Compiler Explorer

Source code

; Function Attrs: nounwind readnone
declare i32 @llvm.x86.sse.movmsk.ps(<4 x float>) #0

; Function Attrs: noinline nounwind
define void @__soa_to_aos3_double4(<4 x double> %v0, <4 x double> %v1, <4 x double> %v2, <4 x double>* noalias %out0, <4 x double>* noalias %out1, <4 x double>* noalias %out2) #1 {
  %t0 = shufflevector <4 x double> %v0, <4 x double> %v1, <4 x i32> <i32 0, i32 1, i32 2, i32 4> 
  %t1 = shufflevector <4 x double> %v1, <4 x double> %v2, <4 x i32> <i32 1, i32 2, i32 4, i32 5>
  %r0 = shufflevector <4 x double> %t0, <4 x double> %t1, <4 x i32> <i32 0, i32 3, i32 6, i32 1>
  store <4 x double> %r0, <4 x double>* %out0, align 32
  %r1 = shufflevector <4 x double> %t0, <4 x double> %t1, <4 x i32> <i32 4, i32 7, i32 2, i32 5>
  store <4 x double> %r1, <4 x double>* %out1, align 32
  %t2 = shufflevector <4 x double> %v0, <4 x double> %v1, <4 x i32> <i32 3, i32 7, i32 undef, i32 undef>
  %r2 = shufflevector <4 x double> %t2, <4 x double> %v2, <4 x i32> <i32 6, i32 0, i32 1, i32 7>
  store <4 x double> %r2, <4 x double>* %out2, align 32
  ret void
}

; Function Attrs: noinline nounwind
define void @__soa_to_aos3_double(<4 x double> %v0, <4 x double> %v1, <4 x double> %v2, double* noalias %p) #1 {
  %out0 = bitcast double* %p to <4 x double>*
  %out1 = getelementptr <4 x double>, <4 x double>* %out0, i32 1
  %out2 = getelementptr <4 x double>, <4 x double>* %out0, i32 2
  call void @__soa_to_aos3_double4(<4 x double> %v0, <4 x double> %v1, <4 x double> %v2, <4 x double>* %out0, <4 x double>* %out1, <4 x double>* %out2)
  ret void
}

; Function Attrs: noinline nounwind
define internal void @soa_to_aos3___vydvydvydun_3C_und_3E_(<4 x double> %v0, <4 x double> %v1, <4 x double> %v2, double* noalias %a, <4 x i32> %__mask) #1 {
allocas:
  %internal_mask_memory = alloca <4 x i32>, align 16
  %full_mask_memory = alloca <4 x i32>, align 16
  %returned_lanes_memory = alloca <4 x i32>, align 16
  %launch_group_handle = alloca i8*, align 8
  %v01 = alloca <4 x double>, align 32
  %v12 = alloca <4 x double>, align 32
  %v23 = alloca <4 x double>, align 32
  %a4 = alloca double*, align 8
  store <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32>* %internal_mask_memory, align 16
  store <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32>* %full_mask_memory, align 16
  store <4 x i32> zeroinitializer, <4 x i32>* %returned_lanes_memory, align 16
  store i8* null, i8** %launch_group_handle, align 8
  store <4 x double> %v0, <4 x double>* %v01, align 32
  store <4 x double> %v1, <4 x double>* %v12, align 32
  store <4 x double> %v2, <4 x double>* %v23, align 32
  store double* %a, double** %a4, align 8
  %load_mask = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask" = and <4 x i32> %load_mask, %__mask
  store <4 x i32> %"internal_mask&function_mask", <4 x i32>* %full_mask_memory, align 16
  %v0_load = load <4 x double>, <4 x double>* %v01, align 32
  %v1_load = load <4 x double>, <4 x double>* %v12, align 32
  %v2_load = load <4 x double>, <4 x double>* %v23, align 32
  %a_load = load double*, double** %a4, align 8
  call void @__soa_to_aos3_double(<4 x double> %v0_load, <4 x double> %v1_load, <4 x double> %v2_load, double* %a_load)
  ret void
}

; Function Attrs: nounwind
define void @f_v___un_3C_unf_3E_(float* noalias %RET, <4 x i32> %__mask) #2 {
allocas:
  %internal_mask_memory = alloca <4 x i32>, align 16
  %full_mask_memory = alloca <4 x i32>, align 16
  %returned_lanes_memory = alloca <4 x i32>, align 16
  %launch_group_handle = alloca i8*, align 8
  %RET1 = alloca float*, align 8
  %a = alloca [12 x double], align 32
  %r = alloca [12 x double], align 32
  %i = alloca i32, align 4
  %x = alloca <4 x double>, align 32
  %y = alloca <4 x double>, align 32
  %z = alloca <4 x double>, align 32
  %a22 = alloca [12 x double], align 32
  %r23 = alloca [12 x double], align 32
  %i29 = alloca i32, align 4
  %x35 = alloca <4 x double>, align 32
  %y36 = alloca <4 x double>, align 32
  %z37 = alloca <4 x double>, align 32
  store <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32>* %internal_mask_memory, align 16
  store <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32>* %full_mask_memory, align 16
  store <4 x i32> zeroinitializer, <4 x i32>* %returned_lanes_memory, align 16
  store i8* null, i8** %launch_group_handle, align 8
  store float* %RET, float** %RET1, align 8
  %load_mask = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask" = and <4 x i32> %load_mask, %__mask
  store <4 x i32> %"internal_mask&function_mask", <4 x i32>* %full_mask_memory, align 16
  %floatmask.i = bitcast <4 x i32> %__mask to <4 x float>
  %v.i = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %floatmask.i) #0
  %cmp.i = icmp eq i32 %v.i, 15
  br i1 %cmp.i, label %all_on, label %some_on

all_on:                                           ; preds = %allocas
  %load_mask2 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask3" = and <4 x i32> %load_mask2, <i32 -1, i32 -1, i32 -1, i32 -1>
  store <4 x i32> %"internal_mask&function_mask3", <4 x i32>* %full_mask_memory, align 16
  %load_mask4 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  store i32 0, i32* %i, align 4
  br label %for_test

for_test:                                         ; preds = %for_loop, %all_on
  %i_load = load i32, i32* %i, align 4
  %less_i_load_ = icmp slt i32 %i_load, 12
  br i1 %less_i_load_, label %for_loop, label %for_exit

for_loop:                                         ; preds = %for_test
  %load_mask5 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask6" = and <4 x i32> %load_mask5, <i32 -1, i32 -1, i32 -1, i32 -1>
  %i_load7 = load i32, i32* %i, align 4
  %i_load7_to_int64 = sext i32 %i_load7 to i64
  %a_offset = getelementptr [12 x double], [12 x double]* %a, i32 0, i64 %i_load7_to_int64
  store double -1.000000e+00, double* %a_offset, align 8
  %i_load8 = load i32, i32* %i, align 4
  %i_load8_plus1 = add i32 %i_load8, 1
  store i32 %i_load8_plus1, i32* %i, align 4
  br label %for_test

