[AVX512]: definitions through util.m4 was added
This commit is contained in:
Vsevolod Livinskiy
committed by
Anton Mitrokhin
Anton Mitrokhin
parent
46528caa5a
commit
9a03cd3590
@@ -29,8 +29,6 @@
|
||||
;; NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
||||
;; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
|
||||
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128-v16:16:16-v32:32:32-v4:128:128";
|
||||
|
||||
define(`MASK',`i1')
|
||||
define(`HAVE_GATHER',`1')
|
||||
define(`HAVE_SCATTER',`1')
|
||||
@@ -40,7 +38,7 @@ include(`util.m4')
|
||||
stdlib_core()
|
||||
scans()
|
||||
reduce_equal(WIDTH)
|
||||
rdrand_decls()
|
||||
rdrand_definition()
|
||||
|
||||
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
||||
;; broadcast/rotate/shuffle
|
||||
@@ -66,64 +64,15 @@ declare <WIDTH x i16> @__undef_i16() nounwind readnone
|
||||
declare <WIDTH x i32> @__undef_i32() nounwind readnone
|
||||
declare <WIDTH x i64> @__undef_i64() nounwind readnone
|
||||
|
||||
declare <WIDTH x float> @__broadcast_float(<WIDTH x float>, i32) nounwind readnone
|
||||
declare <WIDTH x double> @__broadcast_double(<WIDTH x double>, i32) nounwind readnone
|
||||
declare <WIDTH x i8> @__broadcast_i8(<WIDTH x i8>, i32) nounwind readnone
|
||||
declare <WIDTH x i16> @__broadcast_i16(<WIDTH x i16>, i32) nounwind readnone
|
||||
declare <WIDTH x i32> @__broadcast_i32(<WIDTH x i32>, i32) nounwind readnone
|
||||
declare <WIDTH x i64> @__broadcast_i64(<WIDTH x i64>, i32) nounwind readnone
|
||||
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
||||
;; shuffle
|
||||
|
||||
declare <WIDTH x i8> @__rotate_i8(<WIDTH x i8>, i32) nounwind readnone
|
||||
declare <WIDTH x i16> @__rotate_i16(<WIDTH x i16>, i32) nounwind readnone
|
||||
declare <WIDTH x float> @__rotate_float(<WIDTH x float>, i32) nounwind readnone
|
||||
declare <WIDTH x i32> @__rotate_i32(<WIDTH x i32>, i32) nounwind readnone
|
||||
declare <WIDTH x double> @__rotate_double(<WIDTH x double>, i32) nounwind readnone
|
||||
declare <WIDTH x i64> @__rotate_i64(<WIDTH x i64>, i32) nounwind readnone
|
||||
|
||||
declare <WIDTH x i8> @__shift_i8(<WIDTH x i8>, i32) nounwind readnone
|
||||
declare <WIDTH x i16> @__shift_i16(<WIDTH x i16>, i32) nounwind readnone
|
||||
declare <WIDTH x float> @__shift_float(<WIDTH x float>, i32) nounwind readnone
|
||||
declare <WIDTH x i32> @__shift_i32(<WIDTH x i32>, i32) nounwind readnone
|
||||
declare <WIDTH x double> @__shift_double(<WIDTH x double>, i32) nounwind readnone
|
||||
declare <WIDTH x i64> @__shift_i64(<WIDTH x i64>, i32) nounwind readnone
|
||||
|
||||
declare <WIDTH x i8> @__shuffle_i8(<WIDTH x i8>, <WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i8> @__shuffle2_i8(<WIDTH x i8>, <WIDTH x i8>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i16> @__shuffle_i16(<WIDTH x i16>, <WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i16> @__shuffle2_i16(<WIDTH x i16>, <WIDTH x i16>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x float> @__shuffle_float(<WIDTH x float>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x float> @__shuffle2_float(<WIDTH x float>, <WIDTH x float>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i32> @__shuffle_i32(<WIDTH x i32>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i32> @__shuffle2_i32(<WIDTH x i32>, <WIDTH x i32>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x double> @__shuffle_double(<WIDTH x double>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x double> @__shuffle2_double(<WIDTH x double>,
