[PowerPC] Implement plain VSX load/store builtins.

[jtony]

Summary

Implement all the different 24 overloads for vec_xl and vec_xst.

Signatures:
vector signed char vec_xl (signed long long, signed char *);
vector unsigned char vec_xl (signed long long, unsigned char *);
vector signed int vec_xl (signed long long, signed int *);
vector unsigned int vec_xl (signed long long, unsigned int *);
vector signed int128 vec_xl (signed long long, signed int128 *);
vector unsigned int128 vec_xl (signed long long, unsigned int128 *);
vector signed long long vec_xl (signed long long, signed long long *);
vector unsigned long long vec_xl (signed long long, unsigned long long *);
vector signed short vec_xl (signed long long, signed short *);
vector unsigned short vec_xl (signed long long, unsigned short *);
vector double vec_xl (signed long long, double *);
vector float vec_xl (signed long long, float *);
void vec_xst (vector signed char, signed long long, signed char *);
void vec_xst (vector unsigned char, signed long long, unsigned char *);
void vec_xst (vector signed int, signed long long, signed int *);
void vec_xst (vector unsigned int, signed long long, unsigned int *);
void vec_xst (vector signed int128, signed long long, signed int128 *);
void vec_xst (vector unsigned int128, signed long long, unsigned int128 *);
void vec_xst (vector signed long long, signed long long, signed long long *);
void vec_xst (vector unsigned long long, signed long long, unsigned long long *);
void vec_xst (vector signed short, signed long long, signed short *);
void vec_xst (vector unsigned short, signed long long, unsigned short *);
void vec_xst (vector double, signed long long, double *);
void vec_xst (vector float, signed long long, float *);

[nemanjai]

lib/Headers/altivec.h

15647

Please move the __int128 overloads below the 64-bit ones to keep the overloads ordered by size.

[jtony]

Reorder the overloads according to their size.

[jtony]

I also migrated this commit from the old anonymous repository to the new jtony repository to prepare for committing upstream (use git svn dcommit). Should not have any difference with the previous patch, update it just in case.

[jtony]

Add -U999999 option to git diff to show context for this patch.

[kbarton]

LGTM

[Eugene.Zelenko]

Committed in rL286455. See http://llvm.org/docs/Phabricator.html#committing-a-change for review automatic closing.

lib/Headers/altivec.h

vec_revb(vector unsigned __int128 __a) {
   vector unsigned char __indices =
       { 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 };
   return (vector unsigned __int128)vec_perm((vector signed int)__a,
                                             (vector signed int)__a,
                                              __indices);
 }
 #endif /* END __POWER8_VECTOR__ && __powerpc64__ */
 
+
+/* vec_xl */
+
+static inline __ATTRS_o_ai vector signed char vec_xl(signed long long __offset,
+                                                     signed char *__ptr) {
+  return *(vector signed char *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector unsigned char
+vec_xl(signed long long __offset, unsigned char *__ptr) {
+  return *(vector unsigned char *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector signed short vec_xl(signed long long __offset,
+                                                      signed short *__ptr) {
+  return *(vector signed short *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector unsigned short
+vec_xl(signed long long __offset, unsigned short *__ptr) {
+  return *(vector unsigned short *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector signed int vec_xl(signed long long __offset,
+                                                    signed int *__ptr) {
+  return *(vector signed int *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector unsigned int vec_xl(signed long long __offset,
+                                                      unsigned int *__ptr) {
+  return *(vector unsigned int *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector float vec_xl(signed long long __offset,
+                                               float *__ptr) {
+  return *(vector float *)(__ptr + __offset);
+}
+
+#ifdef __VSX__
+static inline __ATTRS_o_ai vector signed long long
+vec_xl(signed long long __offset, signed long long *__ptr) {
+  return *(vector signed long long *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector unsigned long long
+vec_xl(signed long long __offset, unsigned long long *__ptr) {
+  return *(vector unsigned long long *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector double vec_xl(signed long long __offset,
+                                                double *__ptr) {
+  return *(vector double *)(__ptr + __offset);
+}
+#endif
+
+#if defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+static inline __ATTRS_o_ai vector signed __int128
+vec_xl(signed long long __offset, signed __int128 *__ptr) {
+  return *(vector signed __int128 *)(__ptr + __offset);
+}
+
+static inline __ATTRS_o_ai vector unsigned __int128
+vec_xl(signed long long __offset, unsigned __int128 *__ptr) {
+  return *(vector unsigned __int128 *)(__ptr + __offset);
+}
+#endif
+
+/* vec_xst */
+
+static inline __ATTRS_o_ai void vec_xst(vector signed char __vec,
+                                        signed long long __offset,
+                                        signed char *__ptr) {
+  *(vector signed char *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector unsigned char __vec,
+                                        signed long long __offset,
+                                        unsigned char *__ptr) {
+  *(vector unsigned char *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector signed short __vec,
+                                        signed long long __offset,
+                                        signed short *__ptr) {
+  *(vector signed short *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector unsigned short __vec,
+                                        signed long long __offset,
+                                        unsigned short *__ptr) {
+  *(vector unsigned short *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector signed int __vec,
+                                        signed long long __offset,
+                                        signed int *__ptr) {
+  *(vector signed int *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector unsigned int __vec,
+                                        signed long long __offset,
+                                        unsigned int *__ptr) {
+  *(vector unsigned int *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector float __vec,
+                                        signed long long __offset,
+                                        float *__ptr) {
+  *(vector float *)(__ptr + __offset) = __vec;
+}
+
+#ifdef __VSX__
+static inline __ATTRS_o_ai void vec_xst(vector signed long long __vec,
+                                        signed long long __offset,
+                                        signed long long *__ptr) {
+  *(vector signed long long *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector unsigned long long __vec,
+                                        signed long long __offset,
+                                        unsigned long long *__ptr) {
+  *(vector unsigned long long *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector double __vec,
+                                        signed long long __offset,
+                                        double *__ptr) {
+  *(vector double *)(__ptr + __offset) = __vec;
+}
+#endif
+
+#if defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+static inline __ATTRS_o_ai void vec_xst(vector signed __int128 __vec,
+                                        signed long long __offset,
+                                        signed __int128 *__ptr) {
+  *(vector signed __int128 *)(__ptr + __offset) = __vec;
+}
+
+static inline __ATTRS_o_ai void vec_xst(vector unsigned __int128 __vec,
+                                        signed long long __offset,
+                                        unsigned __int128 *__ptr) {
+  *(vector unsigned __int128 *)(__ptr + __offset) = __vec;
+}
+#endif
 #undef __ATTRS_o_ai
 