for_exit:                                         ; preds = %for_test
  store <4 x double> <double 0.000000e+00, double 3.000000e+00, double 6.000000e+00, double 9.000000e+00>, <4 x double>* %x, align 32
  store <4 x double> <double 1.000000e+00, double 4.000000e+00, double 7.000000e+00, double 1.000000e+01>, <4 x double>* %y, align 32
  store <4 x double> <double 2.000000e+00, double 5.000000e+00, double 8.000000e+00, double 1.100000e+01>, <4 x double>* %z, align 32
  %x_load = load <4 x double>, <4 x double>* %x, align 32
  %y_load = load <4 x double>, <4 x double>* %y, align 32
  %z_load = load <4 x double>, <4 x double>* %z, align 32
  %a_offset9 = getelementptr [12 x double], [12 x double]* %a, i32 0, i64 0
  %load_mask10 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask11" = and <4 x i32> %load_mask10, <i32 -1, i32 -1, i32 -1, i32 -1>
  call void @soa_to_aos3___vydvydvydun_3C_und_3E_(<4 x double> %x_load, <4 x double> %y_load, <4 x double> %z_load, double* %a_offset9, <4 x i32> %"internal_mask&function_mask11")
  %RET_load = load float*, float** %RET1, align 8
  %mul__ = mul <4 x i32> <i32 4, i32 4, i32 4, i32 4>, <i32 0, i32 1, i32 2, i32 3>
  %mul___to_64 = sext <4 x i32> %mul__ to <4 x i64>
  %RET_load_ptr2int = ptrtoint float* %RET_load to i64
  %RET_load_ptr2int_broadcast = insertelement <4 x i64> undef, i64 %RET_load_ptr2int, i32 0
  %RET_load_ptr2int_broadcast12 = shufflevector <4 x i64> %RET_load_ptr2int_broadcast, <4 x i64> undef, <4 x i32> zeroinitializer
  %RET_load_offset = add <4 x i64> %RET_load_ptr2int_broadcast12, %mul___to_64
  %a_offset13 = getelementptr [12 x double], [12 x double]* %a, i32 0, i64 1
  %load_mask14 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %a_offset13_load = load double, double* %a_offset13, align 8
  %a_offset13_load_to_float = fptrunc double %a_offset13_load to float
  %a_offset13_load_to_float_broadcast = insertelement <4 x float> undef, float %a_offset13_load_to_float, i32 0
  %a_offset13_load_to_float_broadcast15 = shufflevector <4 x float> %a_offset13_load_to_float_broadcast, <4 x float> undef, <4 x i32> zeroinitializer
  %load_mask16 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask17" = and <4 x i32> %load_mask16, <i32 -1, i32 -1, i32 -1, i32 -1>
  %load_mask18 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask19" = and <4 x i32> %load_mask18, <i32 -1, i32 -1, i32 -1, i32 -1>
  %new_offsets = add <4 x i64> zeroinitializer, %mul___to_64
  %RET_load_ptr2int_2void = inttoptr i64 %RET_load_ptr2int to i8*, !filename !3, !first_line !4, !first_column !5, !last_line !4, !last_column !6
  %mul__54.elt0 = mul i32 4, 0
  %mul__54 = mul <4 x i32> <i32 4, i32 4, i32 4, i32 4>, <i32 0, i32 1, i32 2, i32 3>
  %mul__54_cast.elt0 = sext i32 %mul__54.elt0 to i64
  %mul__54_cast = sext <4 x i32> %mul__54 to <4 x i64>
  %_trunc = trunc <4 x i64> zeroinitializer to <4 x i32>
  %mul__54_cast_trunc.elt0 = trunc i64 %mul__54_cast.elt0 to i32
  %mul__54_cast_trunc = trunc <4 x i64> %mul__54_cast to <4 x i32>
  %_cast.elt0 = trunc i64 0 to i32
  %_cast = trunc <4 x i64> zeroinitializer to <4 x i32>
  %_cast55.elt0 = mul i32 %_cast.elt0, 1
  %_cast55 = mul <4 x i32> %_cast, <i32 1, i32 1, i32 1, i32 1>
  %mul__54_cast_trunc56.elt0 = add i32 %_cast55.elt0, %mul__54_cast_trunc.elt0
  %mul__54_cast_trunc56 = add <4 x i32> %_cast55, %mul__54_cast_trunc
  %scaled_varying.elt0 = mul i32 1, 0
  %scaled_varying = mul <4 x i32> <i32 1, i32 1, i32 1, i32 1>, zeroinitializer
  %"varying+const_offsets.elt0" = add i32 %scaled_varying.elt0, %mul__54_cast_trunc56.elt0
  %"varying+const_offsets" = add <4 x i32> %scaled_varying, %mul__54_cast_trunc56
  %ptr = getelementptr i8, i8* %RET_load_ptr2int_2void, i32 %"varying+const_offsets.elt0"
  %ptrcast = bitcast i8* %ptr to <4 x float>*
  %ptr.i67 = bitcast <4 x float>* %ptrcast to <4 x i32>*
  %val.i68 = bitcast <4 x float> %a_offset13_load_to_float_broadcast15 to <4 x i32>
  %floatmask.i.i.i69 = bitcast <4 x i32> %"internal_mask&function_mask19" to <4 x float>
  %v.i.i.i70 = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %floatmask.i.i.i69) #0
  %v64.i.i.i71 = zext i32 %v.i.i.i70 to i64
  br label %pl_loop.i.i76

pl_loop.i.i76:                                    ; preds = %pl_loopend.i.i83, %for_exit
  %pl_lane.i.i72 = phi i32 [ 0, %for_exit ], [ %pl_nextlane.i.i80, %pl_loopend.i.i83 ]
  %pl_lanemask.i.i73 = phi i64 [ 1, %for_exit ], [ %pl_nextlanemask.i.i81, %pl_loopend.i.i83 ]
  %pl_and.i.i74 = and i64 %v64.i.i.i71, %pl_lanemask.i.i73
  %pl_doit.i.i75 = icmp eq i64 %pl_and.i.i74, %pl_lanemask.i.i73
  br i1 %pl_doit.i.i75, label %pl_dolane.i.i79, label %pl_loopend.i.i83