|
||||
<WIDTH x double>, <WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i64> @__shuffle_i64(<WIDTH x i64>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i64> @__shuffle2_i64(<WIDTH x i64>, <WIDTH x i64>,
|
||||
<WIDTH x i32>) nounwind readnone
|
||||
define_shuffles()
|
||||
|
||||
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
||||
;; aos/soa
|
||||
|
||||
declare void @__soa_to_aos3_float(<WIDTH x float> %v0, <WIDTH x float> %v1,
|
||||
<WIDTH x float> %v2, float * noalias %p) nounwind
|
||||
declare void @__aos_to_soa3_float(float * noalias %p, <WIDTH x float> * %out0,
|
||||
<WIDTH x float> * %out1, <WIDTH x float> * %out2) nounwind
|
||||
declare void @__soa_to_aos4_float(<WIDTH x float> %v0, <WIDTH x float> %v1,
|
||||
<WIDTH x float> %v2, <WIDTH x float> %v3,
|
||||
float * noalias %p) nounwind
|
||||
declare void @__aos_to_soa4_float(float * noalias %p, <WIDTH x float> * noalias %out0,
|
||||
<WIDTH x float> * noalias %out1,
|
||||
<WIDTH x float> * noalias %out2,
|
||||
<WIDTH x float> * noalias %out3) nounwind
|
||||
aossoa()
|
||||
|
||||
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
||||
;; half conversion routines
|
||||
@@ -157,16 +106,14 @@ declare <WIDTH x double> @__ceil_varying_double(<WIDTH x double>) nounwind readn
|
||||
|
||||
;; min/max
|
||||
|
||||
int64minmax()
|
||||
|
||||
declare float @__max_uniform_float(float, float) nounwind readnone
|
||||
declare float @__min_uniform_float(float, float) nounwind readnone
|
||||
declare i32 @__min_uniform_int32(i32, i32) nounwind readnone
|
||||
declare i32 @__max_uniform_int32(i32, i32) nounwind readnone
|
||||
declare i32 @__min_uniform_uint32(i32, i32) nounwind readnone
|
||||
declare i32 @__max_uniform_uint32(i32, i32) nounwind readnone
|
||||
declare i64 @__min_uniform_int64(i64, i64) nounwind readnone
|
||||
declare i64 @__max_uniform_int64(i64, i64) nounwind readnone
|
||||
declare i64 @__min_uniform_uint64(i64, i64) nounwind readnone
|
||||
declare i64 @__max_uniform_uint64(i64, i64) nounwind readnone
|
||||
declare double @__min_uniform_double(double, double) nounwind readnone
|
||||
declare double @__max_uniform_double(double, double) nounwind readnone
|
||||
|
||||
@@ -178,10 +125,6 @@ declare <WIDTH x i32> @__min_varying_int32(<WIDTH x i32>, <WIDTH x i32>) nounwin
|
||||
declare <WIDTH x i32> @__max_varying_int32(<WIDTH x i32>, <WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i32> @__min_varying_uint32(<WIDTH x i32>, <WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i32> @__max_varying_uint32(<WIDTH x i32>, <WIDTH x i32>) nounwind readnone
|
||||
declare <WIDTH x i64> @__min_varying_int64(<WIDTH x i64>, <WIDTH x i64>) nounwind readnone
|
||||
