i965/nir: Sort uniforms direct-first and use two different uniform registers

[mesa.git] / src / mesa / drivers / dri / i965 / brw_vec4_vp.cpp

diff --git a/src/mesa/drivers/dri/i965/brw_vec4_vp.cpp b/src/mesa/drivers/dri/i965/brw_vec4_vp.cpp
index f9c23cae2bf0cbda590f645dd74b36cde8186b24..c3b0233eba223e8c65414abb2c8588aeb95d4bec 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vec4_vp.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_vp.cpp
@@ -43,8 +43,7 @@ vec4_visitor::emit_vp_sop(enum brw_conditional_mod conditional_mod,
 {
    vec4_instruction *inst;
 
-   inst = emit(BRW_OPCODE_CMP, dst_null_d(), src0, src1);
-   inst->conditional_mod = conditional_mod;
+   inst = emit(CMP(dst_null_f(), src0, src1, conditional_mod));
 
    inst = emit(BRW_OPCODE_SEL, dst, one, src_reg(0.0f));
    inst->predicate = BRW_PREDICATE_NORMAL;
@@ -227,7 +226,7 @@ vec4_vs_visitor::emit_program_code()
                /* if (tmp.y < 0) tmp.y = 0; */
                src_reg tmp_y = swizzle(src[0], BRW_SWIZZLE_YYYY);
                result.writemask = WRITEMASK_Z;
-               emit_minmax(BRW_CONDITIONAL_G, result, tmp_y, src_reg(0.0f));
+               emit_minmax(BRW_CONDITIONAL_GE, result, tmp_y, src_reg(0.0f));
 
                src_reg clamped_y(result);
                clamped_y.swizzle = BRW_SWIZZLE_ZZZZ;
@@ -314,7 +313,7 @@ vec4_vs_visitor::emit_program_code()
       }
 
       case OPCODE_MAX:
-         emit_minmax(BRW_CONDITIONAL_G, dst, src[0], src[1]);
+         emit_minmax(BRW_CONDITIONAL_GE, dst, src[0], src[1]);
          break;
 
       case OPCODE_MIN:
@@ -387,7 +386,7 @@ vec4_vs_visitor::emit_program_code()
       }
 
       /* Copy the temporary back into the actual destination register. */
-      if (vpi->Opcode != OPCODE_END) {
+      if (_mesa_num_inst_dst_regs(vpi->Opcode) != 0) {
          emit(MOV(get_vp_dst_reg(vpi->DstReg), src_reg(dst)));
       }
    }
@@ -401,7 +400,7 @@ vec4_vs_visitor::emit_program_code()
       unsigned i;
       for (i = 0; i < params->NumParameters * 4; i++) {
          stage_prog_data->pull_param[i] =
-            &params->ParameterValues[i / 4][i % 4].f;
+            &params->ParameterValues[i / 4][i % 4];
       }
       stage_prog_data->nr_pull_params = i;
    }
@@ -432,7 +431,7 @@ vec4_vs_visitor::setup_vp_regs()
       this->uniform_vector_size[this->uniforms] = components;
       for (unsigned i = 0; i < 4; i++) {
          stage_prog_data->param[this->uniforms * 4 + i] = i >= components
-            ? 0 : &plist->ParameterValues[p][i].f;
+            ? 0 : &plist->ParameterValues[p][i];
       }
       this->uniforms++; /* counted in vec4 units */
    }
@@ -528,6 +527,15 @@ vec4_vs_visitor::get_vp_src_reg(const prog_src_register &src)
 
          /* Add the small constant index to the address register */
          src_reg reladdr = src_reg(this, glsl_type::int_type);
+
+         /* We have to use a message header on Skylake to get SIMD4x2 mode.
+          * Reserve space for the register.
+          */
+         if (brw->gen >= 9) {
+            reladdr.reg_offset++;
+            alloc.sizes[reladdr.reg] = 2;
+         }
+
          dst_reg dst_reladdr = dst_reg(reladdr);
          dst_reladdr.writemask = WRITEMASK_X;
          emit(ADD(dst_reladdr, this->vp_addr_reg, src_reg(src.Index)));
@@ -545,11 +553,19 @@ vec4_vs_visitor::get_vp_src_reg(const prog_src_register &src)
 
          result = src_reg(this, glsl_type::vec4_type);
          src_reg surf_index = src_reg(unsigned(prog_data->base.binding_table.pull_constants_start));
-         vec4_instruction *load =
-            new(mem_ctx) vec4_instruction(this, VS_OPCODE_PULL_CONSTANT_LOAD,
-                                          dst_reg(result), surf_index, reladdr);
-         load->base_mrf = 14;
-         load->mlen = 1;
+         vec4_instruction *load;
+         if (brw->gen >= 7) {
+            load = new(mem_ctx)
+               vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD_GEN7,
+                                dst_reg(result), surf_index, reladdr);
+            load->mlen = 1;
+         } else {
+            load = new(mem_ctx)
+               vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD,
+                                dst_reg(result), surf_index, reladdr);
+            load->base_mrf = 14;
+            load->mlen = 1;
+         }
          emit(load);
          break;
       }