i965: Push most TES inputs in vec4 mode.

(This is commit 4a1c8a3037cd29938b2a6e2c680c341e9903cfbe for vec4 mode.)

Using the push model for inputs is much more efficient than pulling
inputs - the hardware can simply copy a large chunk into URB registers
at thread creation time, rather than having the thread send messages to
request data from the L3 cache.  Unfortunately, it's possible to have
more TES inputs than fit in registers, so we have to fall back to the
pull model in some cases.

However, it turns out that most tessellation evaluation shaders are
fairly simple, and don't use many inputs.  An arbitrary cut-off of
24 vec4 slots (12 registers) should suffice.  (I chose this instead of
the 32 vec4 slots used in the scalar backend to avoid regressing a few
Piglit tests due to the vec4 register allocator being too stupid to
figure out what to do.  We probably ought to fix that, but it's a
separate issue.)

Improves performance in GPUTest's tessmark_x64 microbenchmark by
41.5394% +/- 0.288519% (n = 115) at 1024x768 on my Clevo W740SU
(with Iris Pro 5200).

Improves performance in Synmark's Gl40TerrainFlyTess microbenchmark by
38.3576% +/- 0.759748% (n = 42).

v2: Simplify abs/negate handling, as requested by Matt.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>
(cherry picked from commit 24994ae926629ac8521df3cab4a02eb81de15907)

2 files changed, 54 insertions, 34 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
index 74ec4f0e87..9b721e5b8a 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
@@ -685,9 +685,7 @@ vec4_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr)
    case nir_intrinsic_load_instance_id:
    case nir_intrinsic_load_base_instance:
    case nir_intrinsic_load_draw_id:
-   case nir_intrinsic_load_invocation_id:
-   case nir_intrinsic_load_tess_level_inner:
-   case nir_intrinsic_load_tess_level_outer: {
+   case nir_intrinsic_load_invocation_id: {
       gl_system_value sv = nir_system_value_from_intrinsic(instr->intrinsic);
       src_reg val = src_reg(nir_system_values[sv]);
       assert(val.file != BAD_FILE);
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_tes.cpp b/src/mesa/drivers/dri/i965/brw_vec4_tes.cpp
index ce5fefc75a..e3c23f1a52 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_tes.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_tes.cpp
@@ -28,6 +28,7 @@
  */
 
 #include "brw_vec4_tes.h"
+#include "brw_cfg.h"
 
 namespace brw {
 
@@ -53,39 +54,10 @@ vec4_tes_visitor::make_reg_for_system_value(int location, const glsl_type *type)
 void
 vec4_tes_visitor::nir_setup_system_value_intrinsic(nir_intrinsic_instr *instr)
 {
-   const struct brw_tes_prog_data *tes_prog_data =
-      (const struct brw_tes_prog_data *) prog_data;
-
    switch (instr->intrinsic) {
-   case nir_intrinsic_load_tess_level_outer: {
-      dst_reg dst(this, glsl_type::vec4_type);
-      nir_system_values[SYSTEM_VALUE_TESS_LEVEL_OUTER] = dst;
-
-      dst_reg temp(this, glsl_type::vec4_type);
-      vec4_instruction *read =
-         emit(VEC4_OPCODE_URB_READ, temp, input_read_header);
-      read->offset = 1;
-      read->urb_write_flags = BRW_URB_WRITE_PER_SLOT_OFFSET;
-      emit(MOV(dst, swizzle(src_reg(temp), BRW_SWIZZLE_WZYX)));
-      break;
-   }
-   case nir_intrinsic_load_tess_level_inner: {
-      dst_reg dst(this, glsl_type::vec2_type);
-      nir_system_values[SYSTEM_VALUE_TESS_LEVEL_INNER] = dst;
-
-      /* Set up the message header to reference the proper parts of the URB */
-      dst_reg temp(this, glsl_type::vec4_type);
-      vec4_instruction *read =
-         emit(VEC4_OPCODE_URB_READ, temp, input_read_header);
-      read->urb_write_flags = BRW_URB_WRITE_PER_SLOT_OFFSET;
-      if (tes_prog_data->domain == BRW_TESS_DOMAIN_QUAD) {
-         emit(MOV(dst, swizzle(src_reg(temp), BRW_SWIZZLE_WZYX)));
-      } else {
-         read->offset = 1;
-         emit(MOV(dst, src_reg(temp)));
-      }
+   case nir_intrinsic_load_tess_level_outer:
+   case nir_intrinsic_load_tess_level_inner:
       break;
-   }
    default:
       vec4_visitor::nir_setup_system_value_intrinsic(instr);
    }
@@ -105,6 +77,25 @@ vec4_tes_visitor::setup_payload()
 
    reg = setup_uniforms(reg);
 
