X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fvc4%2Fvc4_opt_algebraic.c;h=5e7d26923de822921cf39ff567cb0a79f3cc4b36;hb=62e36ba43860fdbfb1a1058c64d0ec71734f7b0a;hp=2bf474ccef9cfe23cda8d2c0390cd48224f0fab2;hpb=b0a1e401a93b7b13870b936bc667b3fc15dba6d5;p=mesa.git

diff --git a/src/gallium/drivers/vc4/vc4_opt_algebraic.c b/src/gallium/drivers/vc4/vc4_opt_algebraic.c
index 2bf474ccef9..5e7d26923de 100644
--- a/src/gallium/drivers/vc4/vc4_opt_algebraic.c
+++ b/src/gallium/drivers/vc4/vc4_opt_algebraic.c
@@ -33,40 +33,253 @@
  */
 
 #include "vc4_qir.h"
+#include "util/u_math.h"
+
+static bool debug;
+
+static void
+dump_from(struct vc4_compile *c, struct qinst *inst)
+{
+        if (!debug)
+                return;
+
+        fprintf(stderr, "optimizing: ");
+        qir_dump_inst(c, inst);
+        fprintf(stderr, "\n");
+}
+
+static void
+dump_to(struct vc4_compile *c, struct qinst *inst)
+{
+        if (!debug)
+                return;
+
+        fprintf(stderr, "to: ");
+        qir_dump_inst(c, inst);
+        fprintf(stderr, "\n");
+}
+
+static bool
+is_constant_value(struct vc4_compile *c, struct qreg reg,
+                  uint32_t val)
+{
+        if (reg.file == QFILE_UNIF &&
+            !reg.pack &&
+            c->uniform_contents[reg.index] == QUNIFORM_CONSTANT &&
+            c->uniform_data[reg.index] == val) {
+                return true;
+        }
+
+        if (reg.file == QFILE_SMALL_IMM && reg.index == val)
+                return true;
+
+        return false;
+}
+
+static bool
+is_zero(struct vc4_compile *c, struct qreg reg)
+{
+        reg = qir_follow_movs(c, reg);
+        return is_constant_value(c, reg, 0);
+}
+
+static bool
+is_1f(struct vc4_compile *c, struct qreg reg)
+{
+        reg = qir_follow_movs(c, reg);
+        return is_constant_value(c, reg, fui(1.0));
+}
+
+static void
+replace_with_mov(struct vc4_compile *c, struct qinst *inst, struct qreg arg)
+{
+        dump_from(c, inst);
+
+        inst->src[0] = arg;
+        if (qir_has_implicit_tex_uniform(inst))
+                inst->src[1] = inst->src[qir_get_tex_uniform_src(inst)];
+
+        if (qir_is_mul(inst))
+                inst->op = QOP_MMOV;
+        else if (qir_is_float_input(inst))
+                inst->op = QOP_FMOV;
+        else
+                inst->op = QOP_MOV;
+        dump_to(c, inst);
+}
+
+static bool
+replace_x_0_with_x(struct vc4_compile *c,
+                 struct qinst *inst,
+                 int arg)
+{
+        if (!is_zero(c, inst->src[arg]))
+                return false;
+        replace_with_mov(c, inst, inst->src[1 - arg]);
+        return true;
+}
+
+static bool
+replace_x_0_with_0(struct vc4_compile *c,
+                  struct qinst *inst,
+                  int arg)
+{
+        if (!is_zero(c, inst->src[arg]))
+                return false;
+        replace_with_mov(c, inst, inst->src[arg]);
+        return true;
+}
+
+static bool
+fmul_replace_one(struct vc4_compile *c,
+                 struct qinst *inst,
+                 int arg)
+{
+        if (!is_1f(c, inst->src[arg]))
+                return false;
+        replace_with_mov(c, inst, inst->src[1 - arg]);
+        return true;
+}
 
 bool
-qir_opt_algebraic(struct qcompile *c)
+qir_opt_algebraic(struct vc4_compile *c)
 {
         bool progress = false;
-        struct simple_node *node;
-        bool debug = false;
-
-        foreach(node, &c->instructions) {
-                struct qinst *inst = (struct qinst *)node;
 
+        qir_for_each_inst_inorder(inst, c) {
                 switch (inst->op) {
-                case QOP_CMP:
-                        /* Turn "dst = (a < 0) ? b : b)" into "dst = b" */
-                        if (qir_reg_equals(inst->src[1], inst->src[2])) {
-                                if (debug) {
-                                        fprintf(stderr, "optimizing: ");
-                                        qir_dump_inst(inst);
-                                        fprintf(stderr, "\n");
-                                }
+                case QOP_FMIN:
+                        if (is_1f(c, inst->src[1]) &&
+                            inst->src[0].pack >= QPU_UNPACK_8D_REP &&
+                            inst->src[0].pack <= QPU_UNPACK_8D) {
+                                replace_with_mov(c, inst, inst->src[0]);
+                                progress = true;
+                        }
+                        break;
 