declare <WIDTH x i64> @__max_varying_int64(<WIDTH x i64>, <WIDTH x i64>) nounwind readnone
|
||||
declare <WIDTH x i64> @__min_varying_uint64(<WIDTH x i64>, <WIDTH x i64>) nounwind readnone
|
||||
declare <WIDTH x i64> @__max_varying_uint64(<WIDTH x i64>, <WIDTH x i64>) nounwind readnone
|
||||
declare <WIDTH x double> @__min_varying_double(<WIDTH x double>,
|
||||
<WIDTH x double>) nounwind readnone
|
||||
declare <WIDTH x double> @__max_varying_double(<WIDTH x double>,
|
||||
@@ -205,10 +148,7 @@ declare <WIDTH x double> @__sqrt_varying_double(<WIDTH x double>) nounwind readn
|
||||
declare i32 @__popcnt_int32(i32) nounwind readnone
|
||||
declare i64 @__popcnt_int64(i64) nounwind readnone
|
||||
|
||||
declare i32 @__count_trailing_zeros_i32(i32) nounwind readnone
|
||||
declare i64 @__count_trailing_zeros_i64(i64) nounwind readnone
|
||||
declare i32 @__count_leading_zeros_i32(i32) nounwind readnone
|
||||
declare i64 @__count_leading_zeros_i64(i64) nounwind readnone
|
||||
ctlztz()
|
||||
|
||||
; FIXME: need either to wire these up to the 8-wide SVML entrypoints,
|
||||
; or, use the macro to call the 4-wide ones twice with our 8-wide
|
||||
@@ -354,30 +294,13 @@ gather_scatter(float)
|
||||
gather_scatter(i64)
|
||||
gather_scatter(double)
|
||||
|
||||
declare i32 @__packed_load_active(i32 * nocapture, <WIDTH x i32> * nocapture,
|
||||
<WIDTH x i1>) nounwind
|
||||
declare i32 @__packed_store_active(i32 * nocapture, <WIDTH x i32> %vals,
|
||||
<WIDTH x i1>) nounwind
|
||||
declare i32 @__packed_store_active2(i32 * nocapture, <WIDTH x i32> %vals,
|
||||
<WIDTH x i1>) nounwind
|
||||
|
||||
packed_load_and_store()
|
||||
|
||||
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
||||
;; prefetch
|
||||
|
||||
declare void @__prefetch_read_uniform_1(i8 * nocapture) nounwind
|
||||
declare void @__prefetch_read_uniform_2(i8 * nocapture) nounwind
|
||||
declare void @__prefetch_read_uniform_3(i8 * nocapture) nounwind
|
||||
declare void @__prefetch_read_uniform_nt(i8 * nocapture) nounwind
|
||||
define_prefetches()
|
||||
|
||||
declare void @__prefetch_read_varying_1(<WIDTH x i64> %addr, <WIDTH x MASK> %mask) nounwind
|
||||
declare void @__prefetch_read_varying_1_native(i8 * %base, i32 %scale, <WIDTH x i32> %offsets, <WIDTH x MASK> %mask) nounwind
|
||||
declare void @__prefetch_read_varying_2(<WIDTH x i64> %addr, <WIDTH x MASK> %mask) nounwind
|
||||
declare void @__prefetch_read_varying_2_native(i8 * %base, i32 %scale, <WIDTH x i32> %offsets, <WIDTH x MASK> %mask) nounwind
|
||||
declare void @__prefetch_read_varying_3(<WIDTH x i64> %addr, <WIDTH x MASK> %mask) nounwind
|
||||
declare void @__prefetch_read_varying_3_native(i8 * %base, i32 %scale, <WIDTH x i32> %offsets, <WIDTH x MASK> %mask) nounwind
|
||||
declare void @__prefetch_read_varying_nt(<WIDTH x i64> %addr, <WIDTH x MASK> %mask) nounwind
|
||||
declare void @__prefetch_read_varying_nt_native(i8 * %base, i32 %scale, <WIDTH x i32> %offsets, <WIDTH x MASK> %mask) nounwind
|
||||
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
||||
;; int8/int16 builtins
|
||||
|
||||
|
||||
Reference in New Issue
Block a user