 #endif /* __ALTIVEC_H */

test/CodeGen/builtins-ppc-altivec.c

 
 signed char param_sc;
 unsigned char param_uc;
 short param_s;
 unsigned short param_us;
 int param_i;
 unsigned int param_ui;
 float param_f;
+signed long long param_sll;
 
 int res_sc;
 int res_uc;
 int res_s;
 int res_us;
 int res_i;
 int res_ui;
 int res_f;
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> {{%.+}}, <4 x i32> {{%.+}}, <16 x i8> {{%.+}})
   res_vf = vec_revb(vf);
 // CHECK: store <16 x i8> <i8 3, i8 2, i8 1, i8 0, i8 7, i8 6, i8 5, i8 4, i8 11, i8 10, i8 9, i8 8, i8 15, i8 14, i8 13, i8 12>, <16 x i8>* {{%.+}}, align 16
 // CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> {{%.+}}, <4 x i32> {{%.+}}, <16 x i8> {{%.+}})
 // CHECK-LE: store <16 x i8> <i8 3, i8 2, i8 1, i8 0, i8 7, i8 6, i8 5, i8 4, i8 11, i8 10, i8 9, i8 8, i8 15, i8 14, i8 13, i8 12>, <16 x i8>* {{%.+}}, align 16
 // CHECK-LE: store <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, <16 x i8>* {{%.+}}, align 16
 // CHECK-LE: xor <16 x i8>
 // CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> {{%.+}}, <4 x i32> {{%.+}}, <16 x i8> {{%.+}})
 }
+
+/* ------------------------------ vec_xl ------------------------------------ */
+void test9() {
+  // CHECK-LABEL: define void @test9
+  // CHECK-LE-LABEL: define void @test9
+  res_vsc = vec_xl(param_sll, &param_sc);
+  // CHECK: load <16 x i8>, <16 x i8>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <16 x i8>, <16 x i8>* %{{[0-9]+}}, align 16
+
+  res_vuc = vec_xl(param_sll, &param_uc);
+  // CHECK: load <16 x i8>, <16 x i8>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <16 x i8>, <16 x i8>* %{{[0-9]+}}, align 16
+
+  res_vs = vec_xl(param_sll, &param_s);
+  // CHECK: load <8 x i16>, <8 x i16>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <8 x i16>, <8 x i16>* %{{[0-9]+}}, align 16
+
+  res_vus = vec_xl(param_sll, &param_us);
+  // CHECK: load <8 x i16>, <8 x i16>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <8 x i16>, <8 x i16>* %{{[0-9]+}}, align 16
+
+  res_vi = vec_xl(param_sll, &param_i);
+  // CHECK: load <4 x i32>, <4 x i32>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <4 x i32>, <4 x i32>* %{{[0-9]+}}, align 16
+
+  res_vui = vec_xl(param_sll, &param_ui);
+  // CHECK: load <4 x i32>, <4 x i32>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <4 x i32>, <4 x i32>* %{{[0-9]+}}, align 16
+
+  res_vf = vec_xl(param_sll, &param_f);
+  // CHECK: load <4 x float>, <4 x float>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <4 x float>, <4 x float>* %{{[0-9]+}}, align 16
+}
+
+/* ------------------------------ vec_xst ------------------------------------ */
+void test10() {
+  // CHECK-LABEL: define void @test10
+  // CHECK-LE-LABEL: define void @test10
+  vec_xst(vsc, param_sll, &param_sc);
+  // CHECK: store <16 x i8> %{{[0-9]+}}, <16 x i8>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <16 x i8> %{{[0-9]+}}, <16 x i8>* %{{[0-9]+}}, align 16
+
+  vec_xst(vuc, param_sll, &param_uc);
+  // CHECK: store <16 x i8> %{{[0-9]+}}, <16 x i8>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <16 x i8> %{{[0-9]+}}, <16 x i8>* %{{[0-9]+}}, align 16
+
+  vec_xst(vs, param_sll, &param_s);
+  // CHECK: store <8 x i16> %{{[0-9]+}}, <8 x i16>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <8 x i16> %{{[0-9]+}}, <8 x i16>* %{{[0-9]+}}, align 16
+
+  vec_xst(vus, param_sll, &param_us);
+  // CHECK: store <8 x i16> %{{[0-9]+}}, <8 x i16>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <8 x i16> %{{[0-9]+}}, <8 x i16>* %{{[0-9]+}}, align 16
+
+  vec_xst(vi, param_sll, &param_i);
+  // CHECK: store <4 x i32> %{{[0-9]+}}, <4 x i32>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <4 x i32> %{{[0-9]+}}, <4 x i32>* %{{[0-9]+}}, align 16
+
+  vec_xst(vui, param_sll, &param_ui);
+  // CHECK: store <4 x i32> %{{[0-9]+}}, <4 x i32>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <4 x i32> %{{[0-9]+}}, <4 x i32>* %{{[0-9]+}}, align 16
+
+  vec_xst(vf, param_sll, &param_f);
+  // CHECK: store <4 x float> %{{[0-9]+}}, <4 x float>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <4 x float> %{{[0-9]+}}, <4 x float>* %{{[0-9]+}}, align 16
+}

test/CodeGen/builtins-ppc-quadword.c

 // RUN: -emit-llvm %s -o - 2>&1 | FileCheck %s -check-prefix=CHECK-PPC
 #include <altivec.h>
 
 // CHECK-PPC: error: __int128 is not supported on this target
 vector signed __int128 vlll = { -1 };
 // CHECK-PPC: error: __int128 is not supported on this target
 vector unsigned __int128 vulll = { 1 };
 
+signed long long param_sll;
+// CHECK-PPC: error: __int128 is not supported on this target
+signed __int128 param_lll;
+// CHECK-PPC: error: __int128 is not supported on this target
+unsigned __int128 param_ulll;
+
 // CHECK-PPC: error: __int128 is not supported on this target
 vector signed __int128 res_vlll;
 // CHECK-PPC: error: __int128 is not supported on this target
 vector unsigned __int128 res_vulll;
 
 
 // CHECK-LABEL: define void @test1
 void test1() {
// CHECK-PPC: error: assigning to '__vector unsigned __int128' (vector of 1 'unsigned __int128' value) from incompatible type 'int'
   res_vulll = vec_revb(vulll);
 // CHECK: store <16 x i8> <i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>, <16 x i8>* {{%.+}}, align 16
 // CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> {{%.+}}, <4 x i32> {{%.+}}, <16 x i8> {{%.+}})
 // CHECK-LE: store <16 x i8> <i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>, <16 x i8>* {{%.+}}, align 16
 // CHECK-LE: store <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, <16 x i8>* {{%.+}}, align 16
 // CHECK-LE: xor <16 x i8>
 // CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> {{%.+}}, <4 x i32> {{%.+}}, <16 x i8> {{%.+}})
 // CHECK_PPC: error: call to 'vec_revb' is ambiguous
+
+  /* vec_xl */
+  res_vlll = vec_xl(param_sll, &param_lll);
+  // CHECK: load <1 x i128>, <1 x i128>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <1 x i128>, <1 x i128>* %{{[0-9]+}}, align 16
+  // CHECK-PPC: error: call to 'vec_xl' is ambiguous
+
+  res_vulll = vec_xl(param_sll, &param_ulll);
+  // CHECK: load <1 x i128>, <1 x i128>* %{{[0-9]+}}, align 16
+  // CHECK-LE: load <1 x i128>, <1 x i128>* %{{[0-9]+}}, align 16
+  // CHECK-PPC: error: call to 'vec_xl' is ambiguous
+
+  /* vec_xst */
+   vec_xst(vlll, param_sll, &param_lll);
+  // CHECK: store <1 x i128> %{{[0-9]+}}, <1 x i128>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <1 x i128> %{{[0-9]+}}, <1 x i128>* %{{[0-9]+}}, align 16
+  // CHECK-PPC: error: call to 'vec_xst' is ambiguous
+
+   vec_xst(vulll, param_sll, &param_ulll);
+  // CHECK: store <1 x i128> %{{[0-9]+}}, <1 x i128>* %{{[0-9]+}}, align 16
+  // CHECK-LE: store <1 x i128> %{{[0-9]+}}, <1 x i128>* %{{[0-9]+}}, align 16
+  // CHECK-PPC: error: call to 'vec_xst' is ambiguous
 }

