aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

structural change

Browse Source
This commit is contained in:
Vsevolod Livinskij
2013-11-30 10:48:18 +04:00
parent 4c330bc38b
commit 4faff1a63c
13 changed files with 285 additions and 660 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
builtins/target-sse4.ll
View File

@@ -41,76 +41,10 @@ stdlib_core()
packed_load_and_store()
scans()
int64minmax()
saturation_arithmetic_vec4()
include(`target-sse4-common.ll')
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;vector saturation arithmetic
define <WIDTH x i8> @__padds_vi8(<WIDTH x i8>, <WIDTH x i8>) {
convert4to16(i8, %0, %v0)
convert4to16(i8, %1, %v1)
%r16 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %v0, <16 x i8> %v1)
convert16to4(i8, %r16, %r)
ret <WIDTH x i8> %r
}
define <WIDTH x i16> @__padds_vi16(<WIDTH x i16>, <WIDTH x i16>) {
convert4to8(i16, %0, %v0)
convert4to8(i16, %1, %v1)
%r16 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %v0, <8 x i16> %v1)
convert8to4(i16, %r16, %r)
ret <WIDTH x i16> %r
}
define <WIDTH x i8> @__paddus_vi8(<WIDTH x i8>, <WIDTH x i8>) {
convert4to16(i8, %0, %v0)
convert4to16(i8, %1, %v1)
%r16 = call <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8> %v0, <16 x i8> %v1)
convert16to4(i8, %r16, %r)
ret <WIDTH x i8> %r
}
define <WIDTH x i16> @__paddus_vi16(<WIDTH x i16>, <WIDTH x i16>) {
convert4to8(i16, %0, %v0)
convert4to8(i16, %1, %v1)
%r16 = call <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16> %v0, <8 x i16> %v1)
convert8to4(i16, %r16, %r)
ret <WIDTH x i16> %r
}
define <WIDTH x i8> @__psubs_vi8(<WIDTH x i8>, <WIDTH x i8>) {
convert4to16(i8, %0, %v0)
convert4to16(i8, %1, %v1)
%r16 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %v0, <16 x i8> %v1)
convert16to4(i8, %r16, %r)
ret <WIDTH x i8> %r
}
define <WIDTH x i16> @__psubs_vi16(<WIDTH x i16>, <WIDTH x i16>) {
convert4to8(i16, %0, %v0)
convert4to8(i16, %1, %v1)
%r16 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %v0, <8 x i16> %v1)
convert8to4(i16, %r16, %r)
ret <WIDTH x i16> %r
}
define <WIDTH x i8> @__psubus_vi8(<WIDTH x i8>, <WIDTH x i8>) {
convert4to16(i8, %0, %v0)
convert4to16(i8, %1, %v1)
%r16 = call <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8> %v0, <16 x i8> %v1)
convert16to4(i8, %r16, %r)
ret <WIDTH x i8> %r
}
define <WIDTH x i16> @__psubus_vi16(<WIDTH x i16>, <WIDTH x i16>) {
convert4to8(i16, %0, %v0)
convert4to8(i16, %1, %v1)
%r16 = call <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16> %v0, <8 x i16> %v1)
convert8to4(i16, %r16, %r)
ret <WIDTH x i16> %r
}
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; half conversion routines