This is the mail archive of the
mailing list for the GCC project.
Re: Support Xilinx PowerPC single-precision FPU
David Edelsohn wrote:
Your change to fix_truncsfsi2 references gen_xil_fix_truncsfsi2,
but no xil_fix_truncsfsi2 pattern is defined in your patches.
xil_fix_truncsfsi2 as well as a couple other patterns are not
being submitted, at least for the time being, since Xilinx has
redefined 'fctiwz' instruction to operate on SF values.
I meant to exclude the modifications to the fix_truncsfsi2
pattern as well. I believe that the support for Xilinx's 'fctiwz'
instruction can be reworked so that it doesn't need any changes to
The attached patch reverts these changes.
Michael Eager email@example.com
1960 Park Blvd., Palo Alto, CA 94306 650-325-8077
--- rs6000.md-orig 2008-10-10 12:43:08.000000000 -0700
+++ rs6000.md 2008-10-10 12:44:06.000000000 -0700
@@ -5862,18 +5862,8 @@
(clobber (match_dup 2))
(clobber (match_dup 3))
(clobber (match_dup 4))])]
- "TARGET_HARD_FLOAT && ! TARGET_DOUBLE_FLOAT && TARGET_FPRS
- && TARGET_SINGLE_FLOAT && TARGET_XILINX_FPU"
- if (TARGET_XILINX_FPU)
- operands = gen_reg_rtx (SFmode);
- operands = assign_stack_temp (SFmode, GET_MODE_SIZE (SFmode), 0);
- operands = gen_rtx_MEM (SImode, XEXP (operands, 0));
- emit_insn (gen_xil_fix_truncsfsi2(operands, operands, operands, operands, operands));
+ "TARGET_HARD_FLOAT && !TARGET_FPRS && TARGET_SINGLE_FLOAT"
; For each of these conversions, there is a define_expand, a define_insn
; with a '#' template, and a define_split (with C code). The idea is