test/CodeGen/builtins-ppc-vsx.c

 vector unsigned short vus = { 0, 1, 2, 3, 4, 5, 6, 7 };
 vector bool int vbi = { 0, 1, 0, 1 };
 vector signed int vsi = { -1, 2, -3, 4 };
 vector unsigned int vui = { 0, 1, 2, 3 };
 vector bool long long vbll = { 1, 0 };
 vector signed long long vsll = { 255LL, -937LL };
 vector unsigned long long vull = { 1447LL, 2894LL };
 double d = 23.4;
+signed long long sll = 618LL;
 float af[4] = {23.4f, 56.7f, 89.0f, 12.3f};
 double ad[2] = {23.4, 56.7};
 signed char asc[16] = { -8,  9, -10, 11, -12, 13, -14, 15,
                         -0,  1,  -2,  3,  -4,  5,  -6,  7};
 unsigned char auc[16] = { 8,  9, 10, 11, 12, 13, 14, 15,
                           0,  1,  2,  3,  4,  5,  6,  7};
 signed short ass[8] = { -1, 2, -3, 4, -5, 6, -7, 8 };
 unsigned short aus[8] = { 0, 1, 2, 3, 4, 5, 6, 7 };
 signed int asi[4] = { -1, 2, -3, 4 };
 unsigned int aui[4] = { 0, 1, 2, 3 };
-signed long asl[2] = { -1L, 2L };
-unsigned long aul[2] = { 1L, 2L };
+signed long long asll[2] = { -1L, 2L };
+unsigned long long aull[2] = { 1L, 2L };
 
 vector float res_vf;
 vector double res_vd;
 vector bool char res_vbc;
 vector signed char res_vsc;
 vector unsigned char res_vuc;
 vector bool short res_vbs;
 vector signed short res_vss;
// CHECK-LE: @llvm.ppc.altivec.vsro
 
   res_vull = vec_sro(vull, vsc);
 // CHECK: @llvm.ppc.altivec.vsro
 // CHECK-LE: @llvm.ppc.altivec.vsro
 
   res_vull = vec_sro(vull, vuc);
 // CHECK: @llvm.ppc.altivec.vsro
 // CHECK-LE: @llvm.ppc.altivec.vsro
+
+res_vsll = vec_xl(sll, asll);
+// CHECK: load <2 x i64>, <2 x i64>* %{{[0-9]+}}, align 16
+// CHECK-LE: load <2 x i64>, <2 x i64>* %{{[0-9]+}}, align 16
+
+res_vull = vec_xl(sll, aull);
+// CHECK: load <2 x i64>, <2 x i64>* %{{[0-9]+}}, align 16
+// CHECK-LE: load <2 x i64>, <2 x i64>* %{{[0-9]+}}, align 16
+
+res_vd = vec_xl(sll, ad);
+// CHECK: load <2 x double>, <2 x double>* %{{[0-9]+}}, align 16
+// CHECK-LE: load <2 x double>, <2 x double>* %{{[0-9]+}}, align 16
+
+vec_xst(vsll, sll, asll);
+// CHECK: store <2 x i64> %{{[0-9]+}}, <2 x i64>* %{{[0-9]+}}, align 16
+// CHECK-LE: store <2 x i64> %{{[0-9]+}}, <2 x i64>* %{{[0-9]+}}, align 16
+
+vec_xst(vull, sll, aull);
+// CHECK: store <2 x i64> %{{[0-9]+}}, <2 x i64>* %{{[0-9]+}}, align 16
+// CHECK-LE: store <2 x i64> %{{[0-9]+}}, <2 x i64>* %{{[0-9]+}}, align 16
+
+vec_xst(vd, sll, ad);
+// CHECK: store <2 x double> %{{[0-9]+}}, <2 x double>* %{{[0-9]+}}, align 16
+// CHECK-LE: store <2 x double> %{{[0-9]+}}, <2 x double>* %{{[0-9]+}}, align 16
 }