| Message ID | 11068946.f4xvach6CP@polaris |
|---|---|
| State | New |
| Headers | show |
Hi! On Sat, Oct 08, 2016 at 07:53:42PM +0200, Eric Botcazou wrote: > this fixes reload failures in 64-bit mode with -mcmodel=medium for targets > that define ASM_OUTPUT_SPECIAL_POOL_ENTRY_P to 0; that's the case for the > VxWorks port we'll submit because of the compatibility with the system linker, > which rejects .got sections in object files. > > When reload ends up calling rs6000_emit_move on a register on the LHS and > (const (symbol_ref) (const_int)) on the RHS, the function invokes: > > operands[1] = force_const_mem (mode, operands[1]); > > creating (mem (symbol_ref)), which then needs to be legitimized on TARGET_TOC > targets. That's done by the immediately following block: > > if (TARGET_TOC > && GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF > && constant_pool_expr_p (XEXP (operands[1], 0)) > && ASM_OUTPUT_SPECIAL_POOL_ENTRY_P ( > get_pool_constant (XEXP (operands[1], 0)), > get_pool_mode (XEXP (operands[1], 0)))) > { > rtx tocref = create_TOC_reference (XEXP (operands[1], 0), > operands[0]); > operands[1] = gen_const_mem (mode, tocref); > set_mem_alias_set (operands[1], get_TOC_alias_set ()); > } > > on most targets, but is disabled if ASM_OUTPUT_SPECIAL_POOL_ENTRY_P is 0. Why does this not fail on darwin? It can reach this code afaics, and it has ASM_OUTPUT_SPECIAL_POOL_ENTRY_P always 0. > Now you don't need to have a special pool to call create_TOC_reference, you > can call it for regular TOC references as well, as done a few lines above: > > /* If this is a SYMBOL_REF that refers to a constant pool entry, > and we have put it in the TOC, we just need to make a TOC-relative > reference to it. */ > if (TARGET_TOC > && GET_CODE (operands[1]) == SYMBOL_REF > && use_toc_relative_ref (operands[1], mode)) > operands[1] = create_TOC_reference (operands[1], operands[0]); > > So the attached patch does it there too. > > Tested on PowerPC64/Linux (LRA) and VxWorks (reload), OK for the mainline? Not sure yet, will get back to you. One comment in the meantime... > --- config/rs6000/rs6000.c (revision 240888) > +++ config/rs6000/rs6000.c (working copy) > @@ -10590,10 +10590,7 @@ rs6000_emit_move (rtx dest, rtx source, > > if (TARGET_TOC > && GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF > - && constant_pool_expr_p (XEXP (operands[1], 0)) > - && ASM_OUTPUT_SPECIAL_POOL_ENTRY_P ( > - get_pool_constant (XEXP (operands[1], 0)), > - get_pool_mode (XEXP (operands[1], 0)))) > + && use_toc_relative_ref (XEXP (operands[1], 0), Pmode)) Use "mode" instead of "Pmode" here? Segher
> Why does this not fail on darwin? It can reach this code afaics, and > it has ASM_OUTPUT_SPECIAL_POOL_ENTRY_P always 0. Probably because Darwin doesn't use the TOC at all. > Use "mode" instead of "Pmode" here? No, "mode" is the mode of the MEM, not that of the SYMBOL_REF.
[ sorry for losing track of this patch ] On Sun, Oct 09, 2016 at 10:32:51AM +0200, Eric Botcazou wrote: > > Use "mode" instead of "Pmode" here? > > No, "mode" is the mode of the MEM, not that of the SYMBOL_REF. I still don't see it, could you explain a bit more? Segher
> > No, "mode" is the mode of the MEM, not that of the SYMBOL_REF. > > I still don't see it, could you explain a bit more? MODE is the mode of operands[1] before: operands[1] = force_const_mem (mode, operands[1]); and after. But the test is on the address of the MEM, not on the MEM itself: if (TARGET_TOC && GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF && use_toc_relative_ref (XEXP (operands[1], 0), Pmode)) because it's the mode of SYMBOL_REF we are interesting in (and force_const_mem guarantees that it's Pmode). IOW you could theoretically have mode == SImode and we would still need to pass Pmode to use_toc_relative_ref (of course the whole thing is guarded with mode == Pmode so that's a little artificial).
