Message ID | 20200917171530.37619-1-david.faust@oracle.com |
---|---|
State | New |
Headers | show |
Series | bpf: use xBPF signed div, mod insns when available | expand |
On 9/17/20 11:15 AM, David Faust via Gcc-patches wrote: > The 'mod' and 'div' operators in eBPF are unsigned, with no signed > counterpart. xBPF adds two new ALU operations, sdiv and smod, for > signed division and modulus, respectively. Update bpf.md with > 'define_insn' blocks for signed div and mod to use them when targetting > xBPF, and add new tests to ensure they are used appropriately. > > 2020-09-17 David Faust <david.faust@oracle.com> > > gcc/ > * config/bpf/bpf.md: Add defines for signed div and mod operators. > > gcc/testsuite/ > * gcc.target/bpf/diag-sdiv.c: New test. > * gcc.target/bpf/diag-smod.c: New test. > * gcc.target/bpf/xbpf-sdiv-1.c: New test. > * gcc.target/bpf/xbpf-smod-1.c: New test. OK. But give Jose 48hrs before committing just in case there's something he wants to comment on. Y'all are far more familiar with bpf than I ;-) jeff
>> The 'mod' and 'div' operators in eBPF are unsigned, with no signed >> counterpart. xBPF adds two new ALU operations, sdiv and smod, for >> signed division and modulus, respectively. Update bpf.md with >> 'define_insn' blocks for signed div and mod to use them when targetting >> xBPF, and add new tests to ensure they are used appropriately. >> >> 2020-09-17 David Faust <david.faust@oracle.com> >> >> gcc/ >> * config/bpf/bpf.md: Add defines for signed div and mod operators. >> >> gcc/testsuite/ >> * gcc.target/bpf/diag-sdiv.c: New test. >> * gcc.target/bpf/diag-smod.c: New test. >> * gcc.target/bpf/xbpf-sdiv-1.c: New test. >> * gcc.target/bpf/xbpf-smod-1.c: New test. > > OK. But give Jose 48hrs before committing just in case there's > something he wants to comment on. Y'all are far more familiar with bpf > than I ;-) Looks good to me! :) But the related pending patch in binutils should go in first.
> >>> The 'mod' and 'div' operators in eBPF are unsigned, with no signed >>> counterpart. xBPF adds two new ALU operations, sdiv and smod, for >>> signed division and modulus, respectively. Update bpf.md with >>> 'define_insn' blocks for signed div and mod to use them when targetting >>> xBPF, and add new tests to ensure they are used appropriately. >>> >>> 2020-09-17 David Faust <david.faust@oracle.com> >>> >>> gcc/ >>> * config/bpf/bpf.md: Add defines for signed div and mod operators. >>> >>> gcc/testsuite/ >>> * gcc.target/bpf/diag-sdiv.c: New test. >>> * gcc.target/bpf/diag-smod.c: New test. >>> * gcc.target/bpf/xbpf-sdiv-1.c: New test. >>> * gcc.target/bpf/xbpf-smod-1.c: New test. >> >> OK. But give Jose 48hrs before committing just in case there's >> something he wants to comment on. Y'all are far more familiar with bpf >> than I ;-) > > Looks good to me! :) > But the related pending patch in binutils should go in first. > Hi! I just checked in the binutils patch, but I don't actually have write access for gcc. If someone could check this in for me I'd appreciate it :) Thanks! David
Hi! On Thu, Sep 17, 2020 at 10:15:30AM -0700, David Faust via Gcc-patches wrote: > The 'mod' and 'div' operators in eBPF are unsigned, with no signed > counterpart. xBPF adds two new ALU operations, sdiv and smod, for > signed division and modulus, respectively. Update bpf.md with > 'define_insn' blocks for signed div and mod to use them when targetting > xBPF, and add new tests to ensure they are used appropriately. So why does xBPF have signed versions of the divides? Is it because it is wanted to have it in eBPF eventually? Is it because the libgcc routines are just too slow? Is it because (the generic) libgcc does not trap for MIN_INT / -1 ? Some other reason? (I'm just curious; I cannot figure it out :-) ) Segher
Hi Segher! > On Thu, Sep 17, 2020 at 10:15:30AM -0700, David Faust via Gcc-patches wrote: >> The 'mod' and 'div' operators in eBPF are unsigned, with no signed >> counterpart. xBPF adds two new ALU operations, sdiv and smod, for >> signed division and modulus, respectively. Update bpf.md with >> 'define_insn' blocks for signed div and mod to use them when targetting >> xBPF, and add new tests to ensure they are used appropriately. > > So why does xBPF have signed versions of the divides? Is it because it > is wanted to have it in eBPF eventually? Is it because the libgcc > routines are just too slow? Is it because (the generic) libgcc does not > trap for MIN_INT / -1 ? Some other reason? > > (I'm just curious; I cannot figure it out :-) ) I don't know if eBPF will be adopting signed division instructions at some point... judging from what can be read in their documentation, most probably they will not. In xBPF we mainly want to avoid the funcall. Linking of BPF objects is still an... eer fuzzy area. Also, we reckon some potential applications of xbpf (like using it instead of dwarf in the Infinity project) may find it simpler to have the instructions than to rely on a software implementation like libgcc's.
