[PATCH] D26534: [PPC] add altivec.h functions for converting a vector of half precision to a vector of single precision
Sean Fertile via cfe-commits
cfe-commits at lists.llvm.org
Thu Nov 10 18:06:48 PST 2016
sfertile created this revision.
sfertile added reviewers: nemanjai, amehsan, kbarton, syzaara, jtony, lei.
sfertile added subscribers: cfe-commits, echristo.
sfertile set the repository for this revision to rL LLVM.
Change adds 2 functions to altivec.h
vector float vec_extract_fp32_from_shorth (vector unsigned short);
Purpose:
Extracts four single-precision floating-point numbers from the high elements of a vector of eight 16-bit elements, interpreting each element as a 16-bit floating-point number in IEEE format.
Result value:
The first four elements are interpreted as 16-bit floating-point numbers in IEEE format, and extended to single-precision format, returning a vector with four single-precision IEEE numbers.
vector float vec_extract_fp32_from_shortl (vector unsigned short);
Purpose
Extracts four single-precision floating-point numbers from the low elements of a vector of eight 16-bit elements, interpreting each element as a 16-bit floating-point number in IEEE format.
Result value:
The last four elements are interpreted as 16-bit floating-point numbers in IEEE format, and extended to single-precision format, returning a vector with four single-precision IEEE numbers.
Adds a builtin that gets mapped to the xscvhpsp instruction. (VSX Scalar Convert Half-Precision to Double-Precision)
Repository:
rL LLVM
https://reviews.llvm.org/D26534
Files:
include/clang/Basic/BuiltinsPPC.def
lib/Headers/altivec.h
test/CodeGen/builtins-ppc-p9vector.c
Index: test/CodeGen/builtins-ppc-p9vector.c
===================================================================
--- test/CodeGen/builtins-ppc-p9vector.c
+++ test/CodeGen/builtins-ppc-p9vector.c
@@ -827,4 +827,21 @@
// CHECK-NEXT: ret <16 x i8>
return vec_srv (vuca, vucb);
}
-
+vector float test74(void) {
+// CHECK-BE: shufflevector <8 x i16> {{.+}}, <8 x i16> {{.+}}, <8 x i32> <i32 undef, i32 0, i32 undef, i32 1, i32 undef, i32 2, i32 undef, i32 3>
+// CHECK-BE: @llvm.ppc.vsx.xvcvhpsp(<8 x i16> {{.+}})
+// CHECK-BE-NEXT: ret <4 x float>
+// CHECK-LE: shufflevector <8 x i16> {{.+}}, <8 x i16> {{.+}}, <8 x i32> <i32 0, i32 undef, i32 1, i32 undef, i32 2, i32 undef, i32 3, i32 undef>
+// CHECK-LE: @llvm.ppc.vsx.xvcvhpsp(<8 x i16> {{.+}})
+// CHECK-LE-NEXT: ret <4 x float>
+ return vec_extract_fp32_from_shorth(vusa);
+}
+vector float test75(void) {
+// CHECK-BE: shufflevector <8 x i16> {{.+}}, <8 x i16> {{.+}}, <8 x i32> <i32 undef, i32 4, i32 undef, i32 5, i32 undef, i32 6, i32 undef, i32 7>
+// CHECK-BE: @llvm.ppc.vsx.xvcvhpsp(<8 x i16> {{.+}})
+// CHECK-BE-NEXT: ret <4 x float>
+// CHECK-LE: shufflevector <8 x i16> {{.+}}, <8 x i16> {{.+}}, <8 x i32> <i32 4, i32 undef, i32 5, i32 undef, i32 6, i32 undef, i32 7, i32 undef>
+// CHECK-LE: @llvm.ppc.vsx.xvcvhpsp(<8 x i16> {{.+}})
+// CHECK-LE-NEXT: ret <4 x float>
+ return vec_extract_fp32_from_shortl(vusa);
+}
Index: lib/Headers/altivec.h
===================================================================
--- lib/Headers/altivec.h
+++ lib/Headers/altivec.h
@@ -11902,6 +11902,30 @@
return __a[__b];
}
+#if defined(__POWER9_VECTOR__)
+static __inline__ vector float __ATTRS_o_ai
+vec_extract_fp32_from_shorth(vector unsigned short __a) {
+ vector unsigned short __b =
+#ifdef __LITTLE_ENDIAN__
+ __builtin_shufflevector(__a, __a, 0, -1, 1, -1, 2, -1, 3, -1);
+#else
+ __builtin_shufflevector(__a, __a, -1, 0, -1, 1, -1, 2, -1, 3);
+#endif
+ return __builtin_vsx_xvcvhpsp(__b);
+}
+
+static __inline__ vector float __ATTRS_o_ai
+vec_extract_fp32_from_shortl(vector unsigned short __a) {
+ vector unsigned short __b =
+#ifdef __LITTLE_ENDIAN__
+ __builtin_shufflevector(__a, __a, 4, -1, 5, -1, 6, -1, 7, -1);
+#else
+ __builtin_shufflevector(__a, __a, -1, 4, -1, 5, -1, 6, -1, 7);
+#endif
+ return __builtin_vsx_xvcvhpsp(__b);
+}
+#endif /* __POWER9_VECTOR__ */
+
/* vec_insert */
static __inline__ vector signed char __ATTRS_o_ai
Index: include/clang/Basic/BuiltinsPPC.def
===================================================================
--- include/clang/Basic/BuiltinsPPC.def
+++ include/clang/Basic/BuiltinsPPC.def
@@ -380,6 +380,8 @@
BUILTIN(__builtin_vsx_xviexpdp, "V2dV2ULLiV2ULLi", "")
BUILTIN(__builtin_vsx_xviexpsp, "V4fV4UiV4Ui", "")
+BUILTIN(__builtin_vsx_xvcvhpsp, "V4fV8Us", "")
+
// HTM builtins
BUILTIN(__builtin_tbegin, "UiUIi", "")
BUILTIN(__builtin_tend, "UiUIi", "")
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D26534.77581.patch
Type: text/x-patch
Size: 2930 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20161111/177f6fbb/attachment.bin>
More information about the cfe-commits
mailing list