[Mesa-dev] [PATCH] gk110/ir: do not overwrite def value with zero for EXCH ops

[Ilia Mirkin]

The SM20/SM30 logic does this for exch || cas. Are you *sure* this
shouldn't be the same? It's pretty silly to do a CAS without a dst,
but it's definitely possible (through DCE).

On Wed, Apr 20, 2016 at 1:47 PM, Samuel Pitoiset
<samuel.pitoiset at gmail.com> wrote:
> This is only valid for other atomic operations (including CAS). This
> fixes an invalid opcode error from dmesg. While we are it, make sure
> to initialize global addr to 0 for other atomic operations.
>
> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> Cc: "11.1 11.2" <mesa-stable at lists.freedesktop.org>
> ---
>  .../drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp  | 21 +++++++++++++++------
>  1 file changed, 15 insertions(+), 6 deletions(-)
>
> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
> index 70f3c3f..e2c3b8e 100644
> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
> @@ -1808,6 +1808,9 @@ uses64bitAddress(const Instruction *ldst)
>  void
>  CodeEmitterGK110::emitATOM(const Instruction *i)
>  {
> +   const bool hasDst = i->defExists(0);
> +   const bool exch = i->subOp == NV50_IR_SUBOP_ATOM_EXCH;
> +
>     code[0] = 0x00000002;
>     if (i->subOp == NV50_IR_SUBOP_ATOM_CAS)
>        code[1] = 0x77800000;
> @@ -1836,15 +1839,21 @@ CodeEmitterGK110::emitATOM(const Instruction *i)
>     /* TODO: cas: flip bits if $r255 is used */
>     srcId(i->src(1), 23);
>
> -   if (i->defExists(0))
> +   if (hasDst) {
>        defId(i->def(0), 2);
> -   else
> +   } else
> +   if (!exch) {
>        code[0] |= 255 << 2;
> +   }
>
> -   const int32_t offset = SDATA(i->src(0)).offset;
> -   assert(offset < 0x80000 && offset >= -0x80000);
> -   code[0] |= (offset & 1) << 31;
> -   code[1] |= (offset & 0xffffe) >> 1;
> +   if (hasDst || !exch) {
> +      const int32_t offset = SDATA(i->src(0)).offset;
> +      assert(offset < 0x80000 && offset >= -0x80000);
> +      code[0] |= (offset & 1) << 31;
> +      code[1] |= (offset & 0xffffe) >> 1;
> +   } else {
> +      srcAddr32(i->src(0), 31);
> +   }
>
>     if (i->getIndirect(0, 0)) {
>        srcId(i->getIndirect(0, 0), 10);
> --
> 2.8.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev