}
[(set_attr "type" "vfminmax")])
+(define_insn_and_split "<ieee_fmaxmin_op><mode>3"
+ [(set (match_operand:V_VLSF 0 "register_operand")
+ (unspec:V_VLSF
+ [(match_operand:V_VLSF 1 "register_operand")
+ (match_operand:V_VLSF 2 "register_operand")] UNSPEC_VFMAXMIN))]
+ "TARGET_VECTOR && !HONOR_SNANS (<MODE>mode) && can_create_pseudo_p ()"
+ "#"
+ "&& 1"
+ [(const_int 0)]
+{
+ riscv_vector::emit_vlmax_insn (code_for_pred (<IEEE_FMAXMIN_OP>, <MODE>mode),
+ riscv_vector::BINARY_OP, operands);
+ DONE;
+}
+[(set_attr "type" "vfminmax")])
+
;; -------------------------------------------------------------------------------
;; ---- [FP] Sign copying
;; -------------------------------------------------------------------------------
DONE;
})
+(define_expand "cond_<ieee_fmaxmin_op><mode>"
+ [(match_operand:V_VLSF 0 "register_operand")
+ (match_operand:<VM> 1 "vector_mask_operand")
+ (unspec:V_VLSF
+ [(match_operand:V_VLSF 2 "register_operand")
+ (match_operand:V_VLSF 3 "register_operand")] UNSPEC_VFMAXMIN)
+ (match_operand:V_VLSF 4 "autovec_else_operand")]
+ "TARGET_VECTOR && !HONOR_SNANS (<MODE>mode)"
+{
+ insn_code icode = code_for_pred (<IEEE_FMAXMIN_OP>, <MODE>mode);
+ riscv_vector::expand_cond_binop (icode, operands);
+ DONE;
+})
+
+(define_expand "cond_len_<ieee_fmaxmin_op><mode>"
+ [(match_operand:VF 0 "register_operand")
+ (match_operand:<VM> 1 "vector_mask_operand")
+ (unspec:VF
+ [(match_operand:VF 2 "register_operand")
+ (match_operand:VF 3 "register_operand")] UNSPEC_VFMAXMIN)
+ (match_operand:VF 4 "autovec_else_operand")
+ (match_operand 5 "autovec_length_operand")
+ (match_operand 6 "const_0_operand")]
+ "TARGET_VECTOR && !HONOR_SNANS (<MODE>mode)"
+{
+ insn_code icode = code_for_pred (<IEEE_FMAXMIN_OP>, <MODE>mode);
+ riscv_vector::expand_cond_len_binop (icode, operands);
+ DONE;
+})
+
;; -------------------------------------------------------------------------
;; ---- [INT] Conditional ternary operations
;; -------------------------------------------------------------------------
DONE;
})
+(define_expand "reduc_fmax_scal_<mode>"
+ [(match_operand:<VEL> 0 "register_operand")
+ (match_operand:V_VLSF 1 "register_operand")]
+ "TARGET_VECTOR && !HONOR_SNANS (<MODE>mode)"
+{
+ REAL_VALUE_TYPE rv;
+ real_inf (&rv, true);
+ rtx f = const_double_from_real_value (rv, <VEL>mode);
+ riscv_vector::expand_reduction (UNSPEC_REDUC_MAX, riscv_vector::REDUCE_OP,
+ operands, f);
+ DONE;
+})
+
+(define_expand "reduc_fmin_scal_<mode>"
+ [(match_operand:<VEL> 0 "register_operand")
+ (match_operand:V_VLSF 1 "register_operand")]
+ "TARGET_VECTOR && !HONOR_SNANS (<MODE>mode)"
+{
+ REAL_VALUE_TYPE rv;
+ real_inf (&rv, false);
+ rtx f = const_double_from_real_value (rv, <VEL>mode);
+ riscv_vector::expand_reduction (UNSPEC_REDUC_MIN, riscv_vector::REDUCE_OP,
+ operands, f);
+ DONE;
+})
+
;; -------------------------------------------------------------------------
;; ---- [FP] Left-to-right reductions
;; -------------------------------------------------------------------------
return false;
return icode != maybe_code_for_pred (SMIN, mode)
+ && icode != maybe_code_for_pred (UNSPEC_VFMIN, mode)
&& icode != maybe_code_for_pred (SMAX, mode)
+ && icode != maybe_code_for_pred (UNSPEC_VFMAX, mode)
&& icode != maybe_code_for_pred (NEG, mode)
&& icode != maybe_code_for_pred (ABS, mode)
/* narrower-FP -> FP */
UNSPEC_VFFMA
+ UNSPEC_VFMAX
+ UNSPEC_VFMIN
+
;; Integer and Float Reduction
UNSPEC_REDUC
UNSPEC_REDUC_SUM
(define_int_attr UNSPEC [(UNSPEC_VSLIDE1UP "UNSPEC_VSLIDE1UP")
(UNSPEC_VSLIDE1DOWN "UNSPEC_VSLIDE1DOWN")])
+(define_int_iterator UNSPEC_VFMAXMIN [UNSPEC_VFMAX UNSPEC_VFMIN])
+
+(define_int_attr ieee_fmaxmin_op [(UNSPEC_VFMAX "fmax") (UNSPEC_VFMIN "fmin")])
+(define_int_attr IEEE_FMAXMIN_OP [(UNSPEC_VFMAX "UNSPEC_VFMAX") (UNSPEC_VFMIN "UNSPEC_VFMIN")])
+
(define_code_iterator any_int_binop [plus minus and ior xor ashift ashiftrt lshiftrt
smax umax smin umin mult div udiv mod umod
])
[(set_attr "type" "<float_insn_type>")
(set_attr "mode" "<MODE>")])
+(define_insn "@pred_<ieee_fmaxmin_op><mode>"
+ [(set (match_operand:V_VLSF 0 "register_operand" "=vd, vd, vr, vr")
+ (if_then_else:V_VLSF
+ (unspec:<VM>
+ [(match_operand:<VM> 1 "vector_mask_operand" " vm, vm,Wc1,Wc1")
+ (match_operand 5 "vector_length_operand" " rK, rK, rK, rK")
+ (match_operand 6 "const_int_operand" " i, i, i, i")
+ (match_operand 7 "const_int_operand" " i, i, i, i")
+ (match_operand 8 "const_int_operand" " i, i, i, i")
+ (reg:SI VL_REGNUM)
+ (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
+ (unspec:V_VLSF
+ [(match_operand:V_VLSF 3 "register_operand" " vr, vr, vr, vr")
+ (match_operand:V_VLSF 4 "register_operand" " vr, vr, vr, vr")]
+ UNSPEC_VFMAXMIN)
+ (match_operand:V_VLSF 2 "vector_merge_operand" " vu, 0, vu, 0")))]
+ "TARGET_VECTOR"
+ "v<ieee_fmaxmin_op>.vv\t%0,%3,%4%p1"
+ [(set_attr "type" "vfminmax")
+ (set_attr "mode" "<MODE>")])
+
(define_insn "@pred_<optab><mode>_scalar"
[(set (match_operand:VF 0 "register_operand" "=vd, vd, vr, vr")
(if_then_else:VF
[(set_attr "type" "<float_insn_type>")
(set_attr "mode" "<MODE>")])
+(define_insn "@pred_<ieee_fmaxmin_op><mode>_scalar"
+ [(set (match_operand:VF 0 "register_operand" "=vd, vd, vr, vr")
+ (if_then_else:VF
+ (unspec:<VM>
+ [(match_operand:<VM> 1 "vector_mask_operand" " vm, vm,Wc1,Wc1")
+ (match_operand 5 "vector_length_operand" " rK, rK, rK, rK")
+ (match_operand 6 "const_int_operand" " i, i, i, i")
+ (match_operand 7 "const_int_operand" " i, i, i, i")
+ (match_operand 8 "const_int_operand" " i, i, i, i")
+ (reg:SI VL_REGNUM)
+ (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
+ (unspec:VF
+ [(match_operand:VF 3 "register_operand" " vr, vr, vr, vr")
+ (vec_duplicate:VF
+ (match_operand:<VEL> 4 "register_operand" " f, f, f, f"))]
+ UNSPEC_VFMAXMIN)
+ (match_operand:VF 2 "vector_merge_operand" " vu, 0, vu, 0")))]
+ "TARGET_VECTOR"
+ "v<ieee_fmaxmin_op>.vf\t%0,%3,%4%p1"
+ [(set_attr "type" "vfminmax")
+ (set_attr "mode" "<MODE>")])
+
(define_insn "@pred_<optab><mode>_scalar"
[(set (match_operand:VF 0 "register_operand" "=vd, vd, vr, vr")
(if_then_else:VF
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=scalable --param vect-epilogues-nomask=0 -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+
+#ifndef FN
+#define FN(X) __builtin_fmax##X
+#endif
+
+#define DEF_LOOP(FN, SUFFIX, TYPE) \
+ void __attribute__ ((noipa)) \
+ test_##TYPE (TYPE *__restrict x, TYPE *__restrict y, int n) \
+ { \
+ for (int i = 0; i < n; ++i) \
+ x[i] = FN (SUFFIX) (x[i], y[i]); \
+ }
+
+#define TEST_ALL(T) \
+ T (FN, f, float) \
+ T (FN, , double)
+
+TEST_ALL (DEF_LOOP)
+
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+} 2 } } */
--- /dev/null
+/* { dg-do run { target { riscv_v } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-signaling-nans" } */
+
+#include <math.h>
+#include "fmax-1.c"
+
+#define N 99
+
+#define TEST_LOOP(FN, SUFFIX, TYPE) \
+ { \
+ TYPE dst[N], x[N], y[N]; \
+ for (int i = 0; i < N; ++i) \
+ { \
+ x[i] = i; \
+ dst[i] = i; \
+ y[i] = ((i & 1) - 1) * i * i; \
+ } \
+ y[0] = -0.0; \
+ y[1] = 0.0; \
+ y[2] = nan ("0.0"); \
+ y[3] = INFINITY; \
+ y[4] = -INFINITY; \
+ x[5] = -0.0; \
+ x[6] = 0.0; \
+ x[7] = nan ("0.0"); \
+ x[8] = INFINITY; \
+ x[9] = -INFINITY; \
+ dst[5] = -0.0; \
+ dst[6] = 0.0; \
+ dst[7] = nan ("0.0"); \
+ dst[8] = INFINITY; \
+ dst[9] = -INFINITY; \
+ test_##TYPE (dst, y, N); \
+ for (int i = 0; i < N; ++i) \
+ { \
+ double ref = FN (SUFFIX) (x[i], y[i]); \
+ if (dst[i] != ref) \
+ __builtin_abort (); \
+ asm volatile ("" ::: "memory"); \
+ } \
+ }
+
+int __attribute__ ((optimize ("1"))) main (void)
+{
+ TEST_ALL (TEST_LOOP)
+ return 0;
+}
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+
+#ifndef FN
+#define FN(X) __builtin_fmax##X
+#endif
+
+#define DEF_LOOP(FN, SUFFIX, TYPE) \
+ void __attribute__ ((noipa)) \
+ test_##TYPE (TYPE *__restrict x, TYPE *__restrict y, int n) \
+ { \
+ for (int i = 0; i < n; ++i) \
+ x[i] = FN (SUFFIX) (x[i], y[i]); \
+ }
+
+#define TEST_ALL(T) \
+ T (FN, f16, _Float16) \
+
+TEST_ALL (DEF_LOOP)
+
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+} 1 } } */
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-signaling-nans" } */
+
+#include <math.h>
+#include "fmax_zvfh-1.c"
+
+#define N 99
+
+#define TEST_LOOP(FN, SUFFIX, TYPE) \
+ { \
+ TYPE dst[N], x[N], y[N]; \
+ for (int i = 0; i < N; ++i) \
+ { \
+ x[i] = i; \
+ dst[i] = i; \
+ y[i] = ((i & 1) - 1) * i * i; \
+ } \
+ y[0] = -0.0; \
+ y[1] = 0.0; \
+ y[2] = nan ("0.0"); \
+ y[3] = INFINITY; \
+ y[4] = -INFINITY; \
+ x[5] = -0.0; \
+ x[6] = 0.0; \
+ x[7] = nan ("0.0"); \
+ x[8] = INFINITY; \
+ x[9] = -INFINITY; \
+ dst[5] = -0.0; \
+ dst[6] = 0.0; \
+ dst[7] = nan ("0.0"); \
+ dst[8] = INFINITY; \
+ dst[9] = -INFINITY; \
+ kest_##TYPE (dst, y, N); \
+ for (int i = 0; i < N; ++i) \
+ { \
+ double ref = FN (SUFFIX) (x[i], y[i]); \
+ if (dst[i] != ref) \
+ __builtin_abort (); \
+ asm volatile ("" ::: "memory"); \
+ } \
+ }
+
+
+int __attribute__ ((optimize ("1"))) main (void)
+{
+ TEST_ALL (TEST_LOOP)
+ return 0;
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=scalable --param vect-epilogues-nomask=0 -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+
+#define FN(X) __builtin_fmin##X
+#include "fmax-1.c"
+
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+} 2 } } */
+
--- /dev/null
+/* { dg-do run { target { riscv_v } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#define FN(X) __builtin_fmin##X
+#include "fmax_run-1.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable --param vect-epilogues-nomask=0 -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+
+#define FN(X) __builtin_fmin##X
+#include "fmax_zvfh-1.c"
+
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+} 1 } } */
+
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-signaling-nans" } */
+
+#define FN(X) __builtin_fmin##X
+#include "fmax_zvfh_run-1.c"
/* { dg-do compile } */
-/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include <stdint-gcc.h>
+#include <math.h>
#ifndef FN
#define FN(X) __builtin_fmax##X
T (FN, TYPE, PRED_TYPE, two, 2)
#define TEST_ALL(T) \
- TEST_TYPE (T, FN (f16), _Float16, int16_t) \
TEST_TYPE (T, FN (f32), float, int32_t) \
TEST_TYPE (T, FN (f64), double, int64_t)
TEST_ALL (DEF_LOOP)
-/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 9 } } */
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 6 } } */
/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
/* { dg-do compile } */
-/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include <stdint-gcc.h>
+#include <math.h>
#ifndef FN
#define FN(X) __builtin_fmax##X
/* { dg-do compile } */
-/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include <stdint-gcc.h>
+#include <math.h>
#ifndef FN
#define FN(X) __builtin_fmax##X
T (FN, TYPE, PRED_TYPE, two, 2)
#define TEST_ALL(T) \
- TEST_TYPE (T, FN (f16), _Float16, int16_t) \
TEST_TYPE (T, FN (f32), float, int32_t) \
TEST_TYPE (T, FN (f64), double, int64_t)
TEST_ALL (DEF_LOOP)
-/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 9 } } */
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 6 } } */
/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
/* { dg-do compile } */
-/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include <stdint-gcc.h>
+#include <math.h>
#ifndef FN
#define FN(X) __builtin_fmax##X
T (FN, TYPE, PRED_TYPE, two, 2)
#define TEST_ALL(T) \
- TEST_TYPE (T, FN (f16), _Float16, int16_t) \
TEST_TYPE (T, FN (f32), float, int32_t) \
TEST_TYPE (T, FN (f64), double, int64_t)
TEST_ALL (DEF_LOOP)
-/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 9 } } */
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 6 } } */
/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
/* { dg-do run { target { riscv_v } } } */
-/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include "cond_fmax-1.c"
+#include <math.h>
#define N 99
/* { dg-do run { target { riscv_v } } } */
-/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include "cond_fmax-2.c"
+#include <math.h>
#define N 99
/* { dg-do run { target { riscv_v } } } */
-/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include "cond_fmax-3.c"
+#include <math.h>
#define N 99
/* { dg-do run { target { riscv_v } } } */
-/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include "cond_fmax-4.c"
+#include <math.h>
#define N 99
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+#include <math.h>
+
+#ifndef FN
+#define FN(X) __builtin_fmax##X
+#endif
+
+#define DEF_LOOP(FN, TYPE, PRED_TYPE, NAME, CONST) \
+ void __attribute__ ((noipa)) \
+ test_##TYPE##_##NAME (TYPE *__restrict x, \
+ TYPE *__restrict y, \
+ PRED_TYPE *__restrict pred, \
+ int n) \
+ { \
+ for (int i = 0; i < n; ++i) \
+ x[i] = pred[i] != 1 ? FN (y[i], CONST) : y[i]; \
+ }
+
+#define TEST_TYPE(T, FN, TYPE, PRED_TYPE) \
+ T (FN, TYPE, PRED_TYPE, zero, 0) \
+ T (FN, TYPE, PRED_TYPE, one, 1) \
+ T (FN, TYPE, PRED_TYPE, two, 2)
+
+#define TEST_ALL(T) \
+ TEST_TYPE (T, FN (f16), _Float16, int16_t)
+
+TEST_ALL (DEF_LOOP)
+
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 3 } } */
+/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+#include <math.h>
+
+#ifndef FN
+#define FN(X) __builtin_fmax##X
+#endif
+
+#define DEF_LOOP(FN, TYPE, NAME, CONST) \
+ void __attribute__ ((noipa)) \
+ test_##TYPE##_##NAME (TYPE *__restrict x, \
+ TYPE *__restrict y, \
+ TYPE *__restrict z, \
+ int n) \
+ { \
+ for (int i = 0; i < n; ++i) \
+ x[i] = y[i] < 8 ? FN (z[i], CONST) : y[i]; \
+ }
+
+#define TEST_TYPE(T, FN, TYPE) \
+ T (FN, TYPE, zero, 0) \
+ T (FN, TYPE, one, 1) \
+ T (FN, TYPE, two, 2)
+
+#define TEST_ALL(T) \
+ TEST_TYPE (T, FN (f16), _Float16)
+
+TEST_ALL (DEF_LOOP)
+
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 3 } } */
+/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+#include <math.h>
+
+#ifndef FN
+#define FN(X) __builtin_fmax##X
+#endif
+
+#define DEF_LOOP(FN, TYPE, PRED_TYPE, NAME, CONST) \
+ void __attribute__ ((noipa)) \
+ test_##TYPE##_##NAME (TYPE *__restrict x, \
+ TYPE *__restrict y, \
+ PRED_TYPE *__restrict pred, \
+ int n) \
+ { \
+ for (int i = 0; i < n; ++i) \
+ x[i] = pred[i] != 1 ? FN (y[i], CONST) : 4; \
+ }
+
+#define TEST_TYPE(T, FN, TYPE, PRED_TYPE) \
+ T (FN, TYPE, PRED_TYPE, zero, 0) \
+ T (FN, TYPE, PRED_TYPE, one, 1) \
+ T (FN, TYPE, PRED_TYPE, two, 2)
+
+#define TEST_ALL(T) \
+ TEST_TYPE (T, FN (f16), _Float16, int16_t)
+
+TEST_ALL (DEF_LOOP)
+
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 3 } } */
+/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+#include <math.h>
+
+#ifndef FN
+#define FN(X) __builtin_fmax##X
+#endif
+
+#define DEF_LOOP(FN, TYPE, PRED_TYPE, NAME, CONST) \
+ void __attribute__ ((noipa)) \
+ test_##TYPE##_##NAME (TYPE *__restrict x, \
+ TYPE *__restrict y, \
+ PRED_TYPE *__restrict pred, \
+ int n) \
+ { \
+ for (int i = 0; i < n; ++i) \
+ x[i] = pred[i] != 1 ? FN (y[i], CONST) : 0; \
+ }
+
+#define TEST_TYPE(T, FN, TYPE, PRED_TYPE) \
+ T (FN, TYPE, PRED_TYPE, zero, 0) \
+ T (FN, TYPE, PRED_TYPE, one, 1) \
+ T (FN, TYPE, PRED_TYPE, two, 2)
+
+#define TEST_ALL(T) \
+ TEST_TYPE (T, FN (f16), _Float16, int16_t)
+
+TEST_ALL (DEF_LOOP)
+
+/* { dg-final { scan-assembler-times {vfmax\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 3 } } */
+/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include "cond_fmax_zvfh-1.c"
+#include <math.h>
+
+#define N 99
+
+#define TEST_LOOP(FN, TYPE, PRED_TYPE, NAME, CONST) \
+ { \
+ TYPE x[N], y[N]; \
+ PRED_TYPE pred[N]; \
+ for (int i = 0; i < N; ++i) \
+ { \
+ y[i] = i * i; \
+ pred[i] = i % 3; \
+ } \
+ test_##TYPE##_##NAME (x, y, pred, N); \
+ for (int i = 0; i < N; ++i) \
+ { \
+ TYPE expected = i % 3 != 1 ? FN (y[i], CONST) : y[i]; \
+ if (x[i] != expected) \
+ __builtin_abort (); \
+ asm volatile ("" ::: "memory"); \
+ } \
+ }
+
+int
+main (void)
+{
+ TEST_ALL (TEST_LOOP)
+ return 0;
+}
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include "cond_fmax_zvfh-2.c"
+#include <math.h>
+
+#define N 99
+
+#define TEST_LOOP(FN, TYPE, NAME, CONST) \
+ { \
+ TYPE x[N], y[N], z[N]; \
+ for (int i = 0; i < N; ++i) \
+ { \
+ y[i] = i % 13; \
+ z[i] = i * i; \
+ } \
+ test_##TYPE##_##NAME (x, y, z, N); \
+ for (int i = 0; i < N; ++i) \
+ { \
+ TYPE expected = y[i] < 8 ? FN (z[i], CONST) : y[i]; \
+ if (x[i] != expected) \
+ __builtin_abort (); \
+ asm volatile ("" ::: "memory"); \
+ } \
+ }
+
+int
+main (void)
+{
+ TEST_ALL (TEST_LOOP)
+ return 0;
+}
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include "cond_fmax_zvfh-3.c"
+#include <math.h>
+
+#define N 99
+
+#define TEST_LOOP(FN, TYPE, PRED_TYPE, NAME, CONST) \
+ { \
+ TYPE x[N], y[N]; \
+ PRED_TYPE pred[N]; \
+ for (int i = 0; i < N; ++i) \
+ { \
+ y[i] = i * i; \
+ pred[i] = i % 3; \
+ } \
+ test_##TYPE##_##NAME (x, y, pred, N); \
+ for (int i = 0; i < N; ++i) \
+ { \
+ TYPE expected = i % 3 != 1 ? FN (y[i], CONST) : 4; \
+ if (x[i] != expected) \
+ __builtin_abort (); \
+ asm volatile ("" ::: "memory"); \
+ } \
+ }
+
+int
+main (void)
+{
+ TEST_ALL (TEST_LOOP)
+ return 0;
+}
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include "cond_fmax_zvfh-4.c"
+#include <math.h>
+
+#define N 99
+
+#define TEST_LOOP(FN, TYPE, PRED_TYPE, NAME, CONST) \
+ { \
+ TYPE x[N], y[N]; \
+ PRED_TYPE pred[N]; \
+ for (int i = 0; i < N; ++i) \
+ { \
+ y[i] = i * i; \
+ pred[i] = i % 3; \
+ } \
+ test_##TYPE##_##NAME (x, y, pred, N); \
+ for (int i = 0; i < N; ++i) \
+ { \
+ TYPE expected = i % 3 != 1 ? FN (y[i], CONST) : 0; \
+ if (x[i] != expected) \
+ __builtin_abort (); \
+ asm volatile ("" ::: "memory"); \
+ } \
+ }
+
+int
+main (void)
+{
+ TEST_ALL (TEST_LOOP)
+ return 0;
+}
/* { dg-do compile } */
-/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include <stdint-gcc.h>
+#include <math.h>
#define FN(X) __builtin_fmin##X
#include "cond_fmax-1.c"
-/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 9 } } */
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 6 } } */
/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
/* { dg-do compile } */
-/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include <stdint-gcc.h>
+#include <math.h>
#define FN(X) __builtin_fmin##X
#include "cond_fmax-2.c"
/* { dg-do compile } */
-/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include <stdint-gcc.h>
+#include <math.h>
#define FN(X) __builtin_fmin##X
#include "cond_fmax-3.c"
-/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 9 } } */
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 6 } } */
/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
/* { dg-do compile } */
-/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#include <stdint-gcc.h>
+#include <math.h>
#define FN(X) __builtin_fmin##X
#include "cond_fmax-4.c"
-/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 9 } } */
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 6 } } */
/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
/* { dg-do run { target { riscv_v } } } */
-/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#define FN(X) __builtin_fmin##X
#include "cond_fmax_run-1.c"
/* { dg-do run { target { riscv_v } } } */
-/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#define FN(X) __builtin_fmin##X
#include "cond_fmax_run-2.c"
/* { dg-do run { target { riscv_v } } } */
-/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#define FN(X) __builtin_fmin##X
#include "cond_fmax_run-3.c"
/* { dg-do run { target { riscv_v } } } */
-/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -ffast-math" } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
#define FN(X) __builtin_fmin##X
#include "cond_fmax_run-4.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+#include <math.h>
+
+#define FN(X) __builtin_fmin##X
+#include "cond_fmax_zvfh-1.c"
+
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 3 } } */
+/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+#include <math.h>
+
+#define FN(X) __builtin_fmin##X
+#include "cond_fmax_zvfh-2.c"
+
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 3 } } */
+/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+#include <math.h>
+
+#define FN(X) __builtin_fmin##X
+#include "cond_fmax_zvfh-3.c"
+
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 3 } } */
+/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+#include <math.h>
+
+#define FN(X) __builtin_fmin##X
+#include "cond_fmax_zvfh-4.c"
+
+/* { dg-final { scan-assembler-times {vfmin\.vv\s+v[0-9]+,v[0-9]+,v[0-9]+,v0.t} 3 } } */
+/* { dg-final { scan-assembler-not {\tvf?merge\.v[vxi]m\t} } } */
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#define FN(X) __builtin_fmin##X
+#include "cond_fmax_zvfh_run-1.c"
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#define FN(X) __builtin_fmin##X
+#include "cond_fmax_zvfh_run-2.c"
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#define FN(X) __builtin_fmin##X
+#include "cond_fmax_zvfh_run-3.c"
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-vect-cost-model -fno-signaling-nans" } */
+
+#define FN(X) __builtin_fmin##X
+#include "cond_fmax_zvfh_run-4.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable --param vect-epilogues-nomask=0 -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+
+#define DEF_REDUC_FMAXMIN(TYPE, NAME, MAXMIN_OP) \
+ TYPE __attribute__ ((noinline, noclone)) \
+ reduc_##NAME##_##TYPE (TYPE *a, int n) \
+ { \
+ TYPE r = -0.0; \
+ for (int i = 0; i < n; ++i) \
+ r = MAXMIN_OP (r, a[i]); \
+ return r; \
