target.h (builtin_vectorization_cost): Add new target builtin.

2007-07-12  Dorit Nuzman  <dorit@il.ibm.com>

        * target.h (builtin_vectorization_cost): Add new target builtin.
        * target-def.h (TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST): New.
        * tree-vectorizer.h (TARG_SCALAR_STMT_COST): New.
        (TARG_SCALAR_LOAD_COST, TARG_SCALAR_STORE_COST): New.
        * tree-vect-analyze.c (vect_analyze_slp_instance): Initisliaze
        uninitialized variables.
        * tree-vect-transform.c (cost_for_stmt): New function.
        (vect_estimate_min_profitable_iters): Call cost_for_stmt instead of
        using cost 1 for all scalar stmts. Be less conservative when
        estimating the number of prologue/epulogue iterations. Call
        targetm.vectorize.builtin_vectorization_cost. Return
        min_profitable_iters-1.
        (vect_model_reduction_cost): Use TARG_SCALAR_TO_VEC_COST for
        initialization cost instead of TARG_VEC_STMT_COST. Use
        TARG_VEC_TO_SCALAR_COST instead of TARG_VEC_STMT_COST for reduction
        epilogue code. Fix epilogue cost computation.
        * config/spu/spu.c (spu_builtin_vectorization_cost): New.
        (TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST): Implement.
        * config/spu/spu.h (TARG_COND_BRANCH_COST, TARG_SCALAR_STMT_COST):
        (TARG_SCALAR_LOAD_COST, TARG_SCALAR_STORE_COST, TARG_VEC_STMT_COST):
        (TARG_VEC_TO_SCALAR_COST, TARG_SCALAR_TO_VEC, TARG_VEC_LOAD_COST):
        (TARG_VEC_UNALIGNED_LOAD_COST, TARG_VEC_STORE_COST): Define.

2007-07-12  Dorit Nuzman  <dorit@il.ibm.com>

        * gcc.dg/vect/costmodel/ppc/costmodel-vect-reduc-1char.c: Loops now
        get vectorized.
        * gcc.dg/vect/costmodel/i386/costmodel-vect-reduc-1char.c: Loops
        now get vectorized.
        * gcc.dg/vect/costmodel/spu/spu-costmodel-vect.exp: New.
        * gcc.dg/vect/costmodel/spu/costmodel-fast-math-vect-pr29925.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-31a.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-31b.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-31c.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-31d.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-iv-9.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-33.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-76a.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-76b.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-76c.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-68a.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-68b.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-68c.c: New.
        * gcc.dg/vect/costmodel/spu/costmodel-vect-68d.c: New.
        * lib/target-supports.exp (check_effective_target_vect_int_mul):
        Add spu.

From-SVN: r126584

2 files changed, 65 insertions, 0 deletions

diff --git a/gcc/config/spu/spu.c b/gcc/config/spu/spu.c
index 7e28167..f963268 100644
--- a/gcc/config/spu/spu.c
+++ b/gcc/config/spu/spu.c
@@ -133,6 +133,7 @@ static void spu_encode_section_info (tree, rtx, int);
 static tree spu_builtin_mul_widen_even (tree);
 static tree spu_builtin_mul_widen_odd (tree);
 static tree spu_builtin_mask_for_load (void);
+static int spu_builtin_vectorization_cost (bool);
 
 extern const char *reg_names[];
 rtx spu_compare_op0, spu_compare_op1;
@@ -261,6 +262,9 @@ const struct attribute_spec spu_attribute_table[];
 #undef TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD
 #define TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD spu_builtin_mask_for_load
 
+#undef TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST
+#define TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST spu_builtin_vectorization_cost
+
 struct gcc_target targetm = TARGET_INITIALIZER;
 
 void
@@ -5191,6 +5195,21 @@ spu_builtin_mask_for_load (void)
   return d->fndecl;
 }
 
+/* Implement targetm.vectorize.builtin_vectorization_cost.  */
+static int 
+spu_builtin_vectorization_cost (bool runtime_test)
+{
+  /* If the branch of the runtime test is taken - i.e. - the vectorized
+     version is skipped - this incurs a misprediction cost (because the
+     vectorized version is expected to be the fall-through).  So we subtract
+     the latency of a mispredicted branch from the costs that are incured
+     when the vectorized version is executed.  */
+  if (runtime_test)
+    return -19;
+  else
+    return 0;
+}
+
 void
 spu_init_expanders (void)
 {   
diff --git a/gcc/config/spu/spu.h b/gcc/config/spu/spu.h
index 1f15472..b8af6b2 100644
--- a/gcc/config/spu/spu.h
+++ b/gcc/config/spu/spu.h
@@ -542,6 +542,52 @@ targetm.resolve_overloaded_builtin = spu_resolve_overloaded_builtin;	\
   do { if (LOG!=0) fprintf (FILE, "\t.align\t%d\n", (LOG)); } while (0)
 
 
+/* Model costs for the vectorizer.  */
+
+/* Cost of conditional branch.  */
+#ifndef TARG_COND_BRANCH_COST
+#define TARG_COND_BRANCH_COST        6
+#endif
+
+/* Cost of any scalar operation, excluding load and store.  */
+#ifndef TARG_SCALAR_STMT_COST
+#define TARG_SCALAR_STMT_COST        1
+#endif
+
+/* Cost of scalar load. */
+#undef TARG_SCALAR_LOAD_COST
+#define TARG_SCALAR_LOAD_COST        2 /* load + rotate */
+
+/* Cost of scalar store.  */
+#undef TARG_SCALAR_STORE_COST
+#define TARG_SCALAR_STORE_COST       10
+
+/* Cost of any vector operation, excluding load, store,
+   or vector to scalar operation.  */
+#undef TARG_VEC_STMT_COST
+#define TARG_VEC_STMT_COST           1
+
+/* Cost of vector to scalar operation.  */
+#undef TARG_VEC_TO_SCALAR_COST
+#define TARG_VEC_TO_SCALAR_COST      1
+
+/* Cost of scalar to vector operation.  */
+#undef TARG_SCALAR_TO_VEC_COST
+#define TARG_SCALAR_TO_VEC_COST      1
+
+/* Cost of aligned vector load.  */
+#undef TARG_VEC_LOAD_COST
+#define TARG_VEC_LOAD_COST           1
+
+/* Cost of misaligned vector load.  */
+#undef TARG_VEC_UNALIGNED_LOAD_COST
+#define TARG_VEC_UNALIGNED_LOAD_COST 2
+
+/* Cost of vector store.  */
+#undef TARG_VEC_STORE_COST
+#define TARG_VEC_STORE_COST          1
+
+
 /* Misc */
 
 #define CASE_VECTOR_MODE SImode