>>>> The 'mod' and 'div' operators in eBPF are unsigned, with no signed >>>> counterpart. xBPF adds two new ALU operations, sdiv and smod, for >>>> signed division and modulus, respectively. Update bpf.md with >>>> 'define_insn' blocks for signed div and mod to use them when targetting >>>> xBPF, and add new tests to ensure they are used appropriately. >>>> >>>> 2020-09-17 David Faust <david.faust@oracle.com> >>>> >>>> gcc/ >>>> * config/bpf/bpf.md: Add defines for signed div and mod operators. >>>> >>>> gcc/testsuite/ >>>> * gcc.target/bpf/diag-sdiv.c: New test. >>>> * gcc.target/bpf/diag-smod.c: New test. >>>> * gcc.target/bpf/xbpf-sdiv-1.c: New test. >>>> * gcc.target/bpf/xbpf-smod-1.c: New test. >>> >>> OK. But give Jose 48hrs before committing just in case there's >>> something he wants to comment on. Y'all are far more familiar with bpf >>> than I ;-) >> >> Looks good to me! :) >> But the related pending patch in binutils should go in first. >> > Hi! > > I just checked in the binutils patch, but I don't actually have write > access for gcc. Just did it on your behalf. Thanks!
diff --git a/gcc/config/bpf/bpf.md b/gcc/config/bpf/bpf.md index 41bb4fcd9a7..ac87e3d9c53 100644 --- a/gcc/config/bpf/bpf.md +++ b/gcc/config/bpf/bpf.md @@ -160,6 +160,16 @@ (define_insn "udiv<AM:mode>3" "div<msuffix>\t%0,%2" [(set_attr "type" "<mtype>")]) +;; However, xBPF does provide a signed division operator, sdiv. + +(define_insn "div<AM:mode>3" + [(set (match_operand:AM 0 "register_operand" "=r,r") + (div:AM (match_operand:AM 1 "register_operand" " 0,0") + (match_operand:AM 2 "reg_or_imm_operand" "r,I")))] + "TARGET_XBPF" + "sdiv<msuffix>\t%0,%2" + [(set_attr "type" "<mtype>")]) + ;;; Modulus ;; Note that eBPF doesn't provide instructions for signed integer @@ -173,6 +183,16 @@ (define_insn "umod<AM:mode>3" "mod<msuffix>\t%0,%2" [(set_attr "type" "<mtype>")]) +;; Again, xBPF provides a signed version, smod. + +(define_insn "mod<AM:mode>3" + [(set (match_operand:AM 0 "register_operand" "=r,r") + (mod:AM (match_operand:AM 1 "register_operand" " 0,0") + (match_operand:AM 2 "reg_or_imm_operand" "r,I")))] + "TARGET_XBPF" + "smod<msuffix>\t%0,%2" + [(set_attr "type" "<mtype>")]) + ;;; Logical AND (define_insn "and<AM:mode>3" [(set (match_operand:AM 0 "register_operand" "=r,r") diff --git a/gcc/testsuite/gcc.target/bpf/diag-sdiv.c b/gcc/testsuite/gcc.target/bpf/diag-sdiv.c new file mode 100644 index 00000000000..db0c494a789 --- /dev/null +++ b/gcc/testsuite/gcc.target/bpf/diag-sdiv.c @@ -0,0 +1,12 @@ +/* Verify signed division does not produce 'sdiv' insn in eBPF. */ +/* { dg-do compile } */ +/* { dg-options "-O0" } */ + +void +foo () +{ + signed int x = 5; + signed int y = 2; + signed int z = x / y; +} +/* { dg-final { scan-assembler-not "sdiv(32)?\t%r" } } */ diff --git a/gcc/testsuite/gcc.target/bpf/diag-smod.c b/gcc/testsuite/gcc.target/bpf/diag-smod.c new file mode 100644 index 00000000000..20234ee39cc --- /dev/null +++ b/gcc/testsuite/gcc.target/bpf/diag-smod.c @@ -0,0 +1,12 @@ +/* Verify signed modulo does not produce 'smod' insn in eBPF. */ +/* { dg-do compile } */ +/* { dg-options "-O0" } */ + +void +foo () +{ + signed int x = 5; + signed int y = 2; + signed int z = x % y; +} +/* { dg-final { scan-assembler-not "smod(32)?\t%r" } } */ diff --git a/gcc/testsuite/gcc.target/bpf/xbpf-sdiv-1.c b/gcc/testsuite/gcc.target/bpf/xbpf-sdiv-1.c new file mode 100644 index 00000000000..f6c5c9e9f1c --- /dev/null +++ b/gcc/testsuite/gcc.target/bpf/xbpf-sdiv-1.c @@ -0,0 +1,14 @@ +/* Verify that sdiv instruction is used for xBPF. */ +/* { dg-do compile } */ +/* { dg-options "-O0 -mxbpf" } */ + +void +foo () +{ + signed int x = 5; + signed int y = 2; + signed int z = x / y; + signed int w = x / 3; +} + +/* { dg-final { scan-assembler "sdiv(32)?\t%r" } } */ diff --git a/gcc/testsuite/gcc.target/bpf/xbpf-smod-1.c b/gcc/testsuite/gcc.target/bpf/xbpf-smod-1.c new file mode 100644 index 00000000000..b3e5816b5cf --- /dev/null +++ b/gcc/testsuite/gcc.target/bpf/xbpf-smod-1.c @@ -0,0 +1,14 @@ +/* Verify that smod instruction is used for xBPF. */ +/* { dg-do compile } */ +/* { dg-options "-O0 -mxbpf" } */ + +void +foo () +{ + signed int x = 5; + signed int y = 2; + signed int z = x % y; + signed int w = x % 3; +} + +/* { dg-final { scan-assembler "smod(32)?\t%r" } } */