[10/14] S/390: arch13: vector load/store byte reversed element for builtins

diff --git a/gcc/config/s390/vecintrin.h b/gcc/config/s390/vecintrin.h
index 1220bf6..05707a1 100644
--- a/gcc/config/s390/vecintrin.h
+++ b/gcc/config/s390/vecintrin.h
@@ -159,6 +159,20 @@  __lcbb(const void *ptr, int bndry)
 			    | __VEC_CLASS_FP_SUBNORMAL_N, &cc);	\
       cc != 3 ? 1 : 0;						\
     })
+
+#define vec_vstbrh vec_vlbrh
+#define vec_vstbrf vec_vlbrf
+#define vec_vstbrg vec_vlbrg
+#define vec_vstbrq vec_vlbrq
+#define vec_vstbrf_flt vec_vlbrf_flt
+#define vec_vstbrg_dbl vec_vlbrg_dbl
+
+#define vec_vsterb vec_vlerb
+#define vec_vsterh vec_vlerh
+#define vec_vsterf vec_vlerh
+#define vec_vsterg vec_vlerh
+#define vec_vsterf_flt vec_vlerf_flt
+#define vec_vsterg_dbl vec_vlerg_dbl
 #define vec_gather_element __builtin_s390_vec_gather_element
 #define vec_xl __builtin_s390_vec_xl
 #define vec_xld2 __builtin_s390_vec_xld2
diff --git a/gcc/config/s390/vx-builtins.md b/gcc/config/s390/vx-builtins.md
index 55b49f4..8d837c4 100644
--- a/gcc/config/s390/vx-builtins.md
+++ b/gcc/config/s390/vx-builtins.md
@@ -181,6 +181,18 @@ 
   "vllez<bhfgq>\t%v0,%1"
   [(set_attr "op_type" "VRX")])
 
+; vec_revb (vec_insert_and_zero(x))             bswap-and-replicate-1.c
+; vllebrzh, vllebrzf, vllebrzg
+(define_insn "*vec_insert_and_zero_bswap<mode>"
+  [(set (match_operand:V_HW_HSD                    0 "register_operand" "=v")
+	(bswap:V_HW_HSD (unspec:V_HW_HSD
+			 [(match_operand:<non_vec> 1 "memory_operand"    "R")]
+			 UNSPEC_VEC_INSERT_AND_ZERO)))]
+  "TARGET_VXE2"
+  "vllebrz<bhfgq>\t%v0,%1"
+  [(set_attr "op_type" "VRX")])
+
+
 (define_insn "vlbb"
   [(set (match_operand:V16QI              0 "register_operand"   "=v")
 	(unspec:V16QI [(match_operand:BLK 1 "memory_operand"      "R")
@@ -2139,3 +2151,54 @@ 
   constv = force_const_mem (V16QImode, gen_rtx_CONST_VECTOR (V16QImode, gen_rtvec_v (16, perm_rtx)));
   emit_move_insn (operands[2], constv);
 })
+
+; vec_insert (__builtin_bswap32 (*a), b, 1)        set-element-bswap-2.c
+; b[1] = __builtin_bswap32 (*a)                    set-element-bswap-3.c
+; vlebrh, vlebrf, vlebrg
+(define_insn "*vec_set_bswap_elem<mode>"
+  [(set (match_operand:V_HW_HSD                                     0 "register_operand" "=v")
+	(unspec:V_HW_HSD [(bswap:<non_vec> (match_operand:<non_vec> 1 "memory_operand"    "R"))
+		                           (match_operand:SI        2 "const_int_operand" "C")
+					   (match_operand:V_HW_HSD  3 "register_operand"  "0")]
+		  UNSPEC_VEC_SET))]
+  "TARGET_VXE2 && UINTVAL (operands[2]) < GET_MODE_NUNITS (<V_HW_HSD:MODE>mode)"
+  "vlebr<bhfgq>\t%v0,%1,%2"
+  [(set_attr "op_type" "VRX")])
+
+; vec_revb (vec_insert (*a, vec_revb (b), 1))      set-element-bswap-1.c
+; vlebrh, vlebrf, vlebrg
+(define_insn "*vec_set_bswap_vec<mode>"
+  [(set (match_operand:V_HW_HSD                                     0 "register_operand" "=v")
+	(bswap:V_HW_HSD
+	 (unspec:V_HW_HSD [(match_operand:<non_vec>                 1 "memory_operand"    "R")
+		           (match_operand:SI                        2 "const_int_operand" "C")
+			   (bswap:V_HW_HSD (match_operand:V_HW_HSD  3 "register_operand"  "0"))]
+			  UNSPEC_VEC_SET)))]
+  "TARGET_VXE2 && UINTVAL (operands[2]) < GET_MODE_NUNITS (<V_HW_HSD:MODE>mode)"
+  "vlebr<bhfgq>\t%v0,%1,%2"
+  [(set_attr "op_type" "VRX")])
+
+; *a = vec_extract (vec_revb (b), 1);              get-element-bswap-3.c
+; *a = vec_revb (b)[1];                            get-element-bswap-4.c
+; vstebrh, vstebrf, vstebrg
+(define_insn "*vec_extract_bswap_vec<mode>"
+  [(set (match_operand:<non_vec>                                    0 "memory_operand"   "=R")
+	(unspec:<non_vec> [(bswap:V_HW_HSD (match_operand:V_HW_HSD  1 "register_operand"  "v"))
+			   (match_operand:SI                        2 "const_int_operand" "C")]
+			   UNSPEC_VEC_EXTRACT))]
+  "TARGET_VXE2 && UINTVAL (operands[2]) < GET_MODE_NUNITS (<V_HW_HSD:MODE>mode)"
+  "vstebr<bhfgq>\t%v1,%0,%2"
+  [(set_attr "op_type" "VRX")])
+
+; *a = __builtin_bswap32 (vec_extract (b, 1));     get-element-bswap-1.c
+; *a = __builtin_bswap32 (b[1]);                   get-element-bswap-2.c
+; vstebrh, vstebrf, vstebrg
+(define_insn "*vec_extract_bswap_elem<mode>"
+  [(set (match_operand:<non_vec>                     0 "memory_operand"   "=R")
+	(bswap:<non_vec>
+	 (unspec:<non_vec> [(match_operand:V_HW_HSD  1 "register_operand"  "v")
+			    (match_operand:SI        2 "const_int_operand" "C")]
+			   UNSPEC_VEC_EXTRACT)))]
+  "TARGET_VXE2 && UINTVAL (operands[2]) < GET_MODE_NUNITS (<V_HW_HSD:MODE>mode)"
+  "vstebr<bhfgq>\t%v1,%0,%2"
+  [(set_attr "op_type" "VRX")])
diff --git a/gcc/testsuite/gcc.target/s390/zvector/bswap-and-replicate-1.c b/gcc/testsuite/gcc.target/s390/zvector/bswap-and-replicate-1.c
new file mode 100644
index 0000000..443b346
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/zvector/bswap-and-replicate-1.c
@@ -0,0 +1,28 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O3 -mzarch -march=arch13 -mzvector -fno-asynchronous-unwind-tables -dp" } */
+
+#include <vecintrin.h>
+
+vector signed short
+vllebrzh (const signed short *a)
+{
+  return vec_revb (vec_insert_and_zero (a));
+}
+
+/* { dg-final { scan-assembler-times "vllebrzh.*\n\tvllebrzh.*vec_insert_and_zero_bswapv8hi" 1 } } */
+
+vector signed int
+vllebrzf (const signed int *a)
+{
+  return vec_revb (vec_insert_and_zero (a));
+}
+
+/* { dg-final { scan-assembler-times "vllebrzf.*\n\tvllebrzf.*vec_insert_and_zero_bswapv4si" 1 } } */
+
+vector signed long long
+vllebrzg (const signed long long *a)
+{
+  return vec_revb (vec_insert_and_zero (a));
+}
+
+/* { dg-final { scan-assembler-times "vllebrzg.*\n\tvllebrzg.*vec_insert_and_zero_bswapv2di" 1 } } */
diff --git a/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-1.c b/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-1.c
new file mode 100644
index 0000000..ce81f10
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-1.c
@@ -0,0 +1,28 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O3 -mzarch -march=arch13 -mzvector -fno-asynchronous-unwind-tables -dp" } */
+
+#include <vecintrin.h>
+
+void
+vstebrh (signed short *a, vector signed short b)
+{
+  *a = __builtin_bswap16 (vec_extract (b, 1));
+}
+
+/* { dg-final { scan-assembler-times "vstebrh.*\n\tvstebrh.*vec_extract_bswap_elemv8hi" 1 } } */
+
+void
+vstebrf (int *a, vector int b)
+{
+  *a = __builtin_bswap32 (vec_extract (b, 1));
+}
+
+/* { dg-final { scan-assembler-times "vstebrf.*\n\tvstebrf.*vec_extract_bswap_elemv4si" 1 } } */
+
+void
+vstebrg (long long *a, vector long long b)
+{
+  *a = __builtin_bswap64 (vec_extract (b, 1));
+}
+
+/* { dg-final { scan-assembler-times "vstebrg.*\n\tvstebrg.*vec_extract_bswap_elemv2di" 1 } } */
diff --git a/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-2.c b/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-2.c
new file mode 100644
index 0000000..e8190b0
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-2.c
@@ -0,0 +1,28 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O3 -mzarch -march=arch13 -mzvector -fno-asynchronous-unwind-tables -dp" } */
+
+#include <vecintrin.h>
+
+void
+vstebrh (signed short *a, vector signed short b)
+{
+  *a = __builtin_bswap16 (b[1]);
+}
+
+/* { dg-final { scan-assembler-times "vstebrh.*\n\tvstebrh.*vec_extract_bswap_elemv8hi" 1 } } */
+
+void
+vstebrf (int *a, vector int b)
+{
+  *a = __builtin_bswap32 (b[1]);
+}
+
+/* { dg-final { scan-assembler-times "vstebrf.*\n\tvstebrf.*vec_extract_bswap_elemv4si" 1 } } */
+
+void
+vstebrg (long long *a, vector long long b)
+{
+  *a = __builtin_bswap64 (b[1]);
+}
+
+/* { dg-final { scan-assembler-times "vstebrg.*\n\tvstebrg.*vec_extract_bswap_elemv2di" 1 } } */
diff --git a/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-3.c b/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-3.c
new file mode 100644
index 0000000..c4f0287
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-3.c
@@ -0,0 +1,28 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O3 -mzarch -march=arch13 -mzvector -fno-asynchronous-unwind-tables -dp" } */
+
+#include <vecintrin.h>
+
+void
+vstebrh (signed short *a, vector signed short b)
+{
+  *a = vec_extract (vec_revb (b), 1);
+}
+
+/* { dg-final { scan-assembler-times "vstebrh.*\n\tvstebrh.*vec_extract_bswap_vecv8hi" 1 } } */
+
+void
+vstebrf (int *a, vector int b)
+{
+  *a = vec_extract (vec_revb (b), 1);
+}
+
+/* { dg-final { scan-assembler-times "vstebrf.*\n\tvstebrf.*vec_extract_bswap_vecv4si" 1 } } */
+
+void
+vstebrg (long long *a, vector long long b)
+{
+  *a = vec_extract (vec_revb (b), 1);
+}
+
+/* { dg-final { scan-assembler-times "vstebrg.*\n\tvstebrg.*vec_extract_bswap_vecv2di" 1 } } */
diff --git a/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-4.c b/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-4.c
new file mode 100644
index 0000000..1be463f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/zvector/get-element-bswap-4.c
@@ -0,0 +1,28 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O3 -mzarch -march=arch13 -mzvector -fno-asynchronous-unwind-tables -dp" } */
+
+#include <vecintrin.h>
+
+void
+vstebrh (signed short *a, vector signed short b)
+{
+  *a = vec_revb (b)[1];
+}
+
+/* { dg-final { scan-assembler-times "vstebrh.*\n\tvstebrh.*vec_extract_bswap_vecv8hi" 1 } } */
+
+void
+vstebrf (int *a, vector int b)
+{
+  *a = vec_revb (b)[1];
+}
+
+/* { dg-final { scan-assembler-times "vstebrf.*\n\tvstebrf.*vec_extract_bswap_vecv4si" 1 } } */
+
+void
+vstebrg (long long *a, vector long long b)
+{
+  *a = vec_revb (b)[1];
+}
+
+/* { dg-final { scan-assembler-times "vstebrg.*\n\tvstebrg.*vec_extract_bswap_vecv2di" 1 } } */
diff --git a/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-1.c b/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-1.c
new file mode 100644
index 0000000..218f829
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-1.c
@@ -0,0 +1,28 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O3 -mzarch -march=arch13 -mzvector -fno-asynchronous-unwind-tables -dp" } */
+
+#include <vecintrin.h>
+
+vector signed short
+vlebrh (const signed short *a, vector signed short b)
+{
+  return vec_revb (vec_insert (*a, vec_revb (b), 1));
+}
+
+/* { dg-final { scan-assembler-times "vlebrh.*\n\tvlebrh.*vec_set_bswap_vecv8hi" 1 } } */
+
+vector signed int
+vlebrf (const signed int *a, vector signed int b)
+{
+  return vec_revb (vec_insert (*a, vec_revb (b), 1));
+}
+
+/* { dg-final { scan-assembler-times "vlebrf.*\n\tvlebrf.*vec_set_bswap_vecv4si" 1 } } */
+
+vector signed long long
+vlebrg (const signed long long *a, vector signed long long b)
+{
+  return vec_revb (vec_insert (*a, vec_revb (b), 1));
+}
+
+/* { dg-final { scan-assembler-times "vlebrg.*\n\tvlebrg.*vec_set_bswap_vecv2di" 1 } } */
diff --git a/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-2.c b/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-2.c
new file mode 100644
index 0000000..8a8c95b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-2.c
@@ -0,0 +1,28 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O3 -mzarch -march=arch13 -mzvector -fno-asynchronous-unwind-tables -dp" } */
+
+#include <vecintrin.h>
+
+vector signed short
+vlebrh (const signed short *a, vector signed short b)
+{
+  return vec_insert (__builtin_bswap16 (*a), b, 1);
+}
+
+/* { dg-final { scan-assembler-times "vlebrh.*\n\tvlebrh.*vec_set_bswap_elemv8hi" 1 } } */
+
+vector signed int
+vlebrf (const signed int *a, vector signed int b)
+{
+  return vec_insert (__builtin_bswap32 (*a), b, 1);
+}
+
+/* { dg-final { scan-assembler-times "vlebrf.*\n\tvlebrf.*vec_set_bswap_elemv4si" 1 } } */
+
+vector signed long long
+vlebrg (const signed long long *a, vector signed long long b)
+{
+  return vec_insert (__builtin_bswap64 (*a), b, 1);
+}
+
+/* { dg-final { scan-assembler-times "vlebrg.*\n\tvlebrg.*vec_set_bswap_elemv2di" 1 } } */
diff --git a/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-3.c b/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-3.c
new file mode 100644
index 0000000..ef87426
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/zvector/set-element-bswap-3.c
@@ -0,0 +1,31 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O3 -mzarch -march=arch13 -mzvector -fno-asynchronous-unwind-tables -dp" } */
+
+#include <vecintrin.h>
+
+vector signed short
+vlebrh (const signed short *a, vector signed short b)
+{
+  b[1] = __builtin_bswap16 (*a);
+  return b;
+}
+
+/* { dg-final { scan-assembler-times "vlebrh.*\n\tvlebrh.*vec_set_bswap_elemv8hi" 1 } } */
+
+vector signed int
+vlebrf (const signed int *a, vector signed int b)
+{
+  b[1] = __builtin_bswap32 (*a);
+  return b;
+}
+
+/* { dg-final { scan-assembler-times "vlebrf.*\n\tvlebrf.*vec_set_bswap_elemv4si" 1 } } */
+
+vector signed long long
+vlebrg (const signed long long *a, vector signed long long b)
+{
+  b[1] = __builtin_bswap64 (*a);
+  return b;
+}
+
+/* { dg-final { scan-assembler-times "vlebrg.*\n\tvlebrg.*vec_set_bswap_elemv2di" 1 } } */