[Mesa-dev] [PATCH 12/13] i965: Add typed surface access opcodes.
Pohjolainen, Topi
topi.pohjolainen at intel.com
Mon May 4 01:22:23 PDT 2015
On Sat, Apr 25, 2015 at 01:45:36AM +0300, Francisco Jerez wrote:
> "Pohjolainen, Topi" <topi.pohjolainen at intel.com> writes:
>
> > On Fri, Feb 27, 2015 at 05:34:55PM +0200, Francisco Jerez wrote:
> >> ---
> >> src/mesa/drivers/dri/i965/brw_defines.h | 4 +
> >> src/mesa/drivers/dri/i965/brw_eu.h | 24 +++
> >> src/mesa/drivers/dri/i965/brw_eu_emit.c | 169 +++++++++++++++++++++
> >> src/mesa/drivers/dri/i965/brw_fs.cpp | 12 ++
> >> src/mesa/drivers/dri/i965/brw_fs_generator.cpp | 17 +++
> >> .../drivers/dri/i965/brw_schedule_instructions.cpp | 3 +
> >> src/mesa/drivers/dri/i965/brw_shader.cpp | 8 +
> >> src/mesa/drivers/dri/i965/brw_vec4.cpp | 6 +
> >> src/mesa/drivers/dri/i965/brw_vec4_generator.cpp | 18 +++
> >> 9 files changed, 261 insertions(+)
> >>
> >> diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
> >> index e56f49c..cf07da9 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_defines.h
> >> +++ b/src/mesa/drivers/dri/i965/brw_defines.h
> >> @@ -906,6 +906,10 @@ enum opcode {
> >> SHADER_OPCODE_UNTYPED_SURFACE_READ,
> >> SHADER_OPCODE_UNTYPED_SURFACE_WRITE,
> >>
> >> + SHADER_OPCODE_TYPED_ATOMIC,
> >> + SHADER_OPCODE_TYPED_SURFACE_READ,
> >> + SHADER_OPCODE_TYPED_SURFACE_WRITE,
> >> +
> >> SHADER_OPCODE_GEN4_SCRATCH_READ,
> >> SHADER_OPCODE_GEN4_SCRATCH_WRITE,
> >> SHADER_OPCODE_GEN7_SCRATCH_READ,
> >> diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h
> >> index cad956b..ce9554b 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_eu.h
> >> +++ b/src/mesa/drivers/dri/i965/brw_eu.h
> >> @@ -421,6 +421,30 @@ brw_untyped_surface_write(struct brw_compile *p,
> >> unsigned num_channels);
> >>
> >> void
> >> +brw_typed_atomic(struct brw_compile *p,
> >> + struct brw_reg dst,
> >> + struct brw_reg payload,
> >> + struct brw_reg surface,
> >> + unsigned atomic_op,
> >> + unsigned msg_length,
> >> + bool response_expected);
> >> +
> >> +void
> >> +brw_typed_surface_read(struct brw_compile *p,
> >> + struct brw_reg dst,
> >> + struct brw_reg payload,
> >> + struct brw_reg surface,
> >> + unsigned msg_length,
> >> + unsigned num_channels);
> >> +
> >> +void
> >> +brw_typed_surface_write(struct brw_compile *p,
> >> + struct brw_reg payload,
> >> + struct brw_reg surface,
> >> + unsigned msg_length,
> >> + unsigned num_channels);
> >> +
> >> +void
> >> brw_pixel_interpolator_query(struct brw_compile *p,
> >> struct brw_reg dest,
> >> struct brw_reg mrf,
> >> diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
> >> index f5b8fa9..74f1fc1 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
> >> +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
> >> @@ -2944,6 +2944,175 @@ brw_untyped_surface_write(struct brw_compile *p,
> >> p, insn, num_channels);
> >> }
> >>
> >> +static void
> >> +brw_set_dp_typed_atomic_message(struct brw_compile *p,
> >> + struct brw_inst *insn,
> >> + unsigned atomic_op,
> >> + bool response_expected)
> >> +{
> >> + const struct brw_context *brw = p->brw;
> >> + unsigned msg_control =
> >> + atomic_op | /* Atomic Operation Type: BRW_AOP_* */
> >> + (response_expected ? 1 << 5 : 0); /* Return data expected */
> >> +
> >> + if (brw->gen >= 8 || brw->is_haswell) {
> >> + if (brw_inst_access_mode(brw, p->current) == BRW_ALIGN_1) {
> >> + if (brw_inst_qtr_control(brw, p->current) == GEN6_COMPRESSION_2Q)
