intel/fs: Add DWord scattered read/write opcodes

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

diff --git a/src/intel/compiler/brw_eu.h b/src/intel/compiler/brw_eu.h
index c35725bfe2b66079bc2bc2afb97ca8efbe7e1d38..ce1368a7fd3e235835955b9495175c3ac8523002 100644 (file)
--- a/src/intel/compiler/brw_eu.h
+++ b/src/intel/compiler/brw_eu.h
@@ -697,6 +697,37 @@ brw_dp_byte_scattered_rw_desc(const struct gen_device_info *devinfo,
    return brw_dp_surface_desc(devinfo, msg_type, msg_control);
 }
 
+static inline uint32_t
+brw_dp_dword_scattered_rw_desc(const struct gen_device_info *devinfo,
+                               unsigned exec_size,
+                               bool write)
+{
+   assert(exec_size == 8 || exec_size == 16);
+
+   unsigned msg_type;
+   if (write) {
+      if (devinfo->gen >= 6) {
+         msg_type = GEN6_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE;
+      } else {
+         msg_type = BRW_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE;
+      }
+   } else {
+      if (devinfo->gen >= 7) {
+         msg_type = GEN7_DATAPORT_DC_DWORD_SCATTERED_READ;
+      } else if (devinfo->gen > 4 || devinfo->is_g4x) {
+         msg_type = G45_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ;
+      } else {
+         msg_type = BRW_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ;
+      }
+   }
+
+   const unsigned msg_control =
+      SET_BITS(1, 1, 1) | /* Legacy SIMD Mode */
+      SET_BITS(exec_size == 16, 0, 0);
+
+   return brw_dp_surface_desc(devinfo, msg_type, msg_control);
+}
+
 static inline uint32_t
 brw_dp_a64_untyped_surface_rw_desc(const struct gen_device_info *devinfo,
                                    unsigned exec_size, /**< 0 for SIMD4x2 */

diff --git a/src/intel/compiler/brw_eu_defines.h b/src/intel/compiler/brw_eu_defines.h
index 8aa2c14fddf3591c19401b2c49b90a3694978ac3..d2d8d877e2cedc0368724cd30bc2a535fb862dd4 100644 (file)
--- a/src/intel/compiler/brw_eu_defines.h
+++ b/src/intel/compiler/brw_eu_defines.h
@@ -446,6 +446,8 @@ enum opcode {
     */
    SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL,
    SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL,
+   SHADER_OPCODE_DWORD_SCATTERED_READ_LOGICAL,
+   SHADER_OPCODE_DWORD_SCATTERED_WRITE_LOGICAL,
 
    /**
     * Memory fence messages.

diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp
index 538745880c6fd911d8749c295dd7cf6a6996cd6f..aedf8314c0d3969275b8ec6017b75eca1634f140 100644 (file)
--- a/src/intel/compiler/brw_fs.cpp
+++ b/src/intel/compiler/brw_fs.cpp
@@ -925,6 +925,7 @@ fs_inst::components_read(unsigned i) const
       }
 
    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
+   case SHADER_OPCODE_DWORD_SCATTERED_READ_LOGICAL:
       /* Scattered logical opcodes use the following params:
        * src[0] Surface coordinates
        * src[1] Surface operation source (ignored for reads)
@@ -937,6 +938,7 @@ fs_inst::components_read(unsigned i) const
       return i == SURFACE_LOGICAL_SRC_DATA ? 0 : 1;
 
    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
+   case SHADER_OPCODE_DWORD_SCATTERED_WRITE_LOGICAL:
       assert(src[SURFACE_LOGICAL_SRC_IMM_DIMS].file == IMM &&
              src[SURFACE_LOGICAL_SRC_IMM_ARG].file == IMM);
       return 1;
@@ -5462,6 +5464,13 @@ lower_surface_logical_send(const fs_builder &bld, fs_inst *inst)
       sfid = GEN7_SFID_DATAPORT_DATA_CACHE;
       break;
 
+   case SHADER_OPCODE_DWORD_SCATTERED_READ_LOGICAL:
+   case SHADER_OPCODE_DWORD_SCATTERED_WRITE_LOGICAL:
+      sfid =  devinfo->gen >= 7 ? GEN7_SFID_DATAPORT_DATA_CACHE :
+              devinfo->gen >= 6 ? GEN6_SFID_DATAPORT_RENDER_CACHE :
+                                  BRW_DATAPORT_READ_TARGET_RENDER_CACHE;
+      break;
+
    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
@@ -5515,6 +5524,18 @@ lower_surface_logical_send(const fs_builder &bld, fs_inst *inst)
                                            true    /* write */);
       break;
 
