i965/fs: Allow saturate propagation to propagate negations into MULs.

Allows us to transform

   mul      res  src0   src1
   mov.sat  dst  -res

into

   mul.sat  dst  src0  -src1

instructions in affected programs: 45246 -> 45054 (-0.42%)
helped: 162

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

diff --git a/src/mesa/drivers/dri/i965/brw_fs_saturate_propagation.cpp b/src/mesa/drivers/dri/i965/brw_fs_saturate_propagation.cpp
index 525709439968b795468756e520b687420bd78752..b9ad6fb8bc83997a4e99d57412934c5fa3047301 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_saturate_propagation.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_saturate_propagation.cpp
@@ -56,8 +56,7 @@ opt_saturate_propagation_local(fs_visitor *v, bblock_t *block)
           inst->dst.file != VGRF ||
           inst->dst.type != inst->src[0].type ||
           inst->src[0].file != VGRF ||
-          inst->src[0].abs ||
-          inst->src[0].negate)
+          inst->src[0].abs)
          continue;
 
       int src_var = v->live_intervals->var_from_reg(inst->src[0]);
@@ -82,6 +81,16 @@ opt_saturate_propagation_local(fs_visitor *v, bblock_t *block)
                         scan_inst->src[i].type = inst->dst.type;
                      }
                   }
+
+                  if (inst->src[0].negate) {
+                     if (scan_inst->opcode == BRW_OPCODE_MUL) {
+                        scan_inst->src[0].negate = !scan_inst->src[0].negate;
+                        inst->src[0].negate = false;
+                     } else {
+                        break;
+                     }
+                  }
+
                   scan_inst->saturate = true;
                   inst->saturate = false;
                   progress = true;
@@ -96,7 +105,9 @@ opt_saturate_propagation_local(fs_visitor *v, bblock_t *block)
                if (scan_inst->opcode != BRW_OPCODE_MOV ||
                    !scan_inst->saturate ||
                    scan_inst->src[0].abs ||
-                   scan_inst->src[0].negate) {
+                   scan_inst->src[0].negate ||
+                   scan_inst->src[0].abs != inst->src[0].abs ||
+                   scan_inst->src[0].negate != inst->src[0].negate) {
                   interfered = true;
                   break;
                }

diff --git a/src/mesa/drivers/dri/i965/test_fs_saturate_propagation.cpp b/src/mesa/drivers/dri/i965/test_fs_saturate_propagation.cpp
index 32e8b8f88673a63a0df6da4dabe8f57f81172148..30c14724ff0f1ab2a7bc70bff0c2d89149ab011b 100644 (file)
--- a/src/mesa/drivers/dri/i965/test_fs_saturate_propagation.cpp
+++ b/src/mesa/drivers/dri/i965/test_fs_saturate_propagation.cpp
@@ -244,6 +244,129 @@ TEST_F(saturate_propagation_test, neg_mov_sat)
    EXPECT_TRUE(instruction(block0, 1)->saturate);
 }
 