+   foreach_block_and_inst(block, vec4_instruction, inst, cfg) {
+      for (int i = 0; i < 3; i++) {
+         if (inst->src[i].file != ATTR)
+            continue;
+
+         struct brw_reg grf =
+            brw_vec4_grf(reg + inst->src[i].nr / 2, 4 * (inst->src[i].nr % 2));
+         grf = stride(grf, 0, 4, 1);
+         grf.swizzle = inst->src[i].swizzle;
+         grf.type = inst->src[i].type;
+         grf.abs = inst->src[i].abs;
+         grf.negate = inst->src[i].negate;
+
+         inst->src[i] = grf;
+      }
+   }
+
+   reg += 8 * prog_data->urb_read_length;
+
    this->first_non_payload_grf = reg;
 }
 
@@ -148,12 +139,30 @@ vec4_tes_visitor::emit_urb_write_opcode(bool complete)
 void
 vec4_tes_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr)
 {
+   const struct brw_tes_prog_data *tes_prog_data =
+      (const struct brw_tes_prog_data *) prog_data;
+
    switch (instr->intrinsic) {
    case nir_intrinsic_load_tess_coord:
       /* gl_TessCoord is part of the payload in g1 channels 0-2 and 4-6. */
       emit(MOV(get_nir_dest(instr->dest, BRW_REGISTER_TYPE_F),
                src_reg(brw_vec8_grf(1, 0))));
       break;
+   case nir_intrinsic_load_tess_level_outer:
+      emit(MOV(get_nir_dest(instr->dest, BRW_REGISTER_TYPE_F),
+               swizzle(src_reg(ATTR, 1, glsl_type::vec4_type),
+                       BRW_SWIZZLE_WZYX)));
+      break;
+   case nir_intrinsic_load_tess_level_inner:
+      if (tes_prog_data->domain == BRW_TESS_DOMAIN_QUAD) {
+         emit(MOV(get_nir_dest(instr->dest, BRW_REGISTER_TYPE_F),
+                  swizzle(src_reg(ATTR, 0, glsl_type::vec4_type),
+                          BRW_SWIZZLE_WZYX)));
+      } else {
+         emit(MOV(get_nir_dest(instr->dest, BRW_REGISTER_TYPE_F),
+                  src_reg(ATTR, 1, glsl_type::float_type)));
+      }
+      break;
    case nir_intrinsic_load_primitive_id:
       emit(TES_OPCODE_GET_PRIMITIVE_ID,
            get_nir_dest(instr->dest, BRW_REGISTER_TYPE_UD));
@@ -169,6 +178,19 @@ vec4_tes_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr)
          header = src_reg(this, glsl_type::uvec4_type);
          emit(TES_OPCODE_ADD_INDIRECT_URB_OFFSET, dst_reg(header),
               input_read_header, indirect_offset);
+      } else {
+         /* Arbitrarily only push up to 24 vec4 slots worth of data,
+          * which is 12 registers (since each holds 2 vec4 slots).
+          */
+         const unsigned max_push_slots = 24;
+         if (imm_offset < max_push_slots) {
+            emit(MOV(get_nir_dest(instr->dest, BRW_REGISTER_TYPE_D),
+                     src_reg(ATTR, imm_offset, glsl_type::ivec4_type)));
+            prog_data->urb_read_length =
+               MAX2(prog_data->urb_read_length,
+                    DIV_ROUND_UP(imm_offset + 1, 2));
+            break;
+         }
       }
 
       dst_reg temp(this, glsl_type::ivec4_type);