> >> + msg_control |= 1 << 4; /* Use high 8 slots of the sample mask */
> >> +
> >> + brw_inst_set_dp_msg_type(brw, insn,
> >> + HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP);
> >> + } else {
> >> + brw_inst_set_dp_msg_type(brw, insn,
> >> + HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP_SIMD4X2);
> >> + }
> >> +
> >> + } else {
> >> + brw_inst_set_dp_msg_type(brw, insn,
> >> + GEN7_DATAPORT_RC_TYPED_ATOMIC_OP);
> >> +
> >> + if (brw_inst_qtr_control(brw, p->current) == GEN6_COMPRESSION_2Q)
> >> + msg_control |= 1 << 4; /* Use high 8 slots of the sample mask */
> >> + }
> >> +
> >> + brw_inst_set_dp_msg_control(brw, insn, msg_control);
> >> +}
> >> +
> >> +void
> >> +brw_typed_atomic(struct brw_compile *p,
> >> + struct brw_reg dst,
> >> + struct brw_reg payload,
> >> + struct brw_reg surface,
> >> + unsigned atomic_op,
> >> + unsigned msg_length,
> >> + bool response_expected) {
> >> + const struct brw_context *brw = p->brw;
> >> + const unsigned sfid = (brw->gen >= 8 || brw->is_haswell ?
> >> + HSW_SFID_DATAPORT_DATA_CACHE_1 :
> >> + GEN6_SFID_DATAPORT_RENDER_CACHE);
> >> + const bool align1 = (brw_inst_access_mode(brw, p->current) == BRW_ALIGN_1);
> >> + /* Mask out unused components -- See comment in brw_untyped_atomic(). */
> >> + const unsigned mask = (align1 ? WRITEMASK_XYZW : WRITEMASK_X);
> >> + struct brw_inst *insn = brw_send_indirect_surface_message(
> >> + p, sfid, brw_writemask(dst, mask), payload, surface, msg_length,
> >> + brw_surface_payload_size(p, response_expected,
> >> + brw->gen >= 8 || brw->is_haswell, false),
> >> + true);
Did you mean to use the otherwise unneeded "align1" here instead of the
hardcoded "true"? This should match the behaviour of brw_untyped_atomic() in
this regard, right?
> >> +
> >> + brw_set_dp_typed_atomic_message(
> >> + p, insn, atomic_op, response_expected);
> >> +}
> >> +
> >> +static void
> >> +brw_set_dp_typed_surface_read_message(struct brw_compile *p,
> >> + struct brw_inst *insn,
> >> + unsigned num_channels)
> >> +{
> >> + const struct brw_context *brw = p->brw;
> >> + /* Set mask of unused channels. */
> >> + unsigned msg_control = (0xf & (0xf << num_channels));
> >> +
> >> + if (brw->gen >= 8 || brw->is_haswell) {
> >> + if (brw_inst_access_mode(brw, p->current) == BRW_ALIGN_1) {
> >> + if (brw_inst_qtr_control(brw, p->current) == GEN6_COMPRESSION_2Q)
> >> + msg_control |= 2 << 4; /* Use high 8 slots of the sample mask */
> >> + else
> >> + msg_control |= 1 << 4; /* Use low 8 slots of the sample mask */
> >> + }
> >> +
> >> + brw_inst_set_dp_msg_type(brw, insn,
> >> + HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_READ);
> >> + } else {
> >> + if (brw_inst_access_mode(brw, p->current) == BRW_ALIGN_1) {
> >> + if (brw_inst_qtr_control(brw, p->current) == GEN6_COMPRESSION_2Q)
> >> + msg_control |= 1 << 5; /* Use high 8 slots of the sample mask */
> >> + }
> >> +
> >> + brw_inst_set_dp_msg_type(brw, insn,
> >> + GEN7_DATAPORT_RC_TYPED_SURFACE_READ);
> >> + }
> >> +
> >> + brw_inst_set_dp_msg_control(brw, insn, msg_control);
> >> +}
> >> +
> >> +void
> >> +brw_typed_surface_read(struct brw_compile *p,
> >> + struct brw_reg dst,
> >> + struct brw_reg payload,
> >> + struct brw_reg surface,
> >> + unsigned msg_length,
> >> + unsigned num_channels)
> >> +{
> >> + const struct brw_context *brw = p->brw;
> >> + const unsigned sfid = (brw->gen >= 8 || brw->is_haswell ?
