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

Add reduce_add() for int8 and int16 types.

Browse Source 
This maps to specialized instructions (e.g. PSADBW) when available.
This commit is contained in:
Matt Pharr
2013-07-25 09:11:39 -07:00
parent 2d063925a1
commit b6df447b55
24 changed files with 464 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
builtins/target-sse4-x2.ll
View File

@@ -309,6 +309,36 @@ define i1 @__none(<8 x i32>) nounwind readnone alwaysinline {
ret i1 %cmp
}
declare <2 x i64> @llvm.x86.sse2.psad.bw(<16 x i8>, <16 x i8>) nounwind readnone
define i16 @__reduce_add_int8(<8 x i8>) nounwind readnone alwaysinline {
%wide8 = shufflevector <8 x i8> %0, <8 x i8> zeroinitializer,
<16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7,
i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
%rv = call <2 x i64> @llvm.x86.sse2.psad.bw(<16 x i8> %wide8,
<16 x i8> zeroinitializer)
%r0 = extractelement <2 x i64> %rv, i32 0
%r1 = extractelement <2 x i64> %rv, i32 1
%r = add i64 %r0, %r1
%r16 = trunc i64 %r to i16
ret i16 %r16
}
define internal <8 x i16> @__add_varying_i16(<8 x i16>,
<8 x i16>) nounwind readnone alwaysinline {
%r = add <8 x i16> %0, %1
ret <8 x i16> %r
}
define internal i16 @__add_uniform_i16(i16, i16) nounwind readnone alwaysinline {
%r = add i16 %0, %1
ret i16 %r
}
define i16 @__reduce_add_int16(<8 x i16>) nounwind readnone alwaysinline {
reduce8(i16, @__add_varying_i16, @__add_uniform_i16)
}
define float @__reduce_min_float(<8 x float>) nounwind readnone alwaysinline {
reduce8by4(float, @llvm.x86.sse.min.ps, @__min_uniform_float)
}