i965: Use greater-equal cmod to implement maximum.
authorMatt Turner <mattst88@gmail.com>
Tue, 10 Feb 2015 05:11:46 +0000 (21:11 -0800)
committerMatt Turner <mattst88@gmail.com>
Fri, 20 Feb 2015 05:16:43 +0000 (21:16 -0800)
The docs specifically call out SEL with .l and .ge as the
implementations of MIN and MAX respectively. Among other things, SEL
with these conditional mods are commutative.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
src/mesa/drivers/dri/i965/brw_vec4_vp.cpp
src/mesa/drivers/dri/i965/brw_vec4_vs_visitor.cpp

index 74860711abd96d844080bbb730688f60e4965d87..4b48f2db0f212cde016b7da665ef132b3aa4da98 100644 (file)
@@ -321,6 +321,9 @@ void
 fs_visitor::emit_minmax(enum brw_conditional_mod conditionalmod, const fs_reg &dst,
                         const fs_reg &src0, const fs_reg &src1)
 {
+   assert(conditionalmod == BRW_CONDITIONAL_GE ||
+          conditionalmod == BRW_CONDITIONAL_L);
+
    fs_inst *inst;
 
    if (brw->gen >= 6) {
@@ -1956,7 +1959,7 @@ fs_visitor::rescale_texcoord(fs_reg coordinate, int coord_components,
            chan = offset(chan, i);
 
            inst = emit(BRW_OPCODE_SEL, chan, chan, fs_reg(0.0f));
-           inst->conditional_mod = BRW_CONDITIONAL_G;
+           inst->conditional_mod = BRW_CONDITIONAL_GE;
 
            /* Our parameter comes in as 1.0/width or 1.0/height,
             * because that's what people normally want for doing
index 6154e43c1a4deabcb073625610074b8c900a8874..be071d74baa2ed1a0db28e3357e804ab57b4f6a0 100644 (file)
@@ -517,7 +517,7 @@ vec4_visitor::emit_unpack_snorm_4x8(const dst_reg &dst, src_reg src0)
    emit(MUL(scaled, src_reg(f), src_reg(1.0f / 127.0f)));
 
    dst_reg max(this, glsl_type::vec4_type);
-   emit_minmax(BRW_CONDITIONAL_G, max, src_reg(scaled), src_reg(-1.0f));
+   emit_minmax(BRW_CONDITIONAL_GE, max, src_reg(scaled), src_reg(-1.0f));
    emit_minmax(BRW_CONDITIONAL_L, dst, src_reg(max), src_reg(1.0f));
 }
 
@@ -545,7 +545,7 @@ void
 vec4_visitor::emit_pack_snorm_4x8(const dst_reg &dst, const src_reg &src0)
 {
    dst_reg max(this, glsl_type::vec4_type);
-   emit_minmax(BRW_CONDITIONAL_G, max, src0, src_reg(-1.0f));
+   emit_minmax(BRW_CONDITIONAL_GE, max, src0, src_reg(-1.0f));
 
    dst_reg min(this, glsl_type::vec4_type);
    emit_minmax(BRW_CONDITIONAL_L, min, src_reg(max), src_reg(1.0f));
@@ -1683,7 +1683,7 @@ vec4_visitor::visit(ir_expression *ir)
       emit_minmax(BRW_CONDITIONAL_L, result_dst, op[0], op[1]);
       break;
    case ir_binop_max:
-      emit_minmax(BRW_CONDITIONAL_G, result_dst, op[0], op[1]);
+      emit_minmax(BRW_CONDITIONAL_GE, result_dst, op[0], op[1]);
       break;
 
    case ir_binop_pow:
index a5a5e7bfd5bfa357f913d36bfa28e989b57d10b4..ba3264dce6fa29b357e867678ef752ed609672ce 100644 (file)
@@ -226,7 +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;
@@ -313,7 +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:
index 129c2db638bf535bc0d0af21c2b9c5701f53fdd7..4baf73ebde17e6eb373a3d377a8b422c31200607 100644 (file)
@@ -97,7 +97,7 @@ vec4_vs_visitor::emit_prolog()
                dst.type = brw_type_for_base_type(glsl_type::vec4_type);
                emit(MOV(dst, src_reg(reg_d)));
                emit(MUL(dst, src_reg(dst), src_reg(es3_normalize_factor)));
-               emit_minmax(BRW_CONDITIONAL_G, dst, src_reg(dst), src_reg(-1.0f));
+               emit_minmax(BRW_CONDITIONAL_GE, dst, src_reg(dst), src_reg(-1.0f));
             } else {
                /* The following equations are from the OpenGL 3.2 specification:
                 *