[Mesa-dev] [PATCH v4 20/44] i965/fs: Add byte scattered read message and fs support
Chema Casanova
jmcasanova at igalia.com
Tue Dec 5 00:50:37 UTC 2017
On 30/11/17 21:45, Jason Ekstrand wrote:
> On Wed, Nov 29, 2017 at 6:50 PM, Jose Maria Casanova Crespo
> <jmcasanova at igalia.com <mailto:jmcasanova at igalia.com>> wrote:
>
> v2: Fix alignment style (Topi Pohjolainen)
> (Jason Ekstrand)
> - Enable bit_size parameter to scattered messages to enable
> different
> bitsizes byte/word/dword.
> - Remove use of brw_send_indirect_scattered_message in favor of
> brw_send_indirect_surface_message.
> - Move scattered messages to surface messages namespace.
> - Assert align1 for scattered messages and assume Gen8+.
> - Inline brw_set_dp_byte_scattered_read.
> ---
> src/intel/compiler/brw_eu.h | 8 +++++++
> src/intel/compiler/brw_eu_defines.h | 2 ++
> src/intel/compiler/brw_eu_emit.c | 30
> ++++++++++++++++++++++++++
> src/intel/compiler/brw_fs.cpp | 19 ++++++++++++++++
> src/intel/compiler/brw_fs_copy_propagation.cpp | 2 ++
> src/intel/compiler/brw_fs_generator.cpp | 6 ++++++
> src/intel/compiler/brw_fs_surface_builder.cpp | 11 +++++++++-
> src/intel/compiler/brw_fs_surface_builder.h | 7 ++++++
> src/intel/compiler/brw_shader.cpp | 6 ++++++
> 9 files changed, 90 insertions(+), 1 deletion(-)
>
> diff --git a/src/intel/compiler/brw_eu.h b/src/intel/compiler/brw_eu.h
> index 3ac3b4342a..2d0f56f793 100644
> --- a/src/intel/compiler/brw_eu.h
> +++ b/src/intel/compiler/brw_eu.h
> @@ -485,6 +485,14 @@ brw_typed_surface_write(struct brw_codegen *p,
> unsigned msg_length,
> unsigned num_channels);
>
> +void
> +brw_byte_scattered_read(struct brw_codegen *p,
> + struct brw_reg dst,
> + struct brw_reg payload,
> + struct brw_reg surface,
> + unsigned msg_length,
> + unsigned bit_size);
> +
> void
> brw_byte_scattered_write(struct brw_codegen *p,
> struct brw_reg payload,
> diff --git a/src/intel/compiler/brw_eu_defines.h
> b/src/intel/compiler/brw_eu_defines.h
> index de6330ee54..aa510ebfa4 100644
> --- a/src/intel/compiler/brw_eu_defines.h
> +++ b/src/intel/compiler/brw_eu_defines.h
> @@ -409,6 +409,8 @@ enum opcode {
> * opcode, but instead of taking a single payload blog they
> expect their
> * arguments separately as individual sources, like untyped
> write/read.
> */
> + SHADER_OPCODE_BYTE_SCATTERED_READ,
> + SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL,
> SHADER_OPCODE_BYTE_SCATTERED_WRITE,
> SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL,
>
> diff --git a/src/intel/compiler/brw_eu_emit.c
> b/src/intel/compiler/brw_eu_emit.c
> index ded7e228cf..bdc516848a 100644
> --- a/src/intel/compiler/brw_eu_emit.c
> +++ b/src/intel/compiler/brw_eu_emit.c
> @@ -2998,6 +2998,36 @@ static enum brw_data_size
> brw_data_size_from_bit_size(unsigned bit_size)
> }
> }
>
> +
> +void
> +brw_byte_scattered_read(struct brw_codegen *p,
> + struct brw_reg dst,
> + struct brw_reg payload,
> + struct brw_reg surface,
> + unsigned msg_length,
> + unsigned bit_size)
> +{
> + assert(brw_inst_access_mode(p->devinfo, p->current) == BRW_ALIGN_1);
> + const struct gen_device_info *devinfo = p->devinfo;
> + const unsigned sfid = GEN7_SFID_DATAPORT_DATA_CACHE;
> +
> + struct brw_inst *insn = brw_send_indirect_surface_message(
> + p, sfid, dst, payload, surface, msg_length,
> + brw_surface_payload_size(p, 1, true, true),
> + false);
> +
> + unsigned msg_control = brw_data_size_from_bit_size(bit_size) << 2;
> +
> + if (brw_inst_exec_size(devinfo, p->current) == BRW_EXECUTE_16)
> + msg_control |= 1; /* SIMD16 mode */
> + else
> + msg_control |= 0; /* SIMD8 mode */
> +
> + brw_inst_set_dp_msg_type(devinfo, insn,
