Alejandro Piñeiro <apinhe...@igalia.com> writes: > Although it is possible to emit them directly as AND/OR on brw_fs_nir, > having a specific opcode makes it easier to remove duplicate settings > later. > > v2: (Curro) > - Set thread control to 'switch' when using the control register > - Use a single SHADER_OPCODE_RND_MODE opcode taking an immediate > with the rounding mode. > - Avoid magic numbers setting rounding mode field at control register. > > Signed-off-by: Alejandro Piñeiro <apinhe...@igalia.com> > Signed-off-by: Jose Maria Casanova Crespo <jmcasan...@igalia.com> > --- > src/intel/compiler/brw_eu.h | 3 +++ > src/intel/compiler/brw_eu_defines.h | 17 +++++++++++++++++ > src/intel/compiler/brw_eu_emit.c | 34 > +++++++++++++++++++++++++++++++++ > src/intel/compiler/brw_fs_generator.cpp | 5 +++++ > src/intel/compiler/brw_shader.cpp | 4 ++++ > 5 files changed, 63 insertions(+) > > diff --git a/src/intel/compiler/brw_eu.h b/src/intel/compiler/brw_eu.h > index 8e597b212a6..106bf03530d 100644 > --- a/src/intel/compiler/brw_eu.h > +++ b/src/intel/compiler/brw_eu.h > @@ -500,6 +500,9 @@ brw_broadcast(struct brw_codegen *p, > struct brw_reg src, > struct brw_reg idx); > > +void > +brw_rounding_mode(struct brw_codegen *p, > + enum brw_rnd_mode mode);
Missing whitespace line.
> /***********************************************************************
> * brw_eu_util.c:
> */
> diff --git a/src/intel/compiler/brw_eu_defines.h
> b/src/intel/compiler/brw_eu_defines.h
> index da482b73c58..91d88fe8952 100644
> --- a/src/intel/compiler/brw_eu_defines.h
> +++ b/src/intel/compiler/brw_eu_defines.h
> @@ -388,6 +388,9 @@ enum opcode {
> SHADER_OPCODE_TYPED_SURFACE_WRITE,
> SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL,
>
> +
Redundant whitespace.
> + SHADER_OPCODE_RND_MODE,
> +
> SHADER_OPCODE_MEMORY_FENCE,
>
> SHADER_OPCODE_GEN4_SCRATCH_READ,
> @@ -1214,4 +1217,18 @@ enum brw_message_target {
> /* R0 */
> # define GEN7_GS_PAYLOAD_INSTANCE_ID_SHIFT 27
>
> +/* CR0.0[5:4] Floating-Point Rounding Modes
> + * Skylake PRM, Volume 7 Part 1, "Control Register", page 756
> + */
> +
> +#define BRW_CR0_RND_MODE_MASK 0x30
> +#define BRW_CR0_RND_MODE_SHIFT 4
> +
> +enum PACKED brw_rnd_mode {
> + BRW_RND_MODE_RTNE = 0, /* Round to Nearest or Even */
> + BRW_RND_MODE_RU = 1, /* Round Up, toward +inf */
> + BRW_RND_MODE_RD = 2, /* Round Down, toward -inf */
> + BRW_RND_MODE_RTZ = 3 /* Round Toward Zero */
> +};
> +
> #endif /* BRW_EU_DEFINES_H */
> diff --git a/src/intel/compiler/brw_eu_emit.c
> b/src/intel/compiler/brw_eu_emit.c
> index 8c952e7da26..12164653e47 100644
> --- a/src/intel/compiler/brw_eu_emit.c
> +++ b/src/intel/compiler/brw_eu_emit.c
> @@ -3530,3 +3530,37 @@ brw_WAIT(struct brw_codegen *p)
> brw_inst_set_exec_size(devinfo, insn, BRW_EXECUTE_1);
> brw_inst_set_mask_control(devinfo, insn, BRW_MASK_DISABLE);
> }
> +
> +/**
> + * Changes the floating point rounding mode updating the control register
> + * field defined at cr0.0[5-6] bits. This function supports the changes to
> + * RTNE (00), RU (01), RD (10) and RTZ (11) rounding using bitwise
> operations.
> + * Only RTNE and RTZ rounding are enabled at nir.
> + */
> +
Redundant whitespace.
> +void
> +brw_rounding_mode(struct brw_codegen *p,
> + enum brw_rnd_mode mode)
> +{
> + const unsigned bits = mode << BRW_CR0_RND_MODE_SHIFT;
> +
> + if (bits != BRW_CR0_RND_MODE_MASK) {
> + brw_inst *inst = brw_AND(p, brw_cr0_reg(0), brw_cr0_reg(0),
> + brw_imm_ud(~BRW_CR0_RND_MODE_MASK));
> +
> + /* From the Skylake PRM, Volume 7, page 760:
> + * "Implementation Restriction on Register Access: When the control
> + * register is used as an explicit source and/or destination,
> hardware
> + * does not ensure execution pipeline coherency. Software must set
> the
> + * thread control field to ‘switch’ for an instruction that uses
> + * control register as an explicit operand."
> + */
> + brw_inst_set_thread_control(p->devinfo, inst, BRW_THREAD_SWITCH);
> + }
> +
> + if (bits) {
> + brw_inst *inst = brw_OR(p, brw_cr0_reg(0), brw_cr0_reg(0),
> + brw_imm_ud(bits));
> + brw_inst_set_thread_control(p->devinfo, inst, BRW_THREAD_SWITCH);
> + }
> +}
> diff --git a/src/intel/compiler/brw_fs_generator.cpp
> b/src/intel/compiler/brw_fs_generator.cpp
> index afaec5c9497..ff9880ebfe8 100644
> --- a/src/intel/compiler/brw_fs_generator.cpp
> +++ b/src/intel/compiler/brw_fs_generator.cpp
> @@ -2144,6 +2144,11 @@ fs_generator::generate_code(const cfg_t *cfg, int
> dispatch_width)
> brw_DIM(p, dst, retype(src[0], BRW_REGISTER_TYPE_F));
> break;
>
> + case SHADER_OPCODE_RND_MODE:
> + assert(src[0].file == BRW_IMMEDIATE_VALUE);
> + brw_rounding_mode(p, (brw_rnd_mode) src[0].d);
> + break;
> +
> default:
> unreachable("Unsupported opcode");
>
> diff --git a/src/intel/compiler/brw_shader.cpp
> b/src/intel/compiler/brw_shader.cpp
> index c62b8ba6140..19dd960be3a 100644
> --- a/src/intel/compiler/brw_shader.cpp
> +++ b/src/intel/compiler/brw_shader.cpp
> @@ -486,6 +486,9 @@ brw_instruction_name(const struct gen_device_info
> *devinfo, enum opcode op)
> return "tes_add_indirect_urb_offset";
> case TES_OPCODE_GET_PRIMITIVE_ID:
> return "tes_get_primitive_id";
> +
> + case SHADER_OPCODE_RND_MODE:
> + return "set_round_mode";
It would make sense for the printed instruction name to match the IR
opcode name as closely as possible, e.g. "rnd_mode". Other than these
minor nit-picks patch looks good:
Reviewed-by: Francisco Jerez <curroje...@riseup.net>
> }
>
> unreachable("not reached");
> @@ -1004,6 +1007,7 @@ backend_instruction::has_side_effects() const
> case SHADER_OPCODE_BARRIER:
> case TCS_OPCODE_URB_WRITE:
> case TCS_OPCODE_RELEASE_INPUT:
> + case SHADER_OPCODE_RND_MODE:
> return true;
> default:
> return false;
> --
> 2.11.0
signature.asc
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
