[Mesa-dev] [PATCH 09/21] intel/fs: Use a logical opcode for IMAGE_SIZE

[Jason Ekstrand]

---
 src/intel/compiler/brw_eu_defines.h |  1 +
 src/intel/compiler/brw_fs.cpp       | 10 ++++++++++
 src/intel/compiler/brw_fs_nir.cpp   | 14 ++++++++------
 src/intel/compiler/brw_shader.cpp   |  2 ++
 4 files changed, 21 insertions(+), 6 deletions(-)

diff --git a/src/intel/compiler/brw_eu_defines.h b/src/intel/compiler/brw_eu_defines.h
index f2190a60217..858aec07827 100644
--- a/src/intel/compiler/brw_eu_defines.h
+++ b/src/intel/compiler/brw_eu_defines.h
@@ -362,6 +362,7 @@ enum opcode {
    SHADER_OPCODE_SAMPLEINFO_LOGICAL,
 
    SHADER_OPCODE_IMAGE_SIZE,
+   SHADER_OPCODE_IMAGE_SIZE_LOGICAL,
 
    /**
     * Combines multiple sources of size 1 into a larger virtual GRF.
diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp
index 48009adaaa0..30aa494f9bf 100644
--- a/src/intel/compiler/brw_fs.cpp
+++ b/src/intel/compiler/brw_fs.cpp
@@ -740,6 +740,7 @@ fs_inst::components_read(unsigned i) const
    case SHADER_OPCODE_TXF_LOGICAL:
    case SHADER_OPCODE_TXL_LOGICAL:
    case SHADER_OPCODE_TXS_LOGICAL:
+   case SHADER_OPCODE_IMAGE_SIZE_LOGICAL:
    case FS_OPCODE_TXB_LOGICAL:
    case SHADER_OPCODE_TXF_CMS_LOGICAL:
    case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
@@ -4602,6 +4603,11 @@ lower_sampler_logical_send_gen7(const fs_builder &bld, fs_inst *inst, opcode op,
       bld.MOV(retype(sources[length], BRW_REGISTER_TYPE_UD), lod);
       length++;
       break;
+   case SHADER_OPCODE_IMAGE_SIZE:
+      /* We need an LOD; just use 0 */
+      bld.MOV(retype(sources[length], BRW_REGISTER_TYPE_UD), brw_imm_ud(0));
+      length++;
+      break;
    case SHADER_OPCODE_TXF:
       /* Unfortunately, the parameters for LD are intermixed: u, lod, v, r.
        * On Gen9 they are u, v, lod, r
@@ -5072,6 +5078,10 @@ fs_visitor::lower_logical_sends()
          lower_sampler_logical_send(ibld, inst, SHADER_OPCODE_TXS);
          break;
 
+      case SHADER_OPCODE_IMAGE_SIZE_LOGICAL:
+         lower_sampler_logical_send(ibld, inst, SHADER_OPCODE_IMAGE_SIZE);
+         break;
+
       case FS_OPCODE_TXB_LOGICAL:
          lower_sampler_logical_send(ibld, inst, FS_OPCODE_TXB);
          break;
diff --git a/src/intel/compiler/brw_fs_nir.cpp b/src/intel/compiler/brw_fs_nir.cpp
index 81e16cb9f63..769b74520b3 100644
--- a/src/intel/compiler/brw_fs_nir.cpp
+++ b/src/intel/compiler/brw_fs_nir.cpp
@@ -3778,18 +3778,20 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
                             BRW_REGISTER_TYPE_UD);
       image = bld.emit_uniformize(image);
 
+      fs_reg srcs[TEX_LOGICAL_NUM_SRCS];
+      srcs[TEX_LOGICAL_SRC_SURFACE] = image;
+      srcs[TEX_LOGICAL_SRC_SAMPLER] = brw_imm_d(0);
+      srcs[TEX_LOGICAL_SRC_COORD_COMPONENTS] = brw_imm_d(0);
+      srcs[TEX_LOGICAL_SRC_GRAD_COMPONENTS] = brw_imm_d(0);
+
       /* Since the image size is always uniform, we can just emit a SIMD8
        * query instruction and splat the result out.
        */
       const fs_builder ubld = bld.exec_all().group(8, 0);
 
-      /* The LOD also serves as the message payload */
-      fs_reg lod = ubld.vgrf(BRW_REGISTER_TYPE_UD);
-      ubld.MOV(lod, brw_imm_ud(0));
-
       fs_reg tmp = ubld.vgrf(BRW_REGISTER_TYPE_UD, 4);
-      fs_inst *inst = ubld.emit(SHADER_OPCODE_IMAGE_SIZE, tmp, lod, image);
-      inst->mlen = 1;
+      fs_inst *inst = ubld.emit(SHADER_OPCODE_IMAGE_SIZE_LOGICAL,
+                                tmp, srcs, ARRAY_SIZE(srcs));
       inst->size_written = 4 * REG_SIZE;
 
       for (unsigned c = 0; c < instr->dest.ssa.num_components; ++c) {
diff --git a/src/intel/compiler/brw_shader.cpp b/src/intel/compiler/brw_shader.cpp
index b33d56c9877..8d8ac0348b5 100644
--- a/src/intel/compiler/brw_shader.cpp
+++ b/src/intel/compiler/brw_shader.cpp
@@ -272,6 +272,8 @@ brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op)
 
    case SHADER_OPCODE_IMAGE_SIZE:
       return "image_size";
+   case SHADER_OPCODE_IMAGE_SIZE_LOGICAL:
+      return "image_size_logical";
 
    case SHADER_OPCODE_SHADER_TIME_ADD:
       return "shader_time_add";
-- 
2.19.1