intel/fs: Use SHADER_OPCODE_SEND for varying UBO pulls on gen7+

Reviewed-by: Iago Toral Quiroga <[email protected]>

7 files changed, 25 insertions, 88 deletions

diff --git a/src/intel/compiler/brw_eu_defines.h b/src/intel/compiler/brw_eu_defines.h
index 4445f388d38..d3dfd6dc7e8 100644
--- a/src/intel/compiler/brw_eu_defines.h
+++ b/src/intel/compiler/brw_eu_defines.h
@@ -526,7 +526,6 @@ enum opcode {
    FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
    FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7,
    FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4,
-   FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7,
    FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL,
    FS_OPCODE_DISCARD_JUMP,
    FS_OPCODE_SET_SAMPLE_ID,
diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp
index d9c339b1f08..9ecabd58763 100644
--- a/src/intel/compiler/brw_fs.cpp
+++ b/src/intel/compiler/brw_fs.cpp
@@ -217,7 +217,6 @@ fs_inst::is_send_from_grf() const
 {
    switch (opcode) {
    case SHADER_OPCODE_SEND:
-   case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
    case SHADER_OPCODE_SHADER_TIME_ADD:
    case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
    case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
@@ -5196,16 +5195,37 @@ lower_varying_pull_constant_logical_send(const fs_builder &bld, fs_inst *inst)
    const gen_device_info *devinfo = bld.shader->devinfo;
 
    if (devinfo->gen >= 7) {
+      fs_reg index = inst->src[0];
       /* We are switching the instruction from an ALU-like instruction to a
        * send-from-grf instruction.  Since sends can't handle strides or
        * source modifiers, we have to make a copy of the offset source.
        */
-      fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD);
-      bld.MOV(tmp, inst->src[1]);
-      inst->src[1] = tmp;
+      fs_reg offset = bld.vgrf(BRW_REGISTER_TYPE_UD);
+      bld.MOV(offset, inst->src[1]);
 
-      inst->opcode = FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7;
+      const unsigned simd_mode =
+         inst->exec_size <= 8 ? BRW_SAMPLER_SIMD_MODE_SIMD8 :
+                                BRW_SAMPLER_SIMD_MODE_SIMD16;
+
+      inst->opcode = SHADER_OPCODE_SEND;
       inst->mlen = inst->exec_size / 8;
+      inst->resize_sources(3);
+
+      inst->sfid = BRW_SFID_SAMPLER;
+      inst->desc = brw_sampler_desc(devinfo, 0, 0,
+                                    GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
+                                    simd_mode, 0);
+      if (index.file == IMM) {
+         inst->desc |= index.ud & 0xff;
+         inst->src[0] = brw_imm_ud(0);
+      } else {
+         const fs_builder ubld = bld.exec_all().group(1, 0);
+         fs_reg tmp = ubld.vgrf(BRW_REGISTER_TYPE_UD);
+         ubld.AND(tmp, index, brw_imm_ud(0xff));
+         inst->src[0] = component(tmp, 0);
+      }
+      inst->src[1] = brw_imm_ud(0); /* ex_desc */
+      inst->src[2] = offset; /* payload */
    } else {
       const fs_reg payload(MRF, FIRST_PULL_LOAD_MRF(devinfo->gen),
                            BRW_REGISTER_TYPE_UD);
@@ -5727,7 +5747,6 @@ get_lowered_simd_width(const struct gen_device_info *devinfo,
    case FS_OPCODE_DDX_FINE:
    case FS_OPCODE_DDY_COARSE:
    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
-   case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
    case FS_OPCODE_PACK_HALF_2x16_SPLIT:
    case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
    case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
diff --git a/src/intel/compiler/brw_fs.h b/src/intel/compiler/brw_fs.h
index 4e913eb8d80..6467b4c6d95 100644
--- a/src/intel/compiler/brw_fs.h
+++ b/src/intel/compiler/brw_fs.h
@@ -444,10 +444,6 @@ private:
    void generate_varying_pull_constant_load_gen4(fs_inst *inst,
                                                  struct brw_reg dst,
                                                  struct brw_reg index);
-   void generate_varying_pull_constant_load_gen7(fs_inst *inst,
-                                                 struct brw_reg dst,
-                                                 struct brw_reg index,
-                                                 struct brw_reg offset);
    void generate_mov_dispatch_to_flags(fs_inst *inst);
 
    void generate_pixel_interpolator_query(fs_inst *inst,
diff --git a/src/intel/compiler/brw_fs_cse.cpp b/src/intel/compiler/brw_fs_cse.cpp
index 19089246c4f..9b897804032 100644
--- a/src/intel/compiler/brw_fs_cse.cpp
+++ b/src/intel/compiler/brw_fs_cse.cpp
@@ -74,7 +74,6 @@ is_expression(const fs_visitor *v, const fs_inst *const inst)
    case FS_OPCODE_FB_READ_LOGICAL:
    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
-   case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
    case FS_OPCODE_LINTERP:
    case SHADER_OPCODE_FIND_LIVE_CHANNEL:
    case SHADER_OPCODE_BROADCAST:
diff --git a/src/intel/compiler/brw_fs_generator.cpp b/src/intel/compiler/brw_fs_generator.cpp
index 35762b43615..544d19826f4 100644
--- a/src/intel/compiler/brw_fs_generator.cpp
+++ b/src/intel/compiler/brw_fs_generator.cpp
@@ -1515,75 +1515,6 @@ fs_generator::generate_varying_pull_constant_load_gen4(fs_inst *inst,
 }
 
 void
-fs_generator::generate_varying_pull_constant_load_gen7(fs_inst *inst,
-                                                       struct brw_reg dst,
-                                                       struct brw_reg index,
-                                                       struct brw_reg offset)
-{
-   assert(devinfo->gen >= 7);
-   /* Varying-offset pull constant loads are treated as a normal expression on
-    * gen7, so the fact that it's a send message is hidden at the IR level.
-    */
-   assert(inst->header_size == 0);
-   assert(inst->mlen);
-   assert(index.type == BRW_REGISTER_TYPE_UD);
-
-   uint32_t simd_mode, rlen;
-   if (inst->exec_size == 16) {
-      rlen = 8;
-      simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD16;
-   } else {
-      assert(inst->exec_size == 8);
-      rlen = 4;
-      simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD8;
-   }
-
-   if (index.file == BRW_IMMEDIATE_VALUE) {
-
-      uint32_t surf_index = index.ud;
-
-      brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
-      brw_inst_set_sfid(devinfo, send, BRW_SFID_SAMPLER);
-      brw_set_dest(p, send, retype(dst, BRW_REGISTER_TYPE_UW));
-      brw_set_src0(p, send, offset);
-      brw_set_desc(p, send,
-                   brw_message_desc(devinfo, inst->mlen, rlen, false) |
-                   brw_sampler_desc(devinfo, surf_index,
-                                    0, /* LD message ignores sampler unit */
-                                    GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
-                                    simd_mode, 0));
-
-   } else {
-
-      struct brw_reg addr = vec1(retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD));
-
-      brw_push_insn_state(p);
-      brw_set_default_mask_control(p, BRW_MASK_DISABLE);
-
-      /* a0.0 = surf_index & 0xff */
-      brw_inst *insn_and = brw_next_insn(p, BRW_OPCODE_AND);
-      brw_inst_set_exec_size(p->devinfo, insn_and, BRW_EXECUTE_1);
-      brw_set_dest(p, insn_and, addr);
-      brw_set_src0(p, insn_and, vec1(retype(index, BRW_REGISTER_TYPE_UD)));
-      brw_set_src1(p, insn_and, brw_imm_ud(0x0ff));
-
-      brw_pop_insn_state(p);
-
-      /* dst = send(offset, a0.0 | <descriptor>) */
-      brw_send_indirect_message(
-         p, BRW_SFID_SAMPLER, retype(dst, BRW_REGISTER_TYPE_UW),
-         offset, addr,
-         brw_message_desc(devinfo, inst->mlen, rlen, false) |
-         brw_sampler_desc(devinfo,
-                          0 /* surface */,
-                          0 /* sampler */,
-                          GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
-                          simd_mode,
-                          0));
-   }
-}
-
-void
 fs_generator::generate_pixel_interpolator_query(fs_inst *inst,
                                                 struct brw_reg dst,
                                                 struct brw_reg src,
@@ -2133,10 +2064,6 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width)
 	 generate_varying_pull_constant_load_gen4(inst, dst, src[0]);
 	 break;
 
-      case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
-	 generate_varying_pull_constant_load_gen7(inst, dst, src[0], src[1]);
-	 break;
-
       case FS_OPCODE_REP_FB_WRITE:
       case FS_OPCODE_FB_WRITE:
 	 generate_fb_write(inst, src[0]);
diff --git a/src/intel/compiler/brw_schedule_instructions.cpp b/src/intel/compiler/brw_schedule_instructions.cpp
index 46d3111045d..be57802b227 100644
--- a/src/intel/compiler/brw_schedule_instructions.cpp
+++ b/src/intel/compiler/brw_schedule_instructions.cpp
@@ -323,7 +323,6 @@ schedule_node::set_latency_gen7(bool is_haswell)
       break;
 
    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4:
-   case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
    case VS_OPCODE_PULL_CONSTANT_LOAD:
diff --git a/src/intel/compiler/brw_shader.cpp b/src/intel/compiler/brw_shader.cpp
index 5b1d50052ff..b4c74871a48 100644
--- a/src/intel/compiler/brw_shader.cpp
+++ b/src/intel/compiler/brw_shader.cpp
@@ -407,8 +407,6 @@ brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op)
       return "uniform_pull_const_gen7";
    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4:
       return "varying_pull_const_gen4";
-   case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
-      return "varying_pull_const_gen7";
    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
       return "varying_pull_const_logical";