Skip to content
Snippets Groups Projects
  1. Apr 23, 2000
    • dtc's avatar
      o Where possible schedule FNSTSW directly after FCOM, FUCOM, and FTST · fe72ccd4
      dtc authored
        instructions as this it reportedly faster on the Athlon.
      
      o Flush the PPro float equality comparison code as it is not possible to
        correctly handle unordered arguments with a FUCOMI based sequence,
        except using multiple JMP instructions which appears to be slower.
      fe72ccd4
  2. Apr 22, 2000
    • dtc's avatar
      o Fix a bug in the long-float comparison VOPs when both arguments are · 381b1e7b
      dtc authored
        in the same register and not in FR0.
      
      o Cleanup the single-float and double-float comparision VOPS.
      
      o Add enhanced support for the Pentium Pro and later processors, using
        the FCOMI and FUCOMI instructions to speed float comparisons; conditional
        upon the :ppro feature. Support for =/float is not enabled yet as it is
        does not handle NaNs as required.
      381b1e7b
  3. Apr 21, 2000
  4. Apr 19, 2000
  5. Apr 14, 2000
  6. Apr 13, 2000
  7. Apr 12, 2000
  8. Apr 07, 2000
    • dtc's avatar
      From Raymond Toy: · 63f036e0
      dtc authored
      o Define %max/%min functions for byte-compilation and constant
        folding.
      o Always evaluate all arguments, even if we can prove we won't use
        that argument.  (Don't shortcut %max/%min.)
      63f036e0
  9. Apr 06, 2000
  10. Apr 05, 2000
  11. Apr 02, 2000
    • dtc's avatar
      o Handle the inspection of general CONS objects. These had been giving · c6a7bd9c
      dtc authored
        an error when checking for the list length; and more work is needed
        to handle improper lists.
      c6a7bd9c
    • dtc's avatar
      From Raymond Toy: · 0dd05a02
      dtc authored
      o Add deftransforms for max/min to call the new FP vops for max/min for float
        arguments; currently only supported on the Sparc V9.
      0dd05a02
    • dtc's avatar
      From Raymond Toy: · dfee6bab
      dtc authored
      o Add support for complex FP vops, enabled by the feature :complex-fp-vop
        which should work for all Sparcs.
      dfee6bab
    • dtc's avatar
      From Raymond Toy: · a12698c0
      dtc authored
      o Add complex FP vops, enabled by the feature :complex-fp-vop.
      o Add VOPS for max/min of floats using conditional moves for the Sparc V9.
      a12698c0
    • dtc's avatar
      From Raymond Toy: · 9ebd376e
      dtc authored
      o Extend ir1-transform-< to handle union types, and apply the transform
        when both args are real.
      9ebd376e
  12. Mar 14, 2000
  13. Mar 10, 2000
  14. Feb 28, 2000
    • dtc's avatar
      From Pierre R. Mai: · 180e2498
      dtc authored
      o Weaken the compiler warnings generated for unused or used variables
        to be style-warnings as specified by ANSI CL.
      o Check for ignore declarations of free variables, and when such
        variables are found generate a style-warning noting that the declaration
        has been ignored.
      180e2498
    • dtc's avatar
      o Use linux-nm to allow the symbols to be tailored to this system, · 20f08ea9
      dtc authored
        rather than hacking the main Makefile.
      o Link with the --export-dynamic flag as this reportedly assists code
        that calls back into lisp.
      o Cleanup of the Config files.
      20f08ea9
    • dtc's avatar
      o Cleanup of the Config files. · c38914d0
      dtc authored
      c38914d0
  15. Feb 25, 2000
    • dtc's avatar
      From Peter Van Eynde's branch: · fbea2049
      dtc authored
      o The to-readtable argument to copy-readtable may be nil in which case a
        new readtable is created.
      o Implement issue GET-MACRO-CHARACTER-READTABLE, the readtable argument
        of get-macro-character and get-dispatch-macro-character is a readtable
        designator and may be nil in which case it denotes the standard
        readtable.
      fbea2049
    • dtc's avatar
      Implement issue GET-MACRO-CHARACTER-READTABLE, the readtable argument · cf2d14bc
      dtc authored
      of get-macro-character and get-dispatch-macro-character is a readtable
      designator and may be nil in which case it denotes the standard
      readtable. From Peter Van Eynde's branch.
      cf2d14bc
  16. Feb 24, 2000
    • dtc's avatar
      From Raymond Toy: · 306bf630
      dtc authored
      o Correct some more instances of the quad FP instruction renaming.
      o Added NEGATE-DOUBLE-REG and ABS-DOUBLE-REG functions to abstract out
        their uses for sparc-v7 and sparc-v8/v9.
      o Rename the fsqrt vop for long-float to another name so that it
        doesn't overwrite the one for double-float.
      306bf630
    • dtc's avatar
      From Raymond Toy: · 755952d0
      dtc authored
      o Correct the printing of the conditional move instruction when using
        the %fcc<n> condition registers.
      755952d0
  17. Feb 21, 2000
    • dtc's avatar
      From Raymond Toy: · 85d71464
      dtc authored
      o Correct the register dependencies for the conditonal move
        instructions.
      o Have the disassembler print "MOV" for the move instruction to be
        consistent with the V9 manual and to prevent confusion with the
        conditional move instructions.
      85d71464
  18. Feb 18, 2000
  19. Feb 17, 2000
    • dtc's avatar
      From Raymond Toy: · 982c45e6
      dtc authored
      o Rename all quad FP instructions to use a Q instead of an X.
      o The f{s,d}toir instruction apparently do not exist on Ultrasparcs,
        so undo the extended register stuff, and remove the definition of
        fxtoir.
      o Add the fxto{s,d,q} and f{s,d,q}tox instructions (convert from/to
        64-bit integer) that I forgot .
      982c45e6
  20. Feb 16, 2000
    • dtc's avatar
      o Added truncate vops for V8 and V9. The generic versions in · 19da2b9e
      dtc authored
        assembly/sparc/arith.lisp won't be used anymore on V8 or V9.
      
      o fast-ash now uses a conditional move instruction on the V9.
        Should help since it gets rid of a branch instruction, but I
        can't really tell.
      
      o The fast-* for signed and unsigned multiplication appropriately
        conditionalized to work on V8 and V9, as appropriate.  (As
        discussed in my earlier patch for assembly/sparc/arith.lisp).
      
      o bignum::%digit-0-or-plusp uses a conditional move on the V9
      
      o Some miscellaneous patches to the VOP guards so that the ones
        that require sparc-64 support are only done on sparc-64
        platforms (of which there aren't any supported yet).
      
      o Added a bignum-floor vop for V8.
      19da2b9e
    • dtc's avatar
      o Branch with prediction instructions are defined. · 71c9883d
      dtc authored
      o On the V9, all branches are replaced with the corresponding
        branch-with-prediciton versions.  To minimize the number of
        changes, I did this by keeping the CMUCL instruction name the
        same, but it actually uses the prediction version.  This means,
        however, that you can't use the deprecated branch instructions
        on the V9.
      
        The disassembler, however, uses the branch-with-prediction
        instruction mnemonics, and should still be able to disassemble
        the deprecated branch instructions (I think).
      
      o Added the conditional move instructions: conditional move
        integer/FP depending on the integer or FP condition code and
        conditional move integer/FP depending on the value of an integer
        register.
      
      o The FCMP instructions can place its result into one of four FP
        condition registers on the V9.
      
      o Unified the extended FP instructions by making the printer print
        the right thing instead of having several separate formats and
        macros for them.
      
      o Added a :print-name keyword to the define-f3-inst macro.
      
      o The ld instruction is printed as lduw on the V9 because that's
        what it is and does.
      
      o Marked the deprecated V9 instructions but left them in.
      
      o For the T instruction, set the type of the target to be the
        right size for the V9.  (It's only 7 bits instead of 13 on the V9.)
      71c9883d
  21. Feb 15, 2000
  22. Feb 14, 2000
  23. Feb 12, 2000
  24. Feb 03, 2000
  25. Jan 30, 2000
  26. Jan 19, 2000
    • dtc's avatar
      Have os_validate pass the MAP_NORESERVE flag to mmap. This allows · 274a7d48
      dtc authored
      users with low swap plus memory to run without enabling the
      non-standard and perhaps undesirable overcommit_memory option. With
      this change the current scheme of splitting mmap operations into
      smaller pieces to bypass the kernel checks is no longer necessary,
      and this code has been cleaned up.
      274a7d48
Loading