intel/compiler: Add ctor to fs_builder that just takes the shader

Uses the dispatch_width from the shader (fs_visitor).  This was not
possible before because the dispatch_width was not part of
backend_shader.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26323>

src/intel/compiler/brw_fs.cpp

@@ -1232,7 +1232,7 @@ fs_visitor::emit_gs_thread_end()
       emit_gs_control_data_bits(this->final_gs_vertex_count);
    }
 
-   const fs_builder abld = fs_builder(this, dispatch_width).at_end().annotate("thread end");
+   const fs_builder abld = fs_builder(this).at_end().annotate("thread end");
    fs_inst *inst;
 
    if (gs_prog_data->static_vertex_count != -1) {
@@ -3308,7 +3308,7 @@ fs_visitor::emit_repclear_shader()
               BRW_VERTICAL_STRIDE_8, BRW_WIDTH_2, BRW_HORIZONTAL_STRIDE_4,
               BRW_SWIZZLE_XYZW, WRITEMASK_XYZW);
 
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
    bld.exec_all().group(4, 0).MOV(color_output, color_input);
 
    if (key->nr_color_regions > 1) {
@@ -5313,7 +5313,7 @@ fs_visitor::lower_simd_width()
           */
          const unsigned max_width = MAX2(inst->exec_size, lower_width);
 
-         const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+         const fs_builder bld = fs_builder(this).at_end();
          const fs_builder ibld = bld.at(block, inst)
                                     .exec_all(inst->force_writemask_all)
                                     .group(max_width, inst->group / max_width);
@@ -6691,7 +6691,7 @@ fs_visitor::set_tcs_invocation_id()
 {
    struct brw_tcs_prog_data *tcs_prog_data = brw_tcs_prog_data(prog_data);
    struct brw_vue_prog_data *vue_prog_data = &tcs_prog_data->base;
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
 
    const unsigned instance_id_mask =
       (devinfo->verx10 >= 125) ? INTEL_MASK(7, 0) :
@@ -6743,7 +6743,7 @@ fs_visitor::emit_tcs_thread_end()
    if (devinfo->ver != 8 && mark_last_urb_write_with_eot())
       return;
 
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
 
    /* Emit a URB write to end the thread.  On Broadwell, we use this to write
     * zero to the "TR DS Cache Disable" bit (we haven't implemented a fancy
@@ -6766,7 +6766,7 @@ fs_visitor::run_tcs()
    assert(stage == MESA_SHADER_TESS_CTRL);
 
    struct brw_vue_prog_data *vue_prog_data = brw_vue_prog_data(prog_data);
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
 
    assert(vue_prog_data->dispatch_mode == DISPATCH_MODE_TCS_SINGLE_PATCH ||
           vue_prog_data->dispatch_mode == DISPATCH_MODE_TCS_MULTI_PATCH);
@@ -6866,7 +6866,7 @@ fs_visitor::run_gs()
        * Otherwise, we need to initialize it to 0 here.
        */
       if (gs_compile->control_data_header_size_bits <= 32) {
-         const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+         const fs_builder bld = fs_builder(this).at_end();
          const fs_builder abld = bld.annotate("initialize control data bits");
          abld.MOV(this->control_data_bits, brw_imm_ud(0u));
       }
@@ -6929,7 +6929,7 @@ fs_visitor::run_fs(bool allow_spilling, bool do_rep_send)
 {
    struct brw_wm_prog_data *wm_prog_data = brw_wm_prog_data(this->prog_data);
    brw_wm_prog_key *wm_key = (brw_wm_prog_key *) this->key;
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
 
    assert(stage == MESA_SHADER_FRAGMENT);
 