> + HSW_DATAPORT_DC_PORT0_BYTE_SCATTERED_READ);
> + brw_inst_set_dp_msg_control(devinfo, insn, msg_control);
> +}
> +
> void
> brw_byte_scattered_write(struct brw_codegen *p,
> struct brw_reg payload,
> diff --git a/src/intel/compiler/brw_fs.cpp
> b/src/intel/compiler/brw_fs.cpp
> index 32f1d757f0..1ca4d416b2 100644
> --- a/src/intel/compiler/brw_fs.cpp
> +++ b/src/intel/compiler/brw_fs.cpp
> @@ -251,6 +251,7 @@ fs_inst::is_send_from_grf() const
> case SHADER_OPCODE_UNTYPED_SURFACE_READ:
> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
> + case SHADER_OPCODE_BYTE_SCATTERED_READ:
> case SHADER_OPCODE_TYPED_ATOMIC:
> case SHADER_OPCODE_TYPED_SURFACE_READ:
> case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> @@ -750,6 +751,16 @@ fs_inst::components_read(unsigned i) const
> else
> return 1;
>
> + case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
> + assert(src[3].file == IMM &&
> + src[4].file == IMM);
> + if (i == 0)
> + return 1;
> + else if (i == 1)
> + return 0;
>
>
> I suppose src[1] is for the data to write which is unused for this
> opcode? If so, I'd rather we just do
>
> /* src[1] is normally for data to write and is unused */
> return i == 1 ? 0 : 1;
Yes, that is the reason, in this case as it is a read operation, it
isn't using the source at the payload, the only special change with '
other surfaces ops is that we are using src[4] as IMM passing the
bitsize for the scattered read, and dimension should always be 1, so the
source would be always 1 component for write operations.
Would it make sense to include a comment with the explanation of the
params to explain why we use src[4]?
Something like:
/* Scattered logical opcodes use the following params:
* src[0] Surface coordinates
* src[1] Surface operation source (ignored for reads)
* src[2] Surface
* src[3] IMM with always 1 dimension.
* src[4] IMM with arg bitsize for scattered read/write 8, 16, 32 */
With your change, Can I consider this patch as R-b?
> + else
> + return 1;
> +
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
> assert(src[3].file == IMM &&
> src[4].file == IMM);
> @@ -798,6 +809,7 @@ fs_inst::size_read(int arg) const
> case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
> + case SHADER_OPCODE_BYTE_SCATTERED_READ:
> if (arg == 0)
> return mlen * REG_SIZE;
> break;
> @@ -4545,6 +4557,12 @@ fs_visitor::lower_logical_sends()
> ibld.sample_mask_reg());
> break;
>
> + case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
> + lower_surface_logical_send(ibld, inst,
> + SHADER_OPCODE_BYTE_SCATTERED_READ,
> + fs_reg());
> + break;
> +
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
> lower_surface_logical_send(ibld, inst,
> SHADER_OPCODE_BYTE_SCATTERED_WRITE,
> @@ -5036,6 +5054,7 @@ get_lowered_simd_width(const struct
> gen_device_info *devinfo,
> case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
> + case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
> return MIN2(16, inst->exec_size);
>
> case SHADER_OPCODE_URB_READ_SIMD8:
> diff --git a/src/intel/compiler/brw_fs_copy_propagation.cpp
> b/src/intel/compiler/brw_fs_copy_propagation.cpp
> index fcf4706b7a..d4d01d783c 100644
> --- a/src/intel/compiler/brw_fs_copy_propagation.cpp
> +++ b/src/intel/compiler/brw_fs_copy_propagation.cpp
> @@ -656,6 +656,7 @@ fs_visitor::try_constant_propagate(fs_inst
> *inst, acp_entry *entry)
> case SHADER_OPCODE_TYPED_SURFACE_READ:
> case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
> + case SHADER_OPCODE_BYTE_SCATTERED_READ:
> /* We only propagate into the surface argument of the
> * instruction. Everything else goes through LOAD_PAYLOAD.
