| Message ID | AM5PR0802MB2610C0616A19887E6FA5928E83A30@AM5PR0802MB2610.eurprd08.prod.outlook.com |
|---|---|
| State | New |
| Headers | show |
On Thu, Nov 3, 2016 at 12:20 PM, Wilco Dijkstra <Wilco.Dijkstra@arm.com> wrote: > Fix ldrd offsets of Thumb-2 - for TARGET_LDRD the range is +-1020, > without -255..4091. This reduces the number of addressing instructions > when using DI mode operations (such as in PR77308). > > Bootstrap & regress OK. > > ChangeLog: > 2015-11-03 Wilco Dijkstra <wdijkstr@arm.com> > > gcc/ > * config/arm/arm.c (arm_legitimate_index_p): Add comment. > (thumb2_legitimate_index_p): Use correct range for DI/DF mode. > -- > > diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c > index 3c4c7042d9c2101619722b5822b3d1ca37d637b9..5d12cf9c46c27d60a278d90584bde36ec86bb3fe 100644 > --- a/gcc/config/arm/arm.c > +++ b/gcc/config/arm/arm.c > @@ -7486,6 +7486,8 @@ arm_legitimate_index_p (machine_mode mode, rtx index, RTX_CODE outer, > { > HOST_WIDE_INT val = INTVAL (index); > > + /* Assume we emit ldrd or 2x ldr if !TARGET_LDRD. > + If vldr is selected it uses arm_coproc_mem_operand. */ > if (TARGET_LDRD) > return val > -256 && val < 256; > else > @@ -7613,11 +7615,13 @@ thumb2_legitimate_index_p (machine_mode mode, rtx index, int strict_p) > if (code == CONST_INT) > { > HOST_WIDE_INT val = INTVAL (index); > - /* ??? Can we assume ldrd for thumb2? */ > - /* Thumb-2 ldrd only has reg+const addressing modes. */ > - /* ldrd supports offsets of +-1020. > - However the ldr fallback does not. */ > - return val > -256 && val < 256 && (val & 3) == 0; > + /* Thumb-2 ldrd only has reg+const addressing modes. > + Assume we emit ldrd or 2x ldr if !TARGET_LDRD. > + If vldr is selected it uses arm_coproc_mem_operand. */ > + if (TARGET_LDRD) I suspect this should be : if (TARGET_LDRD && !fix_cm3_ldrd) - I am a bit worried about this change because of the non-uniformity with ldr and the fallout with other places where things may break with this. I would like a test with -mcpu=cortex-m3/-mthumb as well for an arm-none-eabi target to see what the fallout of this change is on that ... Ramana
Ramana Radhakrishnan wrote: > On Thu, Nov 3, 2016 at 12:20 PM, Wilco Dijkstra <Wilco.Dijkstra@arm.com> wrote: > HOST_WIDE_INT val = INTVAL (index); > - /* ??? Can we assume ldrd for thumb2? */ > - /* Thumb-2 ldrd only has reg+const addressing modes. */ > - /* ldrd supports offsets of +-1020. > - However the ldr fallback does not. */ > - return val > -256 && val < 256 && (val & 3) == 0; > + /* Thumb-2 ldrd only has reg+const addressing modes. > + Assume we emit ldrd or 2x ldr if !TARGET_LDRD. > + If vldr is selected it uses arm_coproc_mem_operand. */ > + if (TARGET_LDRD) > I suspect this should be : if (TARGET_LDRD && !fix_cm3_ldrd) - I am a > bit worried about this change because of the non-uniformity with ldr > and the fallout with other places where things may break with this. I > would like a test with -mcpu=cortex-m3/-mthumb as well for an > arm-none-eabi target to see what the fallout of this change is on that Well it works fine given that Thumb-2 supports add/sub up to 4KB, so the existing expansion into add+ldrd for the fix_cm3_ldrd case works fine. I ran a bootstrap with fix_cm3_ldrd forced to true, and that completed without any issues. Wilco
diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c index 3c4c7042d9c2101619722b5822b3d1ca37d637b9..5d12cf9c46c27d60a278d90584bde36ec86bb3fe 100644 --- a/gcc/config/arm/arm.c +++ b/gcc/config/arm/arm.c @@ -7486,6 +7486,8 @@ arm_legitimate_index_p (machine_mode mode, rtx index, RTX_CODE outer, { HOST_WIDE_INT val = INTVAL (index); + /* Assume we emit ldrd or 2x ldr if !TARGET_LDRD. + If vldr is selected it uses arm_coproc_mem_operand. */ if (TARGET_LDRD) return val > -256 && val < 256; else @@ -7613,11 +7615,13 @@ thumb2_legitimate_index_p (machine_mode mode, rtx index, int strict_p) if (code == CONST_INT) { HOST_WIDE_INT val = INTVAL (index); - /* ??? Can we assume ldrd for thumb2? */ - /* Thumb-2 ldrd only has reg+const addressing modes. */ - /* ldrd supports offsets of +-1020. - However the ldr fallback does not. */ - return val > -256 && val < 256 && (val & 3) == 0; + /* Thumb-2 ldrd only has reg+const addressing modes. + Assume we emit ldrd or 2x ldr if !TARGET_LDRD. + If vldr is selected it uses arm_coproc_mem_operand. */ + if (TARGET_LDRD) + return IN_RANGE (val, -1020, 1020) && (val & 3) == 0; + else + return IN_RANGE (val, -255, 4095 - 4); } else return 0;
Fix ldrd offsets of Thumb-2 - for TARGET_LDRD the range is +-1020, without -255..4091. This reduces the number of addressing instructions when using DI mode operations (such as in PR77308). Bootstrap & regress OK. ChangeLog: 2015-11-03 Wilco Dijkstra <wdijkstr@arm.com> gcc/ * config/arm/arm.c (arm_legitimate_index_p): Add comment. (thumb2_legitimate_index_p): Use correct range for DI/DF mode. --