EXPERIMENTAL
+        depends on (!AEABI || OABI_COMPAT) && EXPERIMENTAL
         ---help---
           Say Y here to include the FAST floating point emulator in the kernel.
-          This is an experimental much faster emulator which now also has full
-          precision for the mantissa.  It does not support any exceptions.
-          It is very simple, and approximately 3-6 times faster than NWFPE.
-
-          It should be sufficient for most programs.  It may be not suitable
-          for scientific calculations, but you have to check this for yourself.
-          If you do not feel you need a faster FP emulation you should better
-          choose NWFPE.
+          This is an experimental much faster emulator which is written
+          completely in ARM assembly. All instructions that are not marked as
+          deprecated in the ARM7500FE data sheet are implemented. It supports
+          single, double and double extended precision. It does support
+          exception flags, but not raising exceptions. It gives an average
+          5 times speed increase over NWFPE for FP only code.
+
+          FastFPE does not require long multiply instruction anymore and is
+          now suitable for all ARM cpus. The presence of the long multiply
+          instruction is detected during initialisation and used to speedup
+          multiply and divide.
+
+          Compliance to IEEE Std 754-1985 was verified using the testfloat
+          program of the SoftFloat package version 2a by John Hauser. All
+          operations except square root were reported to be compliant. However,
+          this is not a proof, and especially does not verify the instruction
+          parser.
+
+          You can compile both emulators into the kernel and choose one
+          of them at boot time by passing "fpe=fastfpe" or "fpe=nwfpe" as
+          kernel parameter.
+
+          It is also possible to say M to build the emulator as a module
+          (fastfpe.o). This was never tested. Only do it if you know what
+          you are doing!
+
 
 config VFP