-                                inst->op = QOP_MOV;
-                                inst->src[0] = inst->src[1];
-                                inst->src[1] = c->undef;
+                case QOP_FMAX:
+                        if (is_zero(c, inst->src[1]) &&
+                            inst->src[0].pack >= QPU_UNPACK_8D_REP &&
+                            inst->src[0].pack <= QPU_UNPACK_8D) {
+                                replace_with_mov(c, inst, inst->src[0]);
                                 progress = true;
+                        }
+                        break;
+
+                case QOP_FSUB:
+                case QOP_SUB:
+                        if (is_zero(c, inst->src[1])) {
+                                replace_with_mov(c, inst, inst->src[0]);
+                                progress = true;
+                        }
+                        break;
+
+                case QOP_ADD:
+                        /* Kernel validation requires that we use an actual
+                         * add instruction.
+                         */
+                        if (inst->dst.file != QFILE_TEX_S_DIRECT &&
+                            (replace_x_0_with_x(c, inst, 0) ||
+                             replace_x_0_with_x(c, inst, 1))) {
+                                progress = true;
+                                break;
+                        }
+                        break;
+
+                case QOP_FADD:
+                        if (replace_x_0_with_x(c, inst, 0) ||
+                            replace_x_0_with_x(c, inst, 1)) {
+                                progress = true;
+                                break;
+                        }
+
+                        /* FADD(a, FSUB(0, b)) -> FSUB(a, b) */
+                        if (inst->src[1].file == QFILE_TEMP &&
+                            c->defs[inst->src[1].index] &&
+                            c->defs[inst->src[1].index]->op == QOP_FSUB) {
+                                struct qinst *fsub = c->defs[inst->src[1].index];
+                                if (is_zero(c, fsub->src[0])) {
+                                        dump_from(c, inst);
+                                        inst->op = QOP_FSUB;
+                                        inst->src[1] = fsub->src[1];
+                                        progress = true;
+                                        dump_to(c, inst);
+                                        break;
+                                }
+                        }
 
-                                if (debug) {
-                                        fprintf(stderr, "to: ");
-                                        qir_dump_inst(inst);
-                                        fprintf(stderr, "\n");
+                        /* FADD(FSUB(0, b), a) -> FSUB(a, b) */
+                        if (inst->src[0].file == QFILE_TEMP &&
+                            c->defs[inst->src[0].index] &&
+                            c->defs[inst->src[0].index]->op == QOP_FSUB) {
+                                struct qinst *fsub = c->defs[inst->src[0].index];
+                                if (is_zero(c, fsub->src[0])) {
+                                        dump_from(c, inst);
+                                        inst->op = QOP_FSUB;
+                                        inst->src[0] = inst->src[1];
+                                        inst->src[1] = fsub->src[1];
+                                        dump_to(c, inst);
+                                        progress = true;
+                                        break;
                                 }
                         }
                         break;
 
+                case QOP_FMUL:
+                        if (!inst->dst.pack &&
+                            (replace_x_0_with_0(c, inst, 0) ||
+                             replace_x_0_with_0(c, inst, 1) ||
+                             fmul_replace_one(c, inst, 0) ||
+                             fmul_replace_one(c, inst, 1))) {
+                                progress = true;
+                                break;
+                        }
+                        break;
+
+                case QOP_MUL24:
+                        if (!inst->dst.pack &&
+                            (replace_x_0_with_0(c, inst, 0) ||
+                             replace_x_0_with_0(c, inst, 1))) {
+                                progress = true;
+                                break;
+                        }
+                        break;
+
+                case QOP_AND:
+                        if (replace_x_0_with_0(c, inst, 0) ||
+                            replace_x_0_with_0(c, inst, 1)) {
+                                progress = true;
+                                break;
+                        }
+
+                        if (is_constant_value(c, inst->src[0], ~0)) {
+                                replace_with_mov(c, inst, inst->src[1]);
+                                progress = true;
+                                break;
+                        }
+                        if (is_constant_value(c, inst->src[1], ~0)) {
+                                replace_with_mov(c, inst, inst->src[0]);
+                                progress = true;
+                                break;
+                        }
+                        break;
+
+                case QOP_OR:
+                        if (replace_x_0_with_x(c, inst, 0) ||
+                            replace_x_0_with_x(c, inst, 1)) {
+                                progress = true;
+                                break;
+                        }
+                        break;
+
+                case QOP_RCP:
+                        if (is_1f(c, inst->src[0])) {
+                                replace_with_mov(c, inst, inst->src[0]);
+                                progress = true;
+                                break;
+                        }
+                        break;
+
                 default:
                         break;
                 }