intel/compiler: Use SHADER_OPCODE_SEND for PI messages

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21094>

src/intel/compiler/brw_eu.h

@@ -1671,9 +1671,13 @@ brw_pixel_interp_desc(UNUSED const struct intel_device_info *devinfo,
                       unsigned msg_type,
                       bool noperspective,
                       bool coarse_pixel_rate,
-                      unsigned simd_mode,
+                      unsigned exec_size,
-                      unsigned slot_group)
+                      unsigned group)
 {
+   assert(exec_size == 8 || exec_size == 16);
+   const bool simd_mode = exec_size == 16;
+   const bool slot_group = group >= 16;
+
    assert(devinfo->ver >= 10 || !coarse_pixel_rate);
    return (SET_BITS(slot_group, 11, 11) |
            SET_BITS(msg_type, 13, 12) |

src/intel/compiler/brw_eu_emit.c

@@ -3345,38 +3345,6 @@ brw_memory_fence(struct brw_codegen *p,
       brw_set_memory_fence_message(p, insn, sfid, commit_enable, bti);
 }
 
-void
-brw_pixel_interpolator_query(struct brw_codegen *p,
-                             struct brw_reg dest,
-                             struct brw_reg mrf,
-                             bool noperspective,
-                             bool coarse_pixel_rate,
-                             unsigned mode,
-                             struct brw_reg data,
-                             unsigned msg_length,
-                             unsigned response_length)
-{
-   const struct intel_device_info *devinfo = p->devinfo;
-   const uint16_t exec_size = brw_get_default_exec_size(p);
-   const unsigned slot_group = brw_get_default_group(p) / 16;
-   const unsigned simd_mode = (exec_size == BRW_EXECUTE_16);
-   const unsigned desc =
-      brw_message_desc(devinfo, msg_length, response_length, false) |
-      brw_pixel_interp_desc(devinfo, mode, noperspective, coarse_pixel_rate,
-                            simd_mode, slot_group);
-
-   /* brw_send_indirect_message will automatically use a direct send message
-    * if data is actually immediate.
-    */
-   brw_send_indirect_message(p,
-                             GFX7_SFID_PIXEL_INTERPOLATOR,
-                             dest,
-                             mrf,
-                             vec1(data),
-                             desc,
-                             false);
-}
-
 void
 brw_find_live_channel(struct brw_codegen *p, struct brw_reg dst, bool last)
 {

src/intel/compiler/brw_fs.h

@@ -625,12 +625,6 @@ private:
                                                  struct brw_reg dst,
                                                  struct brw_reg index);
 
-   void generate_pixel_interpolator_query(fs_inst *inst,
-                                          struct brw_reg dst,
-                                          struct brw_reg src,
-                                          struct brw_reg msg_data,
-                                          unsigned msg_type);
-
    void generate_set_sample_id(fs_inst *inst,
                                struct brw_reg dst,
                                struct brw_reg src0,

src/intel/compiler/brw_fs_generator.cpp

@@ -1611,31 +1611,6 @@ fs_generator::generate_varying_pull_constant_load_gfx4(fs_inst *inst,
                                  msg_type, simd_mode, return_format));
 }
 
-void
-fs_generator::generate_pixel_interpolator_query(fs_inst *inst,
-                                                struct brw_reg dst,
-                                                struct brw_reg src,
-                                                struct brw_reg msg_data,
-                                                unsigned msg_type)
-{
-   const bool has_payload = inst->src[0].file != BAD_FILE;
-   assert(msg_data.type == BRW_REGISTER_TYPE_UD);
-   assert(inst->size_written % REG_SIZE == 0);
-
-   struct brw_wm_prog_data *prog_data = brw_wm_prog_data(this->prog_data);
-
-   brw_pixel_interpolator_query(p,
-         retype(dst, BRW_REGISTER_TYPE_UW),
-         /* If we don't have a payload, what we send doesn't matter */
-         has_payload ? src : brw_vec8_grf(0, 0),
-         inst->pi_noperspective,
-         prog_data->per_coarse_pixel_dispatch,
-         msg_type,
-         msg_data,
-         has_payload ? 2 * inst->exec_size / 8 : 1,
-         inst->size_written / REG_SIZE);
-}
-
 /* Sets vstride=1, width=4, hstride=0 of register src1 during
  * the ADD instruction.
  */
@@ -2389,24 +2364,6 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width,
          }
          break;
 