> */
> @@ -696,6 +697,7 @@ fs_visitor::try_constant_propagate(fs_inst
> *inst, acp_entry *entry)
> case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
> case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
> + case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
> inst->src[i] = val;
> progress = true;
> break;
> diff --git a/src/intel/compiler/brw_fs_generator.cpp
> b/src/intel/compiler/brw_fs_generator.cpp
> index fedc9acf97..a3861cd68e 100644
> --- a/src/intel/compiler/brw_fs_generator.cpp
> +++ b/src/intel/compiler/brw_fs_generator.cpp
> @@ -2073,6 +2073,12 @@ fs_generator::generate_code(const cfg_t *cfg,
> int dispatch_width)
> inst->mlen, src[2].ud);
> break;
>
> + case SHADER_OPCODE_BYTE_SCATTERED_READ:
> + assert(src[2].file == BRW_IMMEDIATE_VALUE);
> + brw_byte_scattered_read(p, dst, src[0], src[1],
> + inst->mlen, src[2].ud);
> + break;
> +
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
> assert(src[2].file == BRW_IMMEDIATE_VALUE);
> brw_byte_scattered_write(p, src[0], src[1],
> diff --git a/src/intel/compiler/brw_fs_surface_builder.cpp
> b/src/intel/compiler/brw_fs_surface_builder.cpp
> index 37cc29e361..c346ef9e70 100644
> --- a/src/intel/compiler/brw_fs_surface_builder.cpp
> +++ b/src/intel/compiler/brw_fs_surface_builder.cpp
> @@ -161,6 +161,16 @@ namespace brw {
> addr, tmp, surface, dims, op, rsize);
> }
>
> + fs_reg
> + emit_byte_scattered_read(const fs_builder &bld,
> + const fs_reg &surface, const fs_reg
> &addr,
> + unsigned dims, unsigned size,
> + unsigned bit_size, brw_predicate pred)
> + {
> + return emit_send(bld,
> SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL,
> + addr, fs_reg(), surface, dims, bit_size,
> size, pred);
> + }
> +
> void
> emit_byte_scattered_write(const fs_builder &bld, const fs_reg
> &surface,
> const fs_reg &addr, const fs_reg &src,
> @@ -1202,4 +1212,3 @@ namespace brw {
> }
> }
> }
> -
> diff --git a/src/intel/compiler/brw_fs_surface_builder.h
> b/src/intel/compiler/brw_fs_surface_builder.h
> index bf9a8c68c8..194d61d489 100644
> --- a/src/intel/compiler/brw_fs_surface_builder.h
> +++ b/src/intel/compiler/brw_fs_surface_builder.h
> @@ -64,6 +64,13 @@ namespace brw {
> unsigned dims, unsigned rsize, unsigned op,
> brw_predicate pred = BRW_PREDICATE_NONE);
>
> + fs_reg
> + emit_byte_scattered_read(const fs_builder &bld,
> + const fs_reg &surface, const fs_reg
> &addr,
> + unsigned dims, unsigned size,
> + unsigned bit_size,
> + brw_predicate pred =
> BRW_PREDICATE_NONE);
> +
> void
> emit_byte_scattered_write(const fs_builder &bld, const fs_reg
> &surface,
> const fs_reg &addr, const fs_reg &src,
> diff --git a/src/intel/compiler/brw_shader.cpp
> b/src/intel/compiler/brw_shader.cpp
> index 209552e1b2..74b52976d7 100644
> --- a/src/intel/compiler/brw_shader.cpp
> +++ b/src/intel/compiler/brw_shader.cpp
> @@ -293,6 +293,10 @@ brw_instruction_name(const struct
> gen_device_info *devinfo, enum opcode op)
> case SHADER_OPCODE_MEMORY_FENCE:
> return "memory_fence";
>
> + case SHADER_OPCODE_BYTE_SCATTERED_READ:
> + return "byte_scattered_read";
> + case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
> + return "byte_scattered_read_logical";
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
> return "byte_scattered_write";
> case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
> @@ -999,6 +1003,8 @@ backend_instruction::is_volatile() const
> case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
> case SHADER_OPCODE_TYPED_SURFACE_READ:
> case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
> + case SHADER_OPCODE_BYTE_SCATTERED_READ:
> + case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
> case SHADER_OPCODE_URB_READ_SIMD8:
> case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
> case VEC4_OPCODE_URB_READ:
> --
> 2.14.3
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org <mailto:mesa-dev at lists.freedesktop.org>
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> <https://lists.freedesktop.org/mailman/listinfo/mesa-dev>
>
>
>
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
More information about the mesa-dev
mailing list