> >> + HSW_SFID_DATAPORT_DATA_CACHE_1 :
> >> + GEN6_SFID_DATAPORT_RENDER_CACHE);
> >> + struct brw_inst *insn = brw_send_indirect_surface_message(
> >
> > This call doesn't exist yet in upstream but I understood from you that
> > everything else up to here in this series is upstreamed. Am I missing
> > something?
> >
>
> No, it's not upstream because it's defined in PATCH 5 you haven't sent
> your R-b for AFAIK. Other patches up to this point are reviewed indeed.
Right. This looks pretty consistent with untyped. I checked the message
control bits against IVB PRM and bspec, and they seem to be fine as well.
>
> >> + p, sfid, dst, payload, surface, msg_length,
> >> + brw_surface_payload_size(p, num_channels,
> >> + brw->gen >= 8 || brw->is_haswell, false),
> >> + true);
Here the same question about considering align1 mode? In
brw_untyped_surface_read() we check the access mode and pass it to
brw_surface_payload_size().
> >> +
> >> + brw_set_dp_typed_surface_read_message(
> >> + p, insn, num_channels);
> >> +}
> >> +
> >> +static void
> >> +brw_set_dp_typed_surface_write_message(struct brw_compile *p,
> >> + struct brw_inst *insn,
> >> + unsigned num_channels)
> >> +{
> >> + const struct brw_context *brw = p->brw;
> >> + /* Set mask of unused channels. */
> >> + unsigned msg_control = (0xf & (0xf << num_channels));
> >> +
> >> + if (brw->gen >= 8 || brw->is_haswell) {
> >> + if (brw_inst_access_mode(brw, p->current) == BRW_ALIGN_1) {
> >> + if (brw_inst_qtr_control(brw, p->current) == GEN6_COMPRESSION_2Q)
> >> + msg_control |= 2 << 4; /* Use high 8 slots of the sample mask */
> >> + else
> >> + msg_control |= 1 << 4; /* Use low 8 slots of the sample mask */
> >> + }
> >> +
> >> + brw_inst_set_dp_msg_type(brw, insn,
> >> + HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_WRITE);
> >> +
> >> + } else {
> >> + if (brw_inst_access_mode(brw, p->current) == BRW_ALIGN_1) {
> >> + if (brw_inst_qtr_control(brw, p->current) == GEN6_COMPRESSION_2Q)
> >> + msg_control |= 1 << 5; /* Use high 8 slots of the sample mask */
> >> + }
> >> +
> >> + brw_inst_set_dp_msg_type(brw, insn,
> >> + GEN7_DATAPORT_RC_TYPED_SURFACE_WRITE);
> >> + }
> >> +
> >> + brw_inst_set_dp_msg_control(brw, insn, msg_control);
> >> +}
> >> +
> >> +void
> >> +brw_typed_surface_write(struct brw_compile *p,
> >> + struct brw_reg payload,
> >> + struct brw_reg surface,
> >> + unsigned msg_length,
> >> + unsigned num_channels)
> >> +{
> >> + const struct brw_context *brw = p->brw;
> >> + const unsigned sfid = (brw->gen >= 8 || brw->is_haswell ?