+ }
+
+#define TEST_FMAXMIN(T) \
+ T (float, max, __builtin_fmaxf) \
+ T (double, max, __builtin_fmax) \
+ T (float, min, __builtin_fminf) \
+ T (double, min, __builtin_fmin)
+
+
+TEST_FMAXMIN (DEF_REDUC_FMAXMIN)
+
+/* { dg-final { scan-assembler-times {vfredmax\.vs\s+v[0-9]+,\s*v[0-9]+,\s*v[0-9]+} 2 } } */
+/* { dg-final { scan-assembler-times {vfredmin\.vs\s+v[0-9]+,\s*v[0-9]+,\s*v[0-9]+} 2 } } */
--- /dev/null
+/* { dg-do run { target { riscv_v } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-signaling-nans" } */
+
+#include <math.h>
+
+#include "reduc-10.c"
+
+#define NUM_ELEMS(TYPE) (73 + sizeof (TYPE))
+
+#define INIT_VECTOR(TYPE) \
+ TYPE a[NUM_ELEMS (TYPE) + 1]; \
+ for (int i = 0; i < NUM_ELEMS (TYPE) + 1; i++) \
+ { \
+ a[i] = ((i * 2) * (i & 1 ? 1 : -1) | 3); \
+ asm volatile ("" ::: "memory"); \
+ } \
+ a[0] = -0.0; \
+ a[1] = nan ("0.0"); \
+ a[2] = nan ("1.0"); \
+ a[3] = 0.0; \
+ a[4] = -INFINITY; \
+ a[5] = INFINITY; \
+
+#define TEST_REDUC_FMAXMIN(TYPE, NAME, MAXMIN_OP) \
+ { \
+ INIT_VECTOR (TYPE); \
+ TYPE r1 = reduc_##NAME##_##TYPE (a, NUM_ELEMS (TYPE)); \
+ volatile TYPE r2 = -0.0; \
+ for (int i = 0; i < NUM_ELEMS (TYPE); ++i) \
+ r2 = MAXMIN_OP (r2, a[i]); \
+ if (r1 != r2) \
+ __builtin_abort (); \
+ }
+
+__attribute__ ((optimize ("1")))
+int main ()
+{
+ TEST_FMAXMIN (TEST_REDUC_FMAXMIN)
+
+ return 0;
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable --param vect-epilogues-nomask=0 -fno-signaling-nans" } */
+
+#include <stdint-gcc.h>
+
+#define DEF_REDUC_FMAXMIN(TYPE, NAME, MAXMIN_OP) \
+ TYPE __attribute__ ((noinline, noclone)) \
+ reduc_##NAME##_##TYPE (TYPE *a, int n) \
+ { \
+ TYPE r = -0.0; \
+ for (int i = 0; i < n; ++i) \
+ r = MAXMIN_OP (r, a[i]); \
+ return r; \
+ }
+
+#define TEST_FMAXMIN(T) \
+ T (_Float16, max, __builtin_fmaxf16) \
+ T (_Float16, min, __builtin_fminf16) \
+
+
+TEST_FMAXMIN (DEF_REDUC_FMAXMIN)
+
+/* { dg-final { scan-assembler-times {vfredmax\.vs\s+v[0-9]+,\s*v[0-9]+,\s*v[0-9]+} 1 } } */
+/* { dg-final { scan-assembler-times {vfredmin\.vs\s+v[0-9]+,\s*v[0-9]+,\s*v[0-9]+} 1 } } */
--- /dev/null
+/* { dg-do run { target { riscv_zvfh } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable -fno-signaling-nans" } */
+
+#include <math.h>
+
+#include "reduc_zvfh-10.c"
+
+#define NUM_ELEMS(TYPE) (73 + sizeof (TYPE))
+
+#define INIT_VECTOR(TYPE) \
+ TYPE a[NUM_ELEMS (TYPE) + 1]; \
+ for (int i = 0; i < NUM_ELEMS (TYPE) + 1; i++) \
+ { \
+ a[i] = ((i * 2) * (i & 1 ? 1 : -1) | 3); \
+ asm volatile ("" ::: "memory"); \
+ } \
+ a[0] = -0.0; \
+ a[1] = nan ("0.0"); \
+ a[2] = nan ("1.0"); \
+ a[3] = 0.0; \
+ a[4] = -INFINITY; \
+ a[5] = INFINITY; \
+
+#define TEST_REDUC_FMAXMIN(TYPE, NAME, MAXMIN_OP) \
+ { \
+ INIT_VECTOR (TYPE); \
+ TYPE r1 = reduc_##NAME##_##TYPE (a, NUM_ELEMS (TYPE)); \
+ volatile TYPE r2 = -0.0; \
+ for (int i = 0; i < NUM_ELEMS (TYPE); ++i) \
+ r2 = MAXMIN_OP (r2, a[i]); \
+ if (r1 != r2) \
+ __builtin_abort (); \
+ }
+
+__attribute__ ((optimize ("1")))
+int main ()
+{
+ TEST_FMAXMIN (TEST_REDUC_FMAXMIN)
+
+ return 0;
+}