@@ -7005,7 +7005,7 @@ fs_visitor::run_cs(bool allow_spilling)
 {
    assert(gl_shader_stage_is_compute(stage));
    assert(devinfo->ver >= 7);
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
 
    payload_ = new cs_thread_payload(*this);
 

src/intel/compiler/brw_fs_builder.h

@@ -54,7 +54,7 @@ namespace brw {
        * Construct an fs_builder that inserts instructions into \p shader.
        * \p dispatch_width gives the native execution width of the program.
        */
-      fs_builder(backend_shader *shader,
+      fs_builder(fs_visitor *shader,
                  unsigned dispatch_width) :
          shader(shader), block(NULL), cursor(NULL),
          _dispatch_width(dispatch_width),
@@ -64,13 +64,15 @@ namespace brw {
       {
       }
 
+      explicit fs_builder(fs_visitor *s) : fs_builder(s, s->dispatch_width) {}
+
       /**
        * Construct an fs_builder that inserts instructions into \p shader
        * before instruction \p inst in basic block \p block.  The default
        * execution controls and debug annotation are initialized from the
        * instruction passed as argument.
        */
-      fs_builder(backend_shader *shader, bblock_t *block, fs_inst *inst) :
+      fs_builder(fs_visitor *shader, bblock_t *block, fs_inst *inst) :
          shader(shader), block(block), cursor(inst),
          _dispatch_width(inst->exec_size),
          _group(inst->group),
@@ -832,7 +834,7 @@ namespace brw {
          return inst;
       }
 
-      backend_shader *shader;
+      fs_visitor *shader;
 
       fs_inst *BREAK()    { return emit(BRW_OPCODE_BREAK); }
       fs_inst *DO()       { return emit(BRW_OPCODE_DO); }

src/intel/compiler/brw_fs_nir.cpp

@@ -2219,7 +2219,7 @@ fs_visitor::emit_gs_control_data_bits(const fs_reg &vertex_count)
 
    struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);
 
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
    const fs_builder abld = bld.annotate("emit control data bits");
    const fs_builder fwa_bld = bld.exec_all();
 
@@ -8470,7 +8470,7 @@ fs_visitor::emit_nir_code()
    ntb->s = this;
    ntb->devinfo = devinfo;
    ntb->nir = nir;
-   ntb->bld = fs_builder(this, this->dispatch_width).at_end();
+   ntb->bld = fs_builder(this).at_end();
 
    emit_shader_float_controls_execution_mode(ntb);
 

src/intel/compiler/brw_fs_thread_payload.cpp

@@ -104,7 +104,7 @@ gs_thread_payload::gs_thread_payload(fs_visitor &v)
 {
    struct brw_vue_prog_data *vue_prog_data = brw_vue_prog_data(v.prog_data);
    struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(v.prog_data);
-   const fs_builder bld = fs_builder(&v, v.dispatch_width).at_end();
+   const fs_builder bld = fs_builder(&v).at_end();
 
    /* R0: thread header. */
    unsigned r = reg_unit(v.devinfo);
@@ -439,7 +439,7 @@ task_mesh_thread_payload::task_mesh_thread_payload(fs_visitor &v)
     * the address to descriptors.
     */
 
-   const fs_builder bld = fs_builder(&v, v.dispatch_width).at_end();
+   const fs_builder bld = fs_builder(&v).at_end();
 
    unsigned r = 0;
    assert(subgroup_id_.file != BAD_FILE);

src/intel/compiler/brw_fs_visitor.cpp

@@ -95,7 +95,7 @@ fs_visitor::emit_interpolation_setup_gfx4()
 {
    struct brw_reg g1_uw = retype(brw_vec1_grf(1, 0), BRW_REGISTER_TYPE_UW);
 
-   fs_builder abld = fs_builder(this, dispatch_width).at_end().annotate("compute pixel centers");
+   fs_builder abld = fs_builder(this).at_end().annotate("compute pixel centers");
    this->pixel_x = vgrf(glsl_type::uint_type);
    this->pixel_y = vgrf(glsl_type::uint_type);
    this->pixel_x.type = BRW_REGISTER_TYPE_UW;
@@ -107,7 +107,7 @@ fs_visitor::emit_interpolation_setup_gfx4()
             fs_reg(stride(suboffset(g1_uw, 5), 2, 4, 0)),
             fs_reg(brw_imm_v(0x11001100)));
 
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
    abld = bld.annotate("compute pixel deltas from v0");
 
    this->delta_xy[BRW_BARYCENTRIC_PERSPECTIVE_PIXEL] =
@@ -153,7 +153,7 @@ fs_visitor::emit_interpolation_setup_gfx4()
 void
 fs_visitor::emit_interpolation_setup_gfx6()
 {
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
    fs_builder abld = bld.annotate("compute pixel centers");
 
    this->pixel_x = vgrf(glsl_type::float_type);
@@ -606,7 +606,7 @@ fs_visitor::emit_alpha_test()
 {
    assert(stage == MESA_SHADER_FRAGMENT);
    brw_wm_prog_key *key = (brw_wm_prog_key*) this->key;
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
    const fs_builder abld = bld.annotate("Alpha test");
 
    fs_inst *cmp;
@@ -680,7 +680,7 @@ fs_visitor::emit_single_fb_write(const fs_builder &bld,
 void
 fs_visitor::do_emit_fb_writes(int nr_color_regions, bool replicate_alpha)
 {
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
    fs_inst *inst = NULL;
 
    for (int target = 0; target < nr_color_regions; target++) {
@@ -815,7 +815,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
       unreachable("invalid stage");
    }
 
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
 
    fs_reg per_slot_offsets;
 
@@ -1091,7 +1091,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
 void
 fs_visitor::emit_urb_fence()
 {
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
    fs_reg dst = bld.vgrf(BRW_REGISTER_TYPE_UD);
    fs_inst *fence = bld.emit(SHADER_OPCODE_MEMORY_FENCE, dst,
                              brw_vec8_grf(0, 0),
@@ -1111,7 +1111,7 @@ void
 fs_visitor::emit_cs_terminate()
 {
    assert(devinfo->ver >= 7);
-   const fs_builder bld = fs_builder(this, dispatch_width).at_end();
+   const fs_builder bld = fs_builder(this).at_end();
 
    /* We can't directly send from g0, since sends with EOT have to use
     * g112-127. So, copy it to a virtual register, The register allocator will

src/intel/compiler/test_fs_cmod_propagation.cpp

@@ -83,7 +83,7 @@ cmod_propagation_test::cmod_propagation_test()
 
    v = new cmod_propagation_fs_visitor(compiler, &params, prog_data, shader);
 
-   bld = fs_builder(v, v->dispatch_width).at_end();
+   bld = fs_builder(v).at_end();
 
    devinfo->ver = 7;
    devinfo->verx10 = devinfo->ver * 10;

src/intel/compiler/test_fs_copy_propagation.cpp

@@ -72,7 +72,7 @@ copy_propagation_test::copy_propagation_test()
 
    v = new copy_propagation_fs_visitor(compiler, &params, prog_data, shader);
 
-   bld = fs_builder(v, v->dispatch_width).at_end();
+   bld = fs_builder(v).at_end();
 
    devinfo->ver = 4;
    devinfo->verx10 = devinfo->ver * 10;

src/intel/compiler/test_fs_saturate_propagation.cpp

@@ -72,7 +72,7 @@ saturate_propagation_test::saturate_propagation_test()
 
    v = new saturate_propagation_fs_visitor(compiler, &params, prog_data, shader);
 
-   bld = fs_builder(v, v->dispatch_width).at_end();
+   bld = fs_builder(v).at_end();
 
    devinfo->ver = 6;
    devinfo->verx10 = devinfo->ver * 10;

src/intel/compiler/test_fs_scoreboard.cpp

@@ -65,7 +65,7 @@ scoreboard_test::scoreboard_test()
    v = new fs_visitor(compiler, &params, NULL, &prog_data->base, shader, 8,
                       false, false);
 
-   bld = fs_builder(v, v->dispatch_width).at_end();
+   bld = fs_builder(v).at_end();
 }
 
 scoreboard_test::~scoreboard_test()