This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: RESUBMIT: PowerPC -- support single-precision FPU


Andreas Tobler wrote:
Hi Michael,

Michael Eager wrote:

Just having TARGET_DOUBLE_FLOAT, instead of (TARGET_DOUBLE_FLOAT &&
mode == DFmode) should be the same as previous behavior.

Other tests have explicit checks for TFmode, which are missing.
This would probably be better:


@@ -5200,7 +5250,9 @@ #define USE_FP_FOR_ARG_P(CUM,MODE,TYPE) \ (SCALAR_FLOAT_MODE_P (MODE) \ && (CUM)->fregno <= FP_ARG_MAX_REG \ - && TARGET_HARD_FLOAT && TARGET_FPRS) + && TARGET_HARD_FLOAT && TARGET_FPRS \ + && ((TARGET_DOUBLE_FLOAT && ((MODE) == DFmode || (MODE) == (TFmode))\ + || (TARGET_SINGLE_FLOAT && (MODE) == SFmode)))


@@ -5745,7 +5797,9 @@ cum->words = align_words + n_words;

       if (SCALAR_FLOAT_MODE_P (mode)
-      && TARGET_HARD_FLOAT && TARGET_FPRS)
+      && TARGET_HARD_FLOAT && TARGET_FPRS
+       && ((TARGET_DOUBLE_FLOAT && (mode == DFmode || mode == TFmode))
+           || (TARGET_SINGLE_FLOAT && mode == SFmode)))


Let me complete a build with current svn, including tests. After that I'll apply the two hunks above and rerun. I'll do this on ppc-darwin 32-bit. We then get an idea if it breaks here or not.

After that, we can see what is needed to get TARGET_SINGLE_FLOAT to work correctly w/o breaking existing ABI's and FP rules, ok?

Sounds good. Unfortunately, I don't have a PPC w/ standard FPU as a test system, so I can't run regression tests.


-- Michael Eager eager@eagercon.com 1960 Park Blvd., Palo Alto, CA 94306 650-325-8077


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]