+TEST_F(saturate_propagation_test, mul_neg_mov_sat)
+{
+   const fs_builder &bld = v->bld;
+   fs_reg dst0 = v->vgrf(glsl_type::float_type);
+   fs_reg dst1 = v->vgrf(glsl_type::float_type);
+   fs_reg src0 = v->vgrf(glsl_type::float_type);
+   fs_reg src1 = v->vgrf(glsl_type::float_type);
+   bld.MUL(dst0, src0, src1);
+   dst0.negate = true;
+   set_saturate(true, bld.MOV(dst1, dst0));
+
+   /* = Before =
+    *
+    * 0: mul(8)        dst0  src0  src1
+    * 1: mov.sat(8)    dst1  -dst0
+    *
+    * = After =
+    * 0: mul.sat(8)    dst0  src0 -src1
+    * 1: mov(8)        dst1  dst0
+    */
+
+   v->calculate_cfg();
+   bblock_t *block0 = v->cfg->blocks[0];
+
+   EXPECT_EQ(0, block0->start_ip);
+   EXPECT_EQ(1, block0->end_ip);
+
+   EXPECT_TRUE(saturate_propagation(v));
+   EXPECT_EQ(0, block0->start_ip);
+   EXPECT_EQ(1, block0->end_ip);
+   EXPECT_EQ(BRW_OPCODE_MUL, instruction(block0, 0)->opcode);
+   EXPECT_TRUE(instruction(block0, 0)->saturate);
+   EXPECT_TRUE(instruction(block0, 0)->src[0].negate);
+   EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
+   EXPECT_FALSE(instruction(block0, 1)->saturate);
+   EXPECT_FALSE(instruction(block0, 1)->src[0].negate);
+}
+
+TEST_F(saturate_propagation_test, mul_mov_sat_neg_mov_sat)
+{
+   const fs_builder &bld = v->bld;
+   fs_reg dst0 = v->vgrf(glsl_type::float_type);
+   fs_reg dst1 = v->vgrf(glsl_type::float_type);
+   fs_reg dst2 = v->vgrf(glsl_type::float_type);
+   fs_reg src0 = v->vgrf(glsl_type::float_type);
+   fs_reg src1 = v->vgrf(glsl_type::float_type);
+   bld.MUL(dst0, src0, src1);
+   set_saturate(true, bld.MOV(dst1, dst0));
+   dst0.negate = true;
+   set_saturate(true, bld.MOV(dst2, dst0));
+
+   /* = Before =
+    *
+    * 0: mul(8)        dst0  src0  src1
+    * 1: mov.sat(8)    dst1  dst0
+    * 2: mov.sat(8)    dst2  -dst0
+    *
+    * = After =
+    * (no changes)
+    */
+
+   v->calculate_cfg();
+   bblock_t *block0 = v->cfg->blocks[0];
+
+   EXPECT_EQ(0, block0->start_ip);
+   EXPECT_EQ(2, block0->end_ip);
+
+   EXPECT_FALSE(saturate_propagation(v));
+   EXPECT_EQ(0, block0->start_ip);
+   EXPECT_EQ(2, block0->end_ip);
+   EXPECT_EQ(BRW_OPCODE_MUL, instruction(block0, 0)->opcode);
+   EXPECT_FALSE(instruction(block0, 0)->saturate);
+   EXPECT_FALSE(instruction(block0, 0)->src[1].negate);
+   EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
+   EXPECT_TRUE(instruction(block0, 1)->saturate);
+   EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 2)->opcode);
+   EXPECT_TRUE(instruction(block0, 2)->src[0].negate);
+   EXPECT_TRUE(instruction(block0, 2)->saturate);
+}
+
+TEST_F(saturate_propagation_test, mul_neg_mov_sat_neg_mov_sat)
+{
+   const fs_builder &bld = v->bld;
+   fs_reg dst0 = v->vgrf(glsl_type::float_type);
+   fs_reg dst1 = v->vgrf(glsl_type::float_type);
+   fs_reg dst2 = v->vgrf(glsl_type::float_type);
+   fs_reg src0 = v->vgrf(glsl_type::float_type);
+   fs_reg src1 = v->vgrf(glsl_type::float_type);
+   bld.MUL(dst0, src0, src1);
+   dst0.negate = true;
+   set_saturate(true, bld.MOV(dst1, dst0));
+   set_saturate(true, bld.MOV(dst2, dst0));
+
+   /* = Before =
+    *
+    * 0: mul(8)        dst0  src0  src1
+    * 1: mov.sat(8)    dst1  -dst0
+    * 2: mov.sat(8)    dst2  -dst0
+    *
+    * = After =
+    * (no changes)
+    */
+
+   v->calculate_cfg();
+   bblock_t *block0 = v->cfg->blocks[0];
+
+   EXPECT_EQ(0, block0->start_ip);
+   EXPECT_EQ(2, block0->end_ip);
+
+   EXPECT_FALSE(saturate_propagation(v));
+   EXPECT_EQ(0, block0->start_ip);
+   EXPECT_EQ(2, block0->end_ip);
+   EXPECT_EQ(BRW_OPCODE_MUL, instruction(block0, 0)->opcode);
+   EXPECT_FALSE(instruction(block0, 0)->saturate);
+   EXPECT_FALSE(instruction(block0, 0)->src[1].negate);
+   EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
+   EXPECT_TRUE(instruction(block0, 1)->src[0].negate);
+   EXPECT_TRUE(instruction(block0, 1)->saturate);
+   EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 2)->opcode);
+   EXPECT_TRUE(instruction(block0, 2)->src[0].negate);
+   EXPECT_TRUE(instruction(block0, 2)->saturate);
+}
+
 TEST_F(saturate_propagation_test, abs_mov_sat)
 {
    const fs_builder &bld = v->bld;