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

Extend gather-scatter optimization with prefetch optimization

Browse Source
This commit is contained in:
Vsevolod Livinskiy
2014-09-25 15:59:31 +04:00
parent 8ff187a7b7
commit 0a6eb61ad0
9 changed files with 204 additions and 127 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
builtins/util.m4
View File

@@ -1584,6 +1584,17 @@ define void @__prefetch_read_uniform_nt(i8 *) alwaysinline {
call void @llvm.prefetch(i8 * %0, i32 0, i32 0, i32 1)
ret void
}
define void @__prefetch_read_varying_1(<WIDTH x i64> %addr, <WIDTH x MASK> %mask) alwaysinline {
per_lane(WIDTH, <WIDTH x MASK> %mask, `
%iptr_LANE_ID = extractelement <WIDTH x i64> %addr, i32 LANE
%ptr_LANE_ID = inttoptr i64 %iptr_LANE_ID to i8*
call void @llvm.prefetch(i8 * %ptr_LANE_ID, i32 0, i32 3, i32 1)
')
ret void
}
declare void @__prefetch_read_varying_1_native(i8 * %base, i32 %scale, <WIDTH x i32> %offsets, <WIDTH x MASK> %mask) nounwind
')
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
@@ -2535,6 +2546,13 @@ declare void
@__pseudo_scatter_base_offsets64_double(i8 * nocapture, i32, <WIDTH x i64>,
<WIDTH x double>, <WIDTH x MASK>) nounwind
declare void @__pseudo_prefetch_read_varying_1(<WIDTH x i64>, <WIDTH x MASK>) nounwind
declare void
@__pseudo_prefetch_read_varying_1_native(i8 *, i32, <WIDTH x i32>,
<WIDTH x MASK>) nounwind
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
declare void @__use8(<WIDTH x i8>)
@@ -3034,6 +3052,17 @@ ifelse(HAVE_SCATTER, `1',
<WIDTH x double> %vd, <WIDTH x MASK> %mask)
')
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; prefetchs
call void @__pseudo_prefetch_read_varying_1(<WIDTH x i64> %v64, <WIDTH x MASK> %mask)
call void @__pseudo_prefetch_read_varying_1_native(i8 * %ptr, i32 0,
<WIDTH x i32> %v32, <WIDTH x MASK> %mask)
call void @__prefetch_read_varying_1_native(i8 * %ptr, i32 0,
<WIDTH x i32> %v32, <WIDTH x MASK> %mask)
call void @__prefetch_read_varying_1(<WIDTH x i64> %v64, <WIDTH x MASK> %mask)
ret void
}