> >> + HSW_SFID_DATAPORT_DATA_CACHE_1 :
> >> + GEN6_SFID_DATAPORT_RENDER_CACHE);
> >> + const bool align1 = (brw_inst_access_mode(brw, p->current) == BRW_ALIGN_1);
> >> + /* Mask out unused components -- See comment in brw_untyped_atomic(). */
> >> + const unsigned mask = (brw->gen == 7 && !brw->is_haswell && !align1 ?
> >> + WRITEMASK_X : WRITEMASK_XYZW);
> >> + struct brw_inst *insn = brw_send_indirect_surface_message(
> >> + p, sfid, brw_writemask(brw_null_reg(), mask),
> >> + payload, surface, msg_length, 0, true);
> >> +
> >> + brw_set_dp_typed_surface_write_message(
> >> + p, insn, num_channels);
> >> +}
> >> +
> >> void
> >> brw_pixel_interpolator_query(struct brw_compile *p,
> >> struct brw_reg dest,
> >> diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
> >> index 6d363f1..191d787 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_fs.cpp
> >> +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
> >> @@ -509,6 +509,9 @@ fs_inst::is_send_from_grf() const
> >> case SHADER_OPCODE_UNTYPED_ATOMIC:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_READ:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> + case SHADER_OPCODE_TYPED_SURFACE_READ:
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> case SHADER_OPCODE_URB_WRITE_SIMD8:
> >> return true;
> >> case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
> >> @@ -933,6 +936,12 @@ fs_inst::regs_read(int arg) const
> >> return mlen;
> >> } else if (opcode == SHADER_OPCODE_UNTYPED_SURFACE_WRITE && arg == 0) {
> >> return mlen;
> >> + } else if (opcode == SHADER_OPCODE_TYPED_ATOMIC && arg == 0) {
> >> + return mlen;
> >> + } else if (opcode == SHADER_OPCODE_TYPED_SURFACE_READ && arg == 0) {
> >> + return mlen;
> >> + } else if (opcode == SHADER_OPCODE_TYPED_SURFACE_WRITE && arg == 0) {
> >> + return mlen;
> >> } else if (opcode == FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET && arg == 0) {
> >> return mlen;
> >> }
> >> @@ -1024,6 +1033,9 @@ fs_visitor::implied_mrf_writes(fs_inst *inst)
> >> case SHADER_OPCODE_UNTYPED_ATOMIC:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_READ:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> + case SHADER_OPCODE_TYPED_SURFACE_READ:
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> case SHADER_OPCODE_URB_WRITE_SIMD8:
> >> case FS_OPCODE_INTERPOLATE_AT_CENTROID:
> >> case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
> >> diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> >> index 6e093cf..e80866b 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> >> +++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> >> @@ -1905,6 +1905,23 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width)
> >> inst->mlen, src[2].dw1.ud);
> >> break;
> >>
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> + assert(src[2].file == BRW_IMMEDIATE_VALUE);
> >> + brw_typed_atomic(p, dst, src[0], src[1],
> >> + src[2].dw1.ud, inst->mlen, !inst->dst.is_null());
> >> + break;
> >> +
> >> + case SHADER_OPCODE_TYPED_SURFACE_READ:
> >> + assert(src[2].file == BRW_IMMEDIATE_VALUE);
> >> + brw_typed_surface_read(p, dst, src[0], src[1],
> >> + inst->mlen, src[2].dw1.ud);
> >> + break;
> >> +
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> + assert(src[2].file == BRW_IMMEDIATE_VALUE);
> >> + brw_typed_surface_write(p, src[0], src[1], inst->mlen, src[2].dw1.ud);
> >> + break;
> >> +
> >> case SHADER_OPCODE_FIND_LIVE_CHANNEL:
> >> brw_find_live_channel(p, dst);
> >> break;
> >> diff --git a/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp b/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> >> index 46b635c..808d839 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> >> +++ b/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> >> @@ -340,6 +340,7 @@ schedule_node::set_latency_gen7(bool is_haswell)
> >> break;
> >>
> >> case SHADER_OPCODE_UNTYPED_ATOMIC:
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> /* Test code:
> >> * mov(8) g112<1>ud 0x00000000ud { align1 WE_all 1Q };
> >> * mov(1) g112.7<1>ud g1.7<0,1,0>ud { align1 WE_all };
> >> @@ -359,6 +360,8 @@ schedule_node::set_latency_gen7(bool is_haswell)
> >>
> >> case SHADER_OPCODE_UNTYPED_SURFACE_READ:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
> >> + case SHADER_OPCODE_TYPED_SURFACE_READ:
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> /* Test code:
> >> * mov(8) g112<1>UD 0x00000000UD { align1 WE_all 1Q };
> >> * mov(1) g112.7<1>UD g1.7<0,1,0>UD { align1 WE_all };
> >> diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
> >> index b58796d..647355d 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_shader.cpp
> >> +++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
> >> @@ -460,6 +460,12 @@ brw_instruction_name(enum opcode op)
> >> return "untyped_surface_read";
> >> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
> >> return "untyped_surface_write";
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> + return "typed_atomic";
> >> + case SHADER_OPCODE_TYPED_SURFACE_READ:
> >> + return "typed_surface_read";
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> + return "typed_surface_write";
> >>
> >> case SHADER_OPCODE_LOAD_PAYLOAD:
> >> return "load_payload";
> >> @@ -969,6 +975,8 @@ backend_instruction::has_side_effects() const
> >> switch (opcode) {
> >> case SHADER_OPCODE_UNTYPED_ATOMIC:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> case SHADER_OPCODE_URB_WRITE_SIMD8:
> >> case FS_OPCODE_FB_WRITE:
> >> return true;
> >> diff --git a/src/mesa/drivers/dri/i965/brw_vec4.cpp b/src/mesa/drivers/dri/i965/brw_vec4.cpp
> >> index 736437f..174257b 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_vec4.cpp
> >> +++ b/src/mesa/drivers/dri/i965/brw_vec4.cpp
> >> @@ -259,6 +259,9 @@ vec4_instruction::is_send_from_grf()
> >> case SHADER_OPCODE_UNTYPED_ATOMIC:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_READ:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> + case SHADER_OPCODE_TYPED_SURFACE_READ:
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> return true;
> >> default:
> >> return false;
> >> @@ -276,6 +279,9 @@ vec4_instruction::regs_read(unsigned arg) const
> >> case SHADER_OPCODE_UNTYPED_ATOMIC:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_READ:
> >> case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> + case SHADER_OPCODE_TYPED_SURFACE_READ:
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> return arg == 0 ? mlen : 1;
> >>
> >> case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
> >> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
> >> index 4c75b95..943c92a 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
> >> +++ b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
> >> @@ -1480,6 +1480,24 @@ vec4_generator::generate_code(const cfg_t *cfg)
> >> src[2].dw1.ud);
> >> break;
> >>
> >> + case SHADER_OPCODE_TYPED_ATOMIC:
> >> + assert(src[2].file == BRW_IMMEDIATE_VALUE);
> >> + brw_typed_atomic(p, dst, src[0], src[1], src[2].dw1.ud, inst->mlen,
> >> + !inst->dst.is_null());
> >> + break;
> >> +
> >> + case SHADER_OPCODE_TYPED_SURFACE_READ:
> >> + assert(src[2].file == BRW_IMMEDIATE_VALUE);
> >> + brw_typed_surface_read(p, dst, src[0], src[1], inst->mlen,
> >> + src[2].dw1.ud);
> >> + break;
> >> +
> >> + case SHADER_OPCODE_TYPED_SURFACE_WRITE:
> >> + assert(src[2].file == BRW_IMMEDIATE_VALUE);
> >> + brw_typed_surface_write(p, src[0], src[1], inst->mlen,
> >> + src[2].dw1.ud);
> >> + break;
> >> +
> >> case SHADER_OPCODE_FIND_LIVE_CHANNEL:
> >> brw_find_live_channel(p, dst);
> >> break;
> >> --
> >> 2.1.3
> >>
> >> _______________________________________________
> >> mesa-dev mailing list
> >> mesa-dev at lists.freedesktop.org
> >> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list