[AArch64] Add more choices for the reciprocal square root approximation

Allow a target to prefer such operation depending on the operation mode.

gcc/
	* config/aarch64/aarch64-protos.h
	(AARCH64_APPROX_MODE): New macro.
	(AARCH64_APPROX_{NONE,ALL}): Likewise.
	(cpu_approx_modes): New structure.
	(tune_params): New member "approx_modes".
	* config/aarch64/aarch64-tuning-flags.def
	(AARCH64_EXTRA_TUNE_APPROX_RSQRT): Remove macro.
	* config/aarch64/aarch64.c
	({generic,exynosm1,xgene1}_approx_modes): New core
	"cpu_approx_modes" structures.
	(generic_tunings): New member "approx_modes".
	(cortexa35_tunings): Likewise.
	(cortexa53_tunings): Likewise.
	(cortexa57_tunings): Likewise.
	(cortexa72_tunings): Likewise.
	(exynosm1_tunings): Likewise.
	(thunderx_tunings): Likewise.
	(xgene1_tunings): Likewise.
	(use_rsqrt_p): New argument for the mode and use new member from
	"tune_params".
	(aarch64_builtin_reciprocal): Devise mode from builtin.
	(aarch64_optab_supported_p): New argument for the mode.
	* doc/invoke.texi (-mlow-precision-recip-sqrt): Reword description.

From-SVN: r237395

diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h
index ab9b37a771f97de547d46cefd8e91099b8120136..1c56a1dc5e8d1b5812bc33184a42887e9f4d398f 100644 (file)
--- a/gcc/config/aarch64/aarch64-protos.h
+++ b/gcc/config/aarch64/aarch64-protos.h
@@ -178,6 +178,23 @@ struct cpu_branch_cost
   const int unpredictable;  /* Unpredictable branch or optimizing for speed.  */
 };
 
+/* Control approximate alternatives to certain FP operators.  */
+#define AARCH64_APPROX_MODE(MODE) \
+  ((MIN_MODE_FLOAT <= (MODE) && (MODE) <= MAX_MODE_FLOAT) \
+   ? (1 << ((MODE) - MIN_MODE_FLOAT)) \
+   : (MIN_MODE_VECTOR_FLOAT <= (MODE) && (MODE) <= MAX_MODE_VECTOR_FLOAT) \
+     ? (1 << ((MODE) - MIN_MODE_VECTOR_FLOAT \
+             + MAX_MODE_FLOAT - MIN_MODE_FLOAT + 1)) \
+     : (0))
+#define AARCH64_APPROX_NONE (0)
+#define AARCH64_APPROX_ALL (-1)
+
+/* Allowed modes for approximations.  */
+struct cpu_approx_modes
+{
+  const unsigned int recip_sqrt; /* Reciprocal square root.  */
+};
+
 struct tune_params
 {
   const struct cpu_cost_table *insn_extra_cost;
@@ -185,6 +202,7 @@ struct tune_params
   const struct cpu_regmove_cost *regmove_cost;
   const struct cpu_vector_cost *vec_costs;
   const struct cpu_branch_cost *branch_costs;
+  const struct cpu_approx_modes *approx_modes;
   int memmov_cost;
   int issue_rate;
   unsigned int fusible_ops;

diff --git a/gcc/config/aarch64/aarch64-tuning-flags.def b/gcc/config/aarch64/aarch64-tuning-flags.def
index 7e45a0c735dfbaaa8d078fd49ee178daa44ac908..048c2a3e3f74beeaa9ff0bf0fd4c49c9a883e158 100644 (file)
--- a/gcc/config/aarch64/aarch64-tuning-flags.def
+++ b/gcc/config/aarch64/aarch64-tuning-flags.def
@@ -29,5 +29,3 @@
      AARCH64_TUNE_ to give an enum name. */
 
 AARCH64_EXTRA_TUNING_OPTION ("rename_fma_regs", RENAME_FMA_REGS)
-AARCH64_EXTRA_TUNING_OPTION ("approx_rsqrt", APPROX_RSQRT)
-

diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index a0db3a4f6326c2e95824db29b9e4dc16f0220eea..6bc99101c908930ca3a12557c38d8a5998e0cad8 100644 (file)
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -393,6 +393,24 @@ static const struct cpu_branch_cost cortexa57_branch_cost =
   3   /* Unpredictable.  */
 };
 
+/* Generic approximation modes.  */
+static const cpu_approx_modes generic_approx_modes =
+{
+  AARCH64_APPROX_NONE  /* recip_sqrt  */
+};
+
+/* Approximation modes for Exynos M1.  */
+static const cpu_approx_modes exynosm1_approx_modes =
+{
+  AARCH64_APPROX_ALL   /* recip_sqrt  */
+};
+
+/* Approximation modes for X-Gene 1.  */
+static const cpu_approx_modes xgene1_approx_modes =
+{
+  AARCH64_APPROX_ALL   /* recip_sqrt  */
+};
+
 static const struct tune_params generic_tunings =
 {
   &cortexa57_extra_costs,
@@ -400,6 +418,7 @@ static const struct tune_params generic_tunings =
   &generic_regmove_cost,
   &generic_vector_cost,
   &generic_branch_cost,
+  &generic_approx_modes,
   4, /* memmov_cost  */
   2, /* issue_rate  */
   AARCH64_FUSE_NOTHING, /* fusible_ops  */
@@ -424,6 +443,7 @@ static const struct tune_params cortexa35_tunings =
   &cortexa53_regmove_cost,
   &generic_vector_cost,
   &generic_branch_cost,
+  &generic_approx_modes,
   4, /* memmov_cost  */
   1, /* issue_rate  */
   (AARCH64_FUSE_MOV_MOVK | AARCH64_FUSE_ADRP_ADD
@@ -449,6 +469,7 @@ static const struct tune_params cortexa53_tunings =
   &cortexa53_regmove_cost,
   &generic_vector_cost,
   &generic_branch_cost,
+  &generic_approx_modes,
   4, /* memmov_cost  */
   2, /* issue_rate  */
   (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_MOV_MOVK | AARCH64_FUSE_ADRP_ADD
@@ -474,6 +495,7 @@ static const struct tune_params cortexa57_tunings =
   &cortexa57_regmove_cost,
   &cortexa57_vector_cost,
   &cortexa57_branch_cost,
+  &generic_approx_modes,
   4, /* memmov_cost  */
   3, /* issue_rate  */
   (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_MOV_MOVK | AARCH64_FUSE_ADRP_ADD
@@ -499,6 +521,7 @@ static const struct tune_params cortexa72_tunings =
   &cortexa57_regmove_cost,
   &cortexa57_vector_cost,
   &generic_branch_cost,
+  &generic_approx_modes,
   4, /* memmov_cost  */
   3, /* issue_rate  */
   (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_MOV_MOVK | AARCH64_FUSE_ADRP_ADD
@@ -524,6 +547,7 @@ static const struct tune_params exynosm1_tunings =
   &exynosm1_regmove_cost,
   &exynosm1_vector_cost,
   &generic_branch_cost,
+  &exynosm1_approx_modes,
   4,   /* memmov_cost  */
   3,   /* issue_rate  */
   (AARCH64_FUSE_AES_AESMC), /* fusible_ops  */
@@ -538,7 +562,7 @@ static const struct tune_params exynosm1_tunings =
   48,  /* max_case_values.  */
   64,  /* cache_line_size.  */
   tune_params::AUTOPREFETCHER_WEAK, /* autoprefetcher_model.  */
-  (AARCH64_EXTRA_TUNE_APPROX_RSQRT) /* tune_flags.  */
+  (AARCH64_EXTRA_TUNE_NONE) /* tune_flags.  */
 };
 
 static const struct tune_params thunderx_tunings =
@@ -548,6 +572,7 @@ static const struct tune_params thunderx_tunings =
   &thunderx_regmove_cost,
   &generic_vector_cost,
   &generic_branch_cost,
+  &generic_approx_modes,
   6, /* memmov_cost  */
   2, /* issue_rate  */
   AARCH64_FUSE_CMP_BRANCH, /* fusible_ops  */
@@ -572,6 +597,7 @@ static const struct tune_params xgene1_tunings =
   &xgene1_regmove_cost,
   &xgene1_vector_cost,
   &generic_branch_cost,
+  &xgene1_approx_modes,
   6, /* memmov_cost  */
   4, /* issue_rate  */
   AARCH64_FUSE_NOTHING, /* fusible_ops  */
@@ -586,7 +612,7 @@ static const struct tune_params xgene1_tunings =
   0,   /* max_case_values.  */
   0,   /* cache_line_size.  */
   tune_params::AUTOPREFETCHER_OFF,     /* autoprefetcher_model.  */
-  (AARCH64_EXTRA_TUNE_APPROX_RSQRT)    /* tune_flags.  */
+  (AARCH64_EXTRA_TUNE_NONE)    /* tune_flags.  */
 };
 
 /* Support for fine-grained override of the tuning structures.  */
@@ -7320,12 +7346,12 @@ aarch64_memory_move_cost (machine_mode mode ATTRIBUTE_UNUSED,
    to optimize 1.0/sqrt.  */
 
 static bool
-use_rsqrt_p (void)
+use_rsqrt_p (machine_mode mode)
 {
   return (!flag_trapping_math
          && flag_unsafe_math_optimizations
-         && ((aarch64_tune_params.extra_tuning_flags
-              & AARCH64_EXTRA_TUNE_APPROX_RSQRT)
+         && ((aarch64_tune_params.approx_modes->recip_sqrt
+              & AARCH64_APPROX_MODE (mode))
              || flag_mrecip_low_precision_sqrt));
 }
 
@@ -7335,7 +7361,9 @@ use_rsqrt_p (void)
 static tree
 aarch64_builtin_reciprocal (tree fndecl)
 {
-  if (!use_rsqrt_p ())
+  machine_mode mode = TYPE_MODE (TREE_TYPE (fndecl));
+
+  if (!use_rsqrt_p (mode))
     return NULL_TREE;
   return aarch64_builtin_rsqrt (DECL_FUNCTION_CODE (fndecl));
 }
@@ -13731,13 +13759,13 @@ aarch64_promoted_type (const_tree t)
 /* Implement the TARGET_OPTAB_SUPPORTED_P hook.  */
 
 static bool
-aarch64_optab_supported_p (int op, machine_mode, machine_mode,
+aarch64_optab_supported_p (int op, machine_mode mode1, machine_mode,
                           optimization_type opt_type)
 {
   switch (op)
     {
     case rsqrt_optab:
-      return opt_type == OPTIMIZE_FOR_SPEED && use_rsqrt_p ();
+      return opt_type == OPTIMIZE_FOR_SPEED && use_rsqrt_p (mode1);
 
     default:
       return true;

diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index aa11209895ce354e08ce90b830de663fdf768514..e107da9aeff149d612c326b01c979267e4e54069 100644 (file)
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -13022,7 +13022,7 @@ corresponding flag to the linker.
 @item -mno-low-precision-recip-sqrt
 @opindex mlow-precision-recip-sqrt
 @opindex mno-low-precision-recip-sqrt
-Enable or disable reciprocal square root approximation.
+Enable or disable the reciprocal square root approximation.
 This option only has an effect if @option{-ffast-math} or
 @option{-funsafe-math-optimizations} is used as well.  Enabling this reduces
 precision of reciprocal square root results to about 16 bits for