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

working on target-nvptx64... need to add nvptx64

Browse Source
This commit is contained in:
egaburov
2013-07-28 15:50:08 +02:00
parent 67b549a937
commit af61c9bae3
1 changed files with 12 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
builtins/target-nvptx64.ll
View File

@@ -319,6 +319,8 @@ declare float @llvm.exp.f32(float %Val)
declare float @llvm.log.f32(float %Val) declare float @llvm.log.f32(float %Val)
declare float @llvm.pow.f32(float %f, float %e) declare float @llvm.pow.f32(float %f, float %e)
declare float @llvm.nvvm.rsqrt.approx.f(float %f) nounwind readonly alwaysinline
@@ -624,6 +626,8 @@ define <1 x float> @__sqrt_varying_float(<1 x float>) nounwind readonly alwaysi
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; rsqrt ; rsqrt
define <1 x float> @__rsqrt_varying_float(<1 x float> %v) nounwind readonly alwaysinline { define <1 x float> @__rsqrt_varying_float(<1 x float> %v) nounwind readonly alwaysinline {
; float is = __rsqrt_v(v); ; float is = __rsqrt_v(v);
;%is = call <1 x float> @llvm.x86.sse.rsqrt.ps(<1 x float> %v) ;%is = call <1 x float> @llvm.x86.sse.rsqrt.ps(<1 x float> %v)
@@ -635,10 +639,14 @@ define <1 x float> @__rsqrt_varying_float(<1 x float> %v) nounwind readonly alw
;%is_mul = fmul <1 x float> %is, %three_sub ;%is_mul = fmul <1 x float> %is, %three_sub
;%half_scale = fmul <1 x float> <float 0.5, float 0.5, float 0.5, float 0.5>, %is_mul ;%half_scale = fmul <1 x float> <float 0.5, float 0.5, float 0.5, float 0.5>, %is_mul
;ret <1 x float> %half_scale ;ret <1 x float> %half_scale
%s = call <1 x float> @__sqrt_varying_float(<1 x float> %v) ;%s = call <1 x float> @__sqrt_varying_float(<1 x float> %v)
%r = call <1 x float> @__rcp_varying_float(<1 x float> %s) ;%r = call <1 x float> @__rcp_varying_float(<1 x float> %s)
ret <1 x float> %r ;ret <1 x float> %r
%vs = extractelement <1 x float> %v, i32 0
%rs = call float @llvm.nvvm.rsqrt.approx.f(float %vs)
; %rs = call float asm "rsqrt.approx.f32 $0,$0", "=f,f"(float %vs)
%rv = insertelement <1 x float> undef , float %rs, i32 0
ret <1 x float> %rv
} }