On Tue, Oct 18, 2016 at 01:09:24PM +0200, Eric Botcazou wrote: > > > No, "mode" is the mode of the MEM, not that of the SYMBOL_REF. > > > > I still don't see it, could you explain a bit more? > > MODE is the mode of operands[1] before: > > operands[1] = force_const_mem (mode, operands[1]); > > and after. But the test is on the address of the MEM, not on the MEM itself: > > if (TARGET_TOC > && GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF > && use_toc_relative_ref (XEXP (operands[1], 0), Pmode)) > > because it's the mode of SYMBOL_REF we are interesting in (and force_const_mem > guarantees that it's Pmode). We need to pass the mode of the actual datum we would put in the TOC to the use_toc_relative_ref function, not the mode of its address. I must be missing something... Segher
> We need to pass the mode of the actual datum we would put in the TOC to > the use_toc_relative_ref function, not the mode of its address. Right, but this mode is not "mode", the TOC contains only Pmode entries if the special constant pool is excluded.
On Tue, Oct 18, 2016 at 08:37:47PM +0200, Eric Botcazou wrote: > > We need to pass the mode of the actual datum we would put in the TOC to > > the use_toc_relative_ref function, not the mode of its address. > > Right, but this mode is not "mode", the TOC contains only Pmode entries if the > special constant pool is excluded. I don't fully understand what you mean. This code was created for PR65810, if that helps? Segher
> I don't fully understand what you mean. This code was created for > PR65810, if that helps? OK, let's turn it into "mode" then, this doesn't change anything.
Index: config/rs6000/rs6000.c =================================================================== --- config/rs6000/rs6000.c (revision 240888) +++ config/rs6000/rs6000.c (working copy) @@ -10590,10 +10590,7 @@ rs6000_emit_move (rtx dest, rtx source, if (TARGET_TOC && GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF - && constant_pool_expr_p (XEXP (operands[1], 0)) - && ASM_OUTPUT_SPECIAL_POOL_ENTRY_P ( - get_pool_constant (XEXP (operands[1], 0)), - get_pool_mode (XEXP (operands[1], 0)))) + && use_toc_relative_ref (XEXP (operands[1], 0), Pmode)) { rtx tocref = create_TOC_reference (XEXP (operands[1], 0), operands[0]);
Hi, this fixes reload failures in 64-bit mode with -mcmodel=medium for targets that define ASM_OUTPUT_SPECIAL_POOL_ENTRY_P to 0; that's the case for the VxWorks port we'll submit because of the compatibility with the system linker, which rejects .got sections in object files. When reload ends up calling rs6000_emit_move on a register on the LHS and (const (symbol_ref) (const_int)) on the RHS, the function invokes: operands[1] = force_const_mem (mode, operands[1]); creating (mem (symbol_ref)), which then needs to be legitimized on TARGET_TOC targets. That's done by the immediately following block: if (TARGET_TOC && GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF && constant_pool_expr_p (XEXP (operands[1], 0)) && ASM_OUTPUT_SPECIAL_POOL_ENTRY_P ( get_pool_constant (XEXP (operands[1], 0)), get_pool_mode (XEXP (operands[1], 0)))) { rtx tocref = create_TOC_reference (XEXP (operands[1], 0), operands[0]); operands[1] = gen_const_mem (mode, tocref); set_mem_alias_set (operands[1], get_TOC_alias_set ()); } on most targets, but is disabled if ASM_OUTPUT_SPECIAL_POOL_ENTRY_P is 0. Now you don't need to have a special pool to call create_TOC_reference, you can call it for regular TOC references as well, as done a few lines above: /* If this is a SYMBOL_REF that refers to a constant pool entry, and we have put it in the TOC, we just need to make a TOC-relative reference to it. */ if (TARGET_TOC && GET_CODE (operands[1]) == SYMBOL_REF && use_toc_relative_ref (operands[1], mode)) operands[1] = create_TOC_reference (operands[1], operands[0]); So the attached patch does it there too. Tested on PowerPC64/Linux (LRA) and VxWorks (reload), OK for the mainline? 2016-10-08 Eric Botcazou <ebotcazou@adacore.com> * config/rs6000/rs6000.c (rs6000_emit_move): Also use a TOC reference after forcing to constant memory when the code model is medium.