[Mesa-dev] [PATCH 09/11] st/glsl_to_tgsi: use LDEXP when available

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle at amd.com>

---
 src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index cf6e8f8fde1..196b547d3d9 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -576,24 +576,24 @@ glsl_to_tgsi_visitor::emit_asm(ir_instruction *ir, unsigned op,
                   dinst->src[j].index++;
                }
 
                if (swz & 1)
                   dinst->src[j].swizzle = MAKE_SWIZZLE4(SWIZZLE_Z, SWIZZLE_W, SWIZZLE_Z, SWIZZLE_W);
                else
                   dinst->src[j].swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_X, SWIZZLE_Y);
 
             } else {
                /* some opcodes are special case in what they use as sources
-                  - [FUI]2D/[UI]2I64 is a float/[u]int src0, DLDEXP is integer src1 */
+                  - [FUI]2D/[UI]2I64 is a float/[u]int src0, (D)LDEXP is integer src1 */
                if (op == TGSI_OPCODE_F2D || op == TGSI_OPCODE_U2D || op == TGSI_OPCODE_I2D ||
                    op == TGSI_OPCODE_I2I64 || op == TGSI_OPCODE_U2I64 ||
-                   op == TGSI_OPCODE_DLDEXP ||
+                   op == TGSI_OPCODE_DLDEXP || op == TGSI_OPCODE_LDEXP ||
                    (op == TGSI_OPCODE_UCMP && dst_is_64bit[0])) {
                   dinst->src[j].swizzle = MAKE_SWIZZLE4(swz, swz, swz, swz);
                }
             }
          }
       }
       inst = dinst;
    } else {
       this->instructions.push_tail(inst);
    }
@@ -2025,20 +2025,22 @@ glsl_to_tgsi_visitor::visit_expression(ir_expression* ir, st_src_reg *op)
    case ir_unop_unpack_sampler_2x32:
    case ir_unop_pack_sampler_2x32:
    case ir_unop_unpack_image_2x32:
    case ir_unop_pack_image_2x32:
       emit_asm(ir, TGSI_OPCODE_MOV, result_dst, op[0]);
       break;
 
    case ir_binop_ldexp:
       if (ir->operands[0]->type->is_double()) {
          emit_asm(ir, TGSI_OPCODE_DLDEXP, result_dst, op[0], op[1]);
+      } else if (ir->operands[0]->type->is_float()) {
+         emit_asm(ir, TGSI_OPCODE_LDEXP, result_dst, op[0], op[1]);
       } else {
          assert(!"Invalid ldexp for non-double opcode in glsl_to_tgsi_visitor::visit()");
       }
       break;
 
    case ir_unop_pack_half_2x16:
       emit_asm(ir, TGSI_OPCODE_PK2H, result_dst, op[0]);
       break;
    case ir_unop_unpack_half_2x16:
       emit_asm(ir, TGSI_OPCODE_UP2H, result_dst, op[0]);
@@ -6704,20 +6706,22 @@ st_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
       struct gl_linked_shader *shader = prog->_LinkedShaders[i];
       exec_list *ir = shader->ir;
       gl_shader_stage stage = shader->Stage;
       const struct gl_shader_compiler_options *options =
             &ctx->Const.ShaderCompilerOptions[stage];
       enum pipe_shader_type ptarget = pipe_shader_type_from_mesa(stage);
       bool have_dround = pscreen->get_shader_param(pscreen, ptarget,
                                                    PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED);
       bool have_dfrexp = pscreen->get_shader_param(pscreen, ptarget,
                                                    PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED);
+      bool have_ldexp = pscreen->get_shader_param(pscreen, ptarget,
+                                                  PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED);
       unsigned if_threshold = pscreen->get_shader_param(pscreen, ptarget,
                                                         PIPE_SHADER_CAP_LOWER_IF_THRESHOLD);
 
       /* If there are forms of indirect addressing that the driver
        * cannot handle, perform the lowering pass.
        */
       if (options->EmitNoIndirectInput || options->EmitNoIndirectOutput ||
           options->EmitNoIndirectTemp || options->EmitNoIndirectUniform) {
          lower_variable_index_to_cond_assign(stage, ir,
                                              options->EmitNoIndirectInput,
@@ -6754,21 +6758,21 @@ st_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
       do_mat_op_to_vec(ir);
 
       if (stage == MESA_SHADER_FRAGMENT)
          lower_blend_equation_advanced(shader);
 
       lower_instructions(ir,
                          MOD_TO_FLOOR |
                          FDIV_TO_MUL_RCP |
                          EXP_TO_EXP2 |
                          LOG_TO_LOG2 |
-                         LDEXP_TO_ARITH |
+                         (have_ldexp ? 0 : LDEXP_TO_ARITH) |
                          (have_dfrexp ? 0 : DFREXP_DLDEXP_TO_ARITH) |
                          CARRY_TO_ARITH |
                          BORROW_TO_ARITH |
                          (have_dround ? 0 : DOPS_TO_DFRAC) |
                          (options->EmitNoPow ? POW_TO_EXP2 : 0) |
                          (!ctx->Const.NativeIntegers ? INT_DIV_TO_MUL_RCP : 0) |
                          (options->EmitNoSat ? SAT_TO_CLAMP : 0) |
                          (ctx->Const.ForceGLSLAbsSqrt ? SQRT_TO_ABS_SQRT : 0) |
                          /* Assume that if ARB_gpu_shader5 is not supported
                           * then all of the extended integer functions need
-- 
2.11.0