[Mesa-dev] [PATCH 2/2] tgsi/lowering: add support to lower CEIL

[Roland Scheidegger]

Am 13.04.2016 um 22:51 schrieb Christian Gmeiner:
> Signed-off-by: Christian Gmeiner <christian.gmeiner at gmail.com>
> ---
>  src/gallium/auxiliary/tgsi/tgsi_lowering.c | 54 ++++++++++++++++++++++++++++++
>  src/gallium/auxiliary/tgsi/tgsi_lowering.h |  1 +
>  2 files changed, 55 insertions(+)
> 
> diff --git a/src/gallium/auxiliary/tgsi/tgsi_lowering.c b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
> index cd5bdc2..54536d1 100644
> --- a/src/gallium/auxiliary/tgsi/tgsi_lowering.c
> +++ b/src/gallium/auxiliary/tgsi/tgsi_lowering.c
> @@ -302,6 +302,50 @@ transform_flr(struct tgsi_transform_context *tctx,
>      }
>  }
>  
> +/* CEIL - Ceiling
> + *
> + *   dst.x = \lceil src.x\rceil
> + *   dst.y = \lceil src.y\rceil
> + *   dst.z = \lceil src.z\rceil
> + *   dst.w = \lceil src.w\rceil
> + *
> + * ; needs: 1 tmp
> + * FRC tmpA, src
> + * ADD dst, src, tmpA

The math here really doesn't add up. Might have more luck when throwing
in a negation first (FRC tmpA, -src).

Roland


> + */
> +#define CEIL_GROW (NINST(1) + NINST(2) - OINST(1))
> +#define CEIL_TMP  1
> +static void
> +transform_ceil(struct tgsi_transform_context *tctx,
> +              struct tgsi_full_instruction *inst)
> +{
> +    struct tgsi_lowering_context *ctx = tgsi_lowering_context(tctx);
> +    struct tgsi_full_dst_register *dst = &inst->Dst[0];
> +    struct tgsi_full_src_register *src = &inst->Src[0];
> +    struct tgsi_full_instruction new_inst;
> +
> +    if (dst->Register.WriteMask & TGSI_WRITEMASK_XYZW) {
> +       /* FRC tmpA, src */
> +       new_inst = tgsi_default_full_instruction();
> +       new_inst.Instruction.Opcode = TGSI_OPCODE_FRC;
> +       new_inst.Instruction.NumDstRegs = 1;
> +       reg_dst(&new_inst.Dst[0], &ctx->tmp[A].dst, TGSI_WRITEMASK_XYZW);
> +       new_inst.Instruction.NumSrcRegs = 1;
> +       reg_src(&new_inst.Src[0], src, SWIZ(X, Y, Z, W));
> +       tctx->emit_instruction(tctx, &new_inst);
> +
> +       /* ADD dst, src, tmpA */
> +       new_inst = tgsi_default_full_instruction();
> +       new_inst.Instruction.Opcode = TGSI_OPCODE_ADD;
> +       new_inst.Instruction.NumDstRegs = 1;
> +       reg_dst(&new_inst.Dst[0], dst, TGSI_WRITEMASK_XYZW);
> +       new_inst.Instruction.NumSrcRegs = 2;
> +       reg_src(&new_inst.Src[0], src, SWIZ(X, Y, Z, W));
> +       reg_src(&new_inst.Src[1], &ctx->tmp[A].src, SWIZ(X, Y, Z, W));
> +       tctx->emit_instruction(tctx, &new_inst);
> +    }
> +}
> +
>  /* XPD - Cross Product
>   *   dst.x = src0.y \times src1.z - src1.y \times src0.z
>   *   dst.y = src0.z \times src1.x - src1.z \times src0.x
> @@ -1385,6 +1429,11 @@ transform_instr(struct tgsi_transform_context *tctx,
>             goto skip;
>         transform_flr(tctx, inst);
>         break;
> +   case TGSI_OPCODE_CEIL:
> +       if (!ctx->config->lower_CEIL)
> +           goto skip;
> +       transform_ceil(tctx, inst);
> +       break;
>     case TGSI_OPCODE_XPD:
>        if (!ctx->config->lower_XPD)
>           goto skip;
> @@ -1510,6 +1559,7 @@ tgsi_transform_lowering(const struct tgsi_lowering_config *config,
>     /* if there are no instructions to lower, then we are done: */
>     if (!(OPCS(DST) ||
>           OPCS(FLR) ||
> +         OPCS(CEIL) ||
>           OPCS(XPD) ||
>           OPCS(SCS) ||
>           OPCS(LRP) ||
> @@ -1543,6 +1593,10 @@ tgsi_transform_lowering(const struct tgsi_lowering_config *config,
>        newlen += FLR_GROW * OPCS(FLR);
>        numtmp = MAX2(numtmp, FLR_TMP);
>     }
> +   if (OPCS(CEIL)) {
> +      newlen += CEIL_GROW * OPCS(CEIL);
> +      numtmp = MAX2(numtmp, CEIL_TMP);
> +   }
>     if (OPCS(XPD)) {
>        newlen += XPD_GROW * OPCS(XPD);
>        numtmp = MAX2(numtmp, XPD_TMP);
> diff --git a/src/gallium/auxiliary/tgsi/tgsi_lowering.h b/src/gallium/auxiliary/tgsi/tgsi_lowering.h
> index 803f84e..f2d02a4 100644
> --- a/src/gallium/auxiliary/tgsi/tgsi_lowering.h
> +++ b/src/gallium/auxiliary/tgsi/tgsi_lowering.h
> @@ -56,6 +56,7 @@ struct tgsi_lowering_config
>      */
>     unsigned lower_DST:1;
>     unsigned lower_FLR:1;
> +   unsigned lower_CEIL:1;
>     unsigned lower_XPD:1;
>     unsigned lower_SCS:1;
>     unsigned lower_LRP:1;
>