[Mesa-dev] [PATCH 4/5] glsl_to_tgsi: use TEX_LZ and TXF_LZ when available

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

---
 src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)

diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index 530007e..c38e705 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -425,20 +425,21 @@ public:
    int image_targets[PIPE_MAX_SHADER_IMAGES];
    unsigned image_formats[PIPE_MAX_SHADER_IMAGES];
    bool indirect_addr_consts;
    int wpos_transform_const;
 
    int glsl_version;
    bool native_integers;
    bool have_sqrt;
    bool have_fma;
    bool use_shared_memory;
+   bool has_tex_txf_lz;
 
    variable_storage *find_variable_storage(ir_variable *var);
 
    int add_constant(gl_register_file file, gl_constant_value values[8],
                     int size, int datatype, uint16_t *swizzle_out);
 
    st_src_reg get_temp(const glsl_type *type);
    void reladdr_to_temp(ir_instruction *ir, st_src_reg *reg, int *num_reladdr);
 
    st_src_reg st_src_reg_for_double(double val);
@@ -4183,23 +4184,27 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
          opcode = TGSI_OPCODE_TXB;
       }
       ir->lod_info.bias->accept(this);
       lod_info = this->result;
       if (ir->offset) {
          ir->offset->accept(this);
          offset[0] = this->result;
       }
       break;
    case ir_txl:
-      opcode = is_cube_array ? TGSI_OPCODE_TXL2 : TGSI_OPCODE_TXL;
-      ir->lod_info.lod->accept(this);
-      lod_info = this->result;
+      if (this->has_tex_txf_lz && ir->lod_info.lod->is_zero()) {
+         opcode = TGSI_OPCODE_TEX_LZ;
+      } else {
+         opcode = is_cube_array ? TGSI_OPCODE_TXL2 : TGSI_OPCODE_TXL;
+         ir->lod_info.lod->accept(this);
+         lod_info = this->result;
+      }
       if (ir->offset) {
          ir->offset->accept(this);
          offset[0] = this->result;
       }
       break;
    case ir_txd:
       opcode = TGSI_OPCODE_TXD;
       ir->lod_info.grad.dPdx->accept(this);
       dx = this->result;
       ir->lod_info.grad.dPdy->accept(this);
@@ -4213,23 +4218,27 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
       opcode = TGSI_OPCODE_TXQ;
       ir->lod_info.lod->accept(this);
       lod_info = this->result;
       break;
    case ir_query_levels:
       opcode = TGSI_OPCODE_TXQ;
       lod_info = undef_src;
       levels_src = get_temp(ir->type);
       break;
    case ir_txf:
-      opcode = TGSI_OPCODE_TXF;
-      ir->lod_info.lod->accept(this);
-      lod_info = this->result;
+      if (this->has_tex_txf_lz && ir->lod_info.lod->is_zero()) {
+         opcode = TGSI_OPCODE_TXF_LZ;
+      } else {
+         opcode = TGSI_OPCODE_TXF;
+         ir->lod_info.lod->accept(this);
+         lod_info = this->result;
+      }
       if (ir->offset) {
          ir->offset->accept(this);
          offset[0] = this->result;
       }
       break;
    case ir_txf_ms:
       opcode = TGSI_OPCODE_TXF;
       ir->lod_info.sample_index->accept(this);
       sample_index = this->result;
       break;
@@ -4552,20 +4561,21 @@ glsl_to_tgsi_visitor::glsl_to_tgsi_visitor()
    native_integers = false;
    mem_ctx = ralloc_context(NULL);
    ctx = NULL;
    prog = NULL;
    shader_program = NULL;
    shader = NULL;
    options = NULL;
    have_sqrt = false;
    have_fma = false;
    use_shared_memory = false;
+   has_tex_txf_lz = false;
 }
 
 glsl_to_tgsi_visitor::~glsl_to_tgsi_visitor()
 {
    free(array_sizes);
    ralloc_free(mem_ctx);
 }
 
 extern "C" void free_glsl_to_tgsi_visitor(glsl_to_tgsi_visitor *v)
 {
@@ -5797,27 +5807,29 @@ compile_tgsi_instruction(struct st_translate *t,
    case TGSI_OPCODE_BGNLOOP:
    case TGSI_OPCODE_ELSE:
    case TGSI_OPCODE_ENDLOOP:
    case TGSI_OPCODE_IF:
    case TGSI_OPCODE_UIF:
       assert(num_dst == 0);
       ureg_insn(ureg, inst->op, NULL, 0, src, num_src);
       return;
 
    case TGSI_OPCODE_TEX:
+   case TGSI_OPCODE_TEX_LZ:
    case TGSI_OPCODE_TXB:
    case TGSI_OPCODE_TXD:
    case TGSI_OPCODE_TXL:
    case TGSI_OPCODE_TXP:
    case TGSI_OPCODE_TXQ:
    case TGSI_OPCODE_TXQS:
    case TGSI_OPCODE_TXF:
+   case TGSI_OPCODE_TXF_LZ:
    case TGSI_OPCODE_TEX2:
    case TGSI_OPCODE_TXB2:
    case TGSI_OPCODE_TXL2:
    case TGSI_OPCODE_TG4:
    case TGSI_OPCODE_LODQ:
       src[num_src] = t->samplers[inst->resource.index];
       assert(src[num_src].File != TGSI_FILE_NULL);
       if (inst->resource.reladdr)
          src[num_src] =
             ureg_src_indirect(src[num_src], ureg_src(t->address[2]));
@@ -6657,20 +6669,22 @@ get_mesa_program_tgsi(struct gl_context *ctx,
    v->shader_program = shader_program;
    v->shader = shader;
    v->options = options;
    v->glsl_version = ctx->Const.GLSLVersion;
    v->native_integers = ctx->Const.NativeIntegers;
 
    v->have_sqrt = pscreen->get_shader_param(pscreen, ptarget,
                                             PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED);
    v->have_fma = pscreen->get_shader_param(pscreen, ptarget,
                                            PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED);
+   v->has_tex_txf_lz = pscreen->get_param(pscreen,
+                                          PIPE_CAP_TGSI_TEX_TXF_LZ);
 
    _mesa_generate_parameters_list_for_uniforms(shader_program, shader,
                                                prog->Parameters);
 
    /* Remove reads from output registers. */
    if (!pscreen->get_param(pscreen, PIPE_CAP_TGSI_CAN_READ_OUTPUTS))
       lower_output_reads(shader->Stage, shader->ir);
 
    /* Emit intermediate IR for main(). */
    visit_exec_list(shader->ir, v);
-- 
2.7.4