+   case SHADER_OPCODE_DWORD_SCATTERED_READ_LOGICAL:
+      assert(arg.ud == 32); /* bit_size */
+      desc = brw_dp_dword_scattered_rw_desc(devinfo, inst->exec_size,
+                                            false  /* write */);
+      break;
+
+   case SHADER_OPCODE_DWORD_SCATTERED_WRITE_LOGICAL:
+      assert(arg.ud == 32); /* bit_size */
+      desc = brw_dp_dword_scattered_rw_desc(devinfo, inst->exec_size,
+                                            true   /* write */);
+      break;
+
    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
       desc = brw_dp_untyped_atomic_desc(devinfo, inst->exec_size,
                                         arg.ud, /* atomic_op */
@@ -5874,6 +5895,8 @@ fs_visitor::lower_logical_sends()
       case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
       case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
       case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
+      case SHADER_OPCODE_DWORD_SCATTERED_READ_LOGICAL:
+      case SHADER_OPCODE_DWORD_SCATTERED_WRITE_LOGICAL:
       case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
       case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL:
       case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
@@ -6468,6 +6491,8 @@ get_lowered_simd_width(const struct gen_device_info *devinfo,
    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
+   case SHADER_OPCODE_DWORD_SCATTERED_WRITE_LOGICAL:
+   case SHADER_OPCODE_DWORD_SCATTERED_READ_LOGICAL:
       return MIN2(16, inst->exec_size);
 
    case SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL:

diff --git a/src/intel/compiler/brw_schedule_instructions.cpp b/src/intel/compiler/brw_schedule_instructions.cpp
index 08e89151e9145ae6ad56acf20a4f53ce1975c5c5..575693fc96101a46b8df7b0d2cddbc9d05258af4 100644 (file)
--- a/src/intel/compiler/brw_schedule_instructions.cpp
+++ b/src/intel/compiler/brw_schedule_instructions.cpp
@@ -424,6 +424,8 @@ schedule_node::set_latency_gen7(bool is_haswell)
 
       case GEN7_SFID_DATAPORT_DATA_CACHE:
          switch ((inst->desc >> 14) & 0x1f) {
+         case GEN7_DATAPORT_DC_DWORD_SCATTERED_READ:
+         case GEN6_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE:
          case HSW_DATAPORT_DC_PORT0_BYTE_SCATTERED_READ:
          case HSW_DATAPORT_DC_PORT0_BYTE_SCATTERED_WRITE:
             /* We have no data for this but assume it's roughly the same as

diff --git a/src/intel/compiler/brw_shader.cpp b/src/intel/compiler/brw_shader.cpp
index 94976a12c000d5c658bbf57af6f9eeecb8e31161..f95fcd99e67ac43435cfdf919ad49041b927421c 100644 (file)
--- a/src/intel/compiler/brw_shader.cpp
+++ b/src/intel/compiler/brw_shader.cpp
@@ -331,6 +331,10 @@ brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op)
       return "byte_scattered_read_logical";
    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
       return "byte_scattered_write_logical";
+   case SHADER_OPCODE_DWORD_SCATTERED_READ_LOGICAL:
+      return "dword_scattered_read_logical";
+   case SHADER_OPCODE_DWORD_SCATTERED_WRITE_LOGICAL:
+      return "dword_scattered_write_logical";
 
    case SHADER_OPCODE_LOAD_PAYLOAD:
       return "load_payload";
@@ -1055,6 +1059,7 @@ backend_instruction::has_side_effects() const
    case SHADER_OPCODE_A64_UNTYPED_ATOMIC_INT64_LOGICAL:
    case SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL:
    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
+   case SHADER_OPCODE_DWORD_SCATTERED_WRITE_LOGICAL:
    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
    case SHADER_OPCODE_MEMORY_FENCE:
@@ -1088,6 +1093,7 @@ backend_instruction::is_volatile() const
    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
+   case SHADER_OPCODE_DWORD_SCATTERED_READ_LOGICAL:
    case SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL:
    case SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL:
    case SHADER_OPCODE_URB_READ_SIMD8: