[Mesa-dev] [PATCH 04/12] i965/fs/skl+: Use lcd2dms_w instead of lcd2dms
Neil Roberts
neil at linux.intel.com
Thu Sep 17 09:00:06 PDT 2015
In order to support 16x MSAA, skl+ has a wider version of lcd2dms that
takes two parameters for the MCS data. This patch makes it allocate a
register that is twice as big for the MCS data and then always use
the wider version.
---
src/mesa/drivers/dri/i965/brw_defines.h | 4 ++++
src/mesa/drivers/dri/i965/brw_disasm.c | 1 +
src/mesa/drivers/dri/i965/brw_fs.cpp | 27 ++++++++++++++++++++++++--
src/mesa/drivers/dri/i965/brw_fs_generator.cpp | 5 +++++
src/mesa/drivers/dri/i965/brw_fs_visitor.cpp | 9 ++++++---
src/mesa/drivers/dri/i965/brw_shader.cpp | 5 +++++
6 files changed, 46 insertions(+), 5 deletions(-)
diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
index 8fc8ceb..2d5b67d 100644
--- a/src/mesa/drivers/dri/i965/brw_defines.h
+++ b/src/mesa/drivers/dri/i965/brw_defines.h
@@ -969,6 +969,8 @@ enum opcode {
FS_OPCODE_TXB_LOGICAL,
SHADER_OPCODE_TXF_CMS,
SHADER_OPCODE_TXF_CMS_LOGICAL,
+ SHADER_OPCODE_TXF_CMS_W,
+ SHADER_OPCODE_TXF_CMS_W_LOGICAL,
SHADER_OPCODE_TXF_UMS,
SHADER_OPCODE_TXF_UMS_LOGICAL,
SHADER_OPCODE_TXF_MCS,
@@ -1517,10 +1519,12 @@ enum brw_message_target {
#define GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO 17
#define GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO_C 18
#define HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE 20
+#define GEN9_SAMPLER_MESSAGE_SAMPLE_LD2DMS_W 28
#define GEN7_SAMPLER_MESSAGE_SAMPLE_LD_MCS 29
#define GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS 30
#define GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DSS 31
+
/* for GEN5 only */
#define BRW_SAMPLER_SIMD_MODE_SIMD4X2 0
#define BRW_SAMPLER_SIMD_MODE_SIMD8 1
diff --git a/src/mesa/drivers/dri/i965/brw_disasm.c b/src/mesa/drivers/dri/i965/brw_disasm.c
index db23a18..5aca728 100644
--- a/src/mesa/drivers/dri/i965/brw_disasm.c
+++ b/src/mesa/drivers/dri/i965/brw_disasm.c
@@ -622,6 +622,7 @@ static const char *const gen5_sampler_msg_type[] = {
[GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO] = "gather4_po",
[GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO_C] = "gather4_po_c",
[HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE] = "sample_d_c",
+ [GEN9_SAMPLER_MESSAGE_SAMPLE_LD2DMS_W] = "ld2dms_w",
[GEN7_SAMPLER_MESSAGE_SAMPLE_LD_MCS] = "ld_mcs",
[GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS] = "ld2dms",
[GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DSS] = "ld2dss",
diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index 32a2adf..82f49b4 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -697,6 +697,7 @@ fs_inst::components_read(unsigned i) const
case SHADER_OPCODE_TXS_LOGICAL:
case FS_OPCODE_TXB_LOGICAL:
case SHADER_OPCODE_TXF_CMS_LOGICAL:
+ case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
case SHADER_OPCODE_TXF_UMS_LOGICAL:
case SHADER_OPCODE_TXF_MCS_LOGICAL:
case SHADER_OPCODE_LOD_LOGICAL:
@@ -712,6 +713,9 @@ fs_inst::components_read(unsigned i) const
/* Texture offset. */
else if (i == 7)
return 2;
+ /* MCS */
+ else if (i == 5 && opcode == SHADER_OPCODE_TXF_CMS_W_LOGICAL)
+ return 2;
else
return 1;
@@ -873,6 +877,7 @@ fs_visitor::implied_mrf_writes(fs_inst *inst)
case SHADER_OPCODE_TXD:
case SHADER_OPCODE_TXF:
case SHADER_OPCODE_TXF_CMS:
+ case SHADER_OPCODE_TXF_CMS_W:
case SHADER_OPCODE_TXF_MCS:
case SHADER_OPCODE_TG4:
case SHADER_OPCODE_TG4_OFFSET:
@@ -3816,17 +3821,31 @@ lower_sampler_logical_send_gen7(const fs_builder &bld, fs_inst *inst, opcode op,
coordinate_done = true;
break;
case SHADER_OPCODE_TXF_CMS:
+ case SHADER_OPCODE_TXF_CMS_W:
case SHADER_OPCODE_TXF_UMS:
case SHADER_OPCODE_TXF_MCS:
- if (op == SHADER_OPCODE_TXF_UMS || op == SHADER_OPCODE_TXF_CMS) {
+ if (op == SHADER_OPCODE_TXF_UMS ||
+ op == SHADER_OPCODE_TXF_CMS ||
+ op == SHADER_OPCODE_TXF_CMS_W) {
bld.MOV(retype(sources[length], BRW_REGISTER_TYPE_UD), sample_index);
length++;
}
- if (op == SHADER_OPCODE_TXF_CMS) {
+ if (op == SHADER_OPCODE_TXF_CMS || op == SHADER_OPCODE_TXF_CMS_W) {
/* Data from the multisample control surface. */
bld.MOV(retype(sources[length], BRW_REGISTER_TYPE_UD), mcs);
length++;
+
+ /* On Gen9+ we'll use lcd2ms_w instead which has two registers for
+ * the MCS data.