-      case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
-         generate_pixel_interpolator_query(inst, dst, src[0], src[1],
-                                           GFX7_PIXEL_INTERPOLATOR_LOC_SAMPLE);
-         send_count++;
-         break;
-
-      case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
-         generate_pixel_interpolator_query(inst, dst, src[0], src[1],
-                                           GFX7_PIXEL_INTERPOLATOR_LOC_SHARED_OFFSET);
-         send_count++;
-         break;
-
-      case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
-         generate_pixel_interpolator_query(inst, dst, src[0], src[1],
-                                           GFX7_PIXEL_INTERPOLATOR_LOC_PER_SLOT_OFFSET);
-         send_count++;
-         break;
-
       case CS_OPCODE_CS_TERMINATE:
          generate_cs_terminate(inst, src[0]);
          send_count++;

src/intel/compiler/brw_ir_performance.cpp

@@ -1107,6 +1107,13 @@ namespace {
                abort();
             }
 
+         case GFX7_SFID_PIXEL_INTERPOLATOR:
+            if (devinfo->ver >= 7)
+               return calculate_desc(info, EU_UNIT_PI, 2, 0, 0, 14 /* XXX */, 0,
+                                     0, 90 /* XXX */, 0, 0, 0, 0);
+            else
+               abort();
+
          case GFX12_SFID_UGM:
          case GFX12_SFID_TGM:
          case GFX12_SFID_SLM:

src/intel/compiler/brw_lower_logical_sends.cpp

@@ -2446,6 +2446,61 @@ lower_math_logical_send(const fs_builder &bld, fs_inst *inst)
    }
 }
 
+static void
+lower_interpolator_logical_send(const fs_builder &bld, fs_inst *inst,
+                                const struct brw_wm_prog_data *wm_prog_data)
+{
+   const intel_device_info *devinfo = bld.shader->devinfo;
+
+   /* We have to send something */
+   fs_reg payload = brw_vec8_grf(0, 0);
+   unsigned mlen = 1;
+
+   const fs_reg desc = inst->src[1];
+
+   unsigned mode;
+   switch (inst->opcode) {
+   case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
+      assert(inst->src[0].file == BAD_FILE);
+      mode = GFX7_PIXEL_INTERPOLATOR_LOC_SAMPLE;
+      break;
+
+   case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
+      assert(inst->src[0].file == BAD_FILE);
+      mode = GFX7_PIXEL_INTERPOLATOR_LOC_SHARED_OFFSET;
+      break;
+
+   case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
+      payload = inst->src[0];
+      mlen = 2 * inst->exec_size / 8;
+      mode = GFX7_PIXEL_INTERPOLATOR_LOC_PER_SLOT_OFFSET;
+      break;
+
+   default:
+      unreachable("Invalid interpolator instruction");
+   }
+
+   uint32_t desc_imm =
+      brw_pixel_interp_desc(devinfo, mode, inst->pi_noperspective,
+                            wm_prog_data->per_coarse_pixel_dispatch,
+                            inst->exec_size, inst->group);
+
+   assert(bld.shader->devinfo->ver >= 7);
+   inst->opcode = SHADER_OPCODE_SEND;
+   inst->sfid = GFX7_SFID_PIXEL_INTERPOLATOR;
+   inst->desc = desc_imm;
+   inst->ex_desc = 0;
+   inst->mlen = mlen;
+   inst->ex_mlen = 0;
+   inst->send_has_side_effects = false;
+   inst->send_is_volatile = false;
+
+   inst->resize_sources(3);
+   inst->src[0] = desc;
+   inst->src[1] = brw_imm_ud(0); /* ex_desc */
+   inst->src[2] = payload;
+}
+
 static void
 lower_btd_logical_send(const fs_builder &bld, fs_inst *inst)
 {
@@ -2749,6 +2804,13 @@ fs_visitor::lower_logical_sends()
             continue;
          }
 
+      case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
+      case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
+      case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
+         lower_interpolator_logical_send(ibld, inst,
+                                         brw_wm_prog_data(prog_data));
+         break;
+
       case SHADER_OPCODE_BTD_SPAWN_LOGICAL:
       case SHADER_OPCODE_BTD_RETIRE_LOGICAL:
          lower_btd_logical_send(ibld, inst);

src/intel/compiler/brw_schedule_instructions.cpp

@@ -532,6 +532,10 @@ schedule_node::set_latency_gfx7(bool is_haswell)
          }
          break;
 
+      case GFX7_SFID_PIXEL_INTERPOLATOR:
+         latency = 50; /* TODO */
+         break;
+
       case GFX12_SFID_UGM:
       case GFX12_SFID_TGM:
       case GFX12_SFID_SLM: