I'm going to get the index of max element in float<4> in avx2-i32x8.
and here is my code.
*export uniform int YetAnother(uniform float<4>& InValue){ varying float
src; *((uniform float<4> *uniform)&src) = *((uniform float<4>
*uniform)&InValue); varying float max = reduce_max(src); uniform
int IMMask; if(src == max) { IMMask = lanemask(); }
return count_trailing_zeros(IMMask);}*
i get the asm by this command
*ispc -O2 -g --target=avx2-i32x8 --emit-asm*
and the asm code like below
* .def YetAnother; .scl 2; .type 32; .endef .globl
YetAnother # -- Begin function YetAnother
.p2align 4, 0x90YetAnother: #
@YetAnother.Lfunc_begin14: .cv_func_id 39# %bb.0:
# %allocas #DEBUG_VALUE: YetAnother:InValue <- undef
.cv_inline_site_id 40 within 39 inlined_at 1 0 0 .cv_loc 40 2 1354 18
# stdlib.ispc:1354:18 vmovaps (%rcx),
%xmm0.Ltmp256: #DEBUG_VALUE: test <- undef #DEBUG_VALUE: __mask <-
undef #DEBUG_VALUE: reduce_max:v <- $ymm0 vmaxps %xmm0, %xmm0,
%xmm1 vpermilpd $1, %xmm1, %xmm2 # xmm2 = xmm1[1,0] vmaxps
%xmm1, %xmm2, %xmm1 vmovshdup %xmm1, %xmm2 # xmm2 =
xmm1[1,1,3,3] vmaxss %xmm1, %xmm2, %xmm1.Ltmp257: #DEBUG_VALUE:
result <- $xmm1 #DEBUG_VALUE: __mask <- undef vbroadcastss %xmm1,
%ymm1.Ltmp258: #DEBUG_VALUE: __mask <- undef #DEBUG_VALUE: max <-
$ymm1 vcmpeqps %ymm1, %ymm0, %ymm0.Ltmp259: vmovmskps %ymm0,
%eax.Ltmp260: #DEBUG_VALUE: __mask <- undef #DEBUG_VALUE:
count_trailing_zeros:v <- $eax #DEBUG_VALUE: IMMask <- $eax
#DEBUG_VALUE: src <- undef #DEBUG_VALUE: __mask <- undef
.cv_inline_site_id 41 within 39 inlined_at 1 196 9 .cv_loc 41 2 877
12 # stdlib.ispc:877:12 tzcntl %eax,
%eax.Ltmp261: #DEBUG_VALUE: iflt_neg_max <- undef #DEBUG_VALUE:
__mask <- undef .cv_loc 39 1 196 9 #
simple.ispc:196:9 vzeroupper.Ltmp262: retq.Ltmp263:.Lfunc_end14:
# -- End function*
I' m wondering if it is the best way to do that.
and the asm use ymm1 register.
how can i use only xmm1 to do vcmpeqps and vmovmskps.
I don‘t want to change target to avx2-i32x4
I want't to keep the TARGET_WIDTH = 8 to get other function in my programe
effecient
--
You received this message because you are subscribed to the Google Groups
"Intel SPMD Program Compiler Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/ispc-users/7d3ea55b-0e3f-4305-996b-8b4536fa1331n%40googlegroups.com.