+ */
+ if (op == SHADER_OPCODE_TXF_CMS_W) {
+ bld.MOV(retype(sources[length], BRW_REGISTER_TYPE_UD),
+ mcs.file == IMM ?
+ mcs :
+ offset(mcs, bld, 1));
+ length++;
+ }
}
/* There is no offsetting for this message; just copy in the integer
@@ -4040,6 +4059,10 @@ fs_visitor::lower_logical_sends()
lower_sampler_logical_send(ibld, inst, SHADER_OPCODE_TXF_CMS);
break;
+ case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
+ lower_sampler_logical_send(ibld, inst, SHADER_OPCODE_TXF_CMS_W);
+ break;
+
case SHADER_OPCODE_TXF_UMS_LOGICAL:
lower_sampler_logical_send(ibld, inst, SHADER_OPCODE_TXF_UMS);
break;
diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
index 90805e4..d3e3e05 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
@@ -612,6 +612,10 @@ fs_generator::generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src
case SHADER_OPCODE_TXF:
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LD;
break;
+ case SHADER_OPCODE_TXF_CMS_W:
+ assert(devinfo->gen >= 9);
+ msg_type = GEN9_SAMPLER_MESSAGE_SAMPLE_LD2DMS_W;
+ break;
case SHADER_OPCODE_TXF_CMS:
if (devinfo->gen >= 7)
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS;
@@ -1916,6 +1920,7 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width)
case SHADER_OPCODE_TXD:
case SHADER_OPCODE_TXF:
case SHADER_OPCODE_TXF_CMS:
+ case SHADER_OPCODE_TXF_CMS_W:
case SHADER_OPCODE_TXF_UMS:
case SHADER_OPCODE_TXF_MCS:
case SHADER_OPCODE_TXL:
diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
index b85b52b..ea7b3eb 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
@@ -200,8 +200,8 @@ fs_visitor::emit_mcs_fetch(const fs_reg &coordinate, unsigned components,
fs_inst *inst = bld.emit(SHADER_OPCODE_TXF_MCS_LOGICAL, dest, srcs,
ARRAY_SIZE(srcs));
- /* We only care about one reg of response, but the sampler always writes
- * 4/8.
+ /* We only care about one or two regs of response, but the sampler always
+ * writes 4/8.
*/
inst->regs_written = 4 * dispatch_width / 8;
@@ -287,7 +287,10 @@ fs_visitor::emit_texture(ir_texture_opcode op,
opcode = SHADER_OPCODE_TXF_LOGICAL;
break;
case ir_txf_ms:
- opcode = SHADER_OPCODE_TXF_CMS_LOGICAL;
+ if (devinfo->gen >= 9)
+ opcode = SHADER_OPCODE_TXF_CMS_W_LOGICAL;
+ else
+ opcode = SHADER_OPCODE_TXF_CMS_LOGICAL;
break;
case ir_txs:
case ir_query_levels:
diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index eed73fb..1e1ea9f 100644
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -607,6 +607,10 @@ brw_instruction_name(enum opcode op)
return "txf_cms";
case SHADER_OPCODE_TXF_CMS_LOGICAL:
return "txf_cms_logical";
+ case SHADER_OPCODE_TXF_CMS_W:
+ return "txf_cms_w";
+ case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
+ return "txf_cms_w_logical";
case SHADER_OPCODE_TXF_UMS:
return "txf_ums";
case SHADER_OPCODE_TXF_UMS_LOGICAL:
@@ -1034,6 +1038,7 @@ backend_instruction::is_tex() const
opcode == SHADER_OPCODE_TXD ||
opcode == SHADER_OPCODE_TXF ||
opcode == SHADER_OPCODE_TXF_CMS ||
+ opcode == SHADER_OPCODE_TXF_CMS_W ||
opcode == SHADER_OPCODE_TXF_UMS ||
opcode == SHADER_OPCODE_TXF_MCS ||
opcode == SHADER_OPCODE_TXL ||
--
1.9.3
More information about the mesa-dev
mailing list