pl_dolane.i.i79:                                  ; preds = %pl_loop.i.i76
  %ptr__id.i.i77 = getelementptr <4 x i32>, <4 x i32>* %ptr.i67, i32 0, i32 %pl_lane.i.i72
  %storeval__id.i.i78 = extractelement <4 x i32> %val.i68, i32 %pl_lane.i.i72
  store i32 %storeval__id.i.i78, i32* %ptr__id.i.i77, align 4
  br label %pl_loopend.i.i83

pl_loopend.i.i83:                                 ; preds = %pl_dolane.i.i79, %pl_loop.i.i76
  %pl_nextlane.i.i80 = add nuw nsw i32 %pl_lane.i.i72, 1
  %pl_nextlanemask.i.i81 = mul i64 %pl_lanemask.i.i73, 2
  %pl_test.i.i82 = icmp ne i32 %pl_nextlane.i.i80, 4
  br i1 %pl_test.i.i82, label %pl_loop.i.i76, label %__masked_store_float.exit84

__masked_store_float.exit84:                      ; preds = %pl_loopend.i.i, %pl_loopend.i.i83
  ret void

some_on:                                          ; preds = %allocas
  %load_mask20 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask21" = and <4 x i32> %load_mask20, %__mask
  store <4 x i32> %"internal_mask&function_mask21", <4 x i32>* %full_mask_memory, align 16
  %load_mask28 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  store i32 0, i32* %i29, align 4
  br label %for_test24

for_test24:                                       ; preds = %for_loop25, %some_on
  %i_load30 = load i32, i32* %i29, align 4
  %less_i_load30_ = icmp slt i32 %i_load30, 12
  br i1 %less_i_load30_, label %for_loop25, label %for_exit27

for_loop25:                                       ; preds = %for_test24
  %load_mask31 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask32" = and <4 x i32> %load_mask31, %__mask
  %i_load33 = load i32, i32* %i29, align 4
  %i_load33_to_int64 = sext i32 %i_load33 to i64
  %a22_offset = getelementptr [12 x double], [12 x double]* %a22, i32 0, i64 %i_load33_to_int64
  store double -1.000000e+00, double* %a22_offset, align 8
  %i_load34 = load i32, i32* %i29, align 4
  %i_load34_plus1 = add i32 %i_load34, 1
  store i32 %i_load34_plus1, i32* %i29, align 4
  br label %for_test24

for_exit27:                                       ; preds = %for_test24
  store <4 x double> <double 0.000000e+00, double 3.000000e+00, double 6.000000e+00, double 9.000000e+00>, <4 x double>* %x35, align 32
  store <4 x double> <double 1.000000e+00, double 4.000000e+00, double 7.000000e+00, double 1.000000e+01>, <4 x double>* %y36, align 32
  store <4 x double> <double 2.000000e+00, double 5.000000e+00, double 8.000000e+00, double 1.100000e+01>, <4 x double>* %z37, align 32
  %x_load38 = load <4 x double>, <4 x double>* %x35, align 32
  %y_load39 = load <4 x double>, <4 x double>* %y36, align 32
  %z_load40 = load <4 x double>, <4 x double>* %z37, align 32
  %a22_offset41 = getelementptr [12 x double], [12 x double]* %a22, i32 0, i64 0
  %load_mask42 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask43" = and <4 x i32> %load_mask42, %__mask
  call void @soa_to_aos3___vydvydvydun_3C_und_3E_(<4 x double> %x_load38, <4 x double> %y_load39, <4 x double> %z_load40, double* %a22_offset41, <4 x i32> %"internal_mask&function_mask43")
  %RET_load44 = load float*, float** %RET1, align 8
  %mul__45 = mul <4 x i32> <i32 4, i32 4, i32 4, i32 4>, <i32 0, i32 1, i32 2, i32 3>
  %mul__45_to_64 = sext <4 x i32> %mul__45 to <4 x i64>
  %RET_load44_ptr2int = ptrtoint float* %RET_load44 to i64
  %RET_load44_ptr2int_broadcast = insertelement <4 x i64> undef, i64 %RET_load44_ptr2int, i32 0
  %RET_load44_ptr2int_broadcast46 = shufflevector <4 x i64> %RET_load44_ptr2int_broadcast, <4 x i64> undef, <4 x i32> zeroinitializer
  %RET_load44_offset = add <4 x i64> %RET_load44_ptr2int_broadcast46, %mul__45_to_64
  %a22_offset47 = getelementptr [12 x double], [12 x double]* %a22, i32 0, i64 1
  %load_mask48 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %a22_offset47_load = load double, double* %a22_offset47, align 8
  %a22_offset47_load_to_float = fptrunc double %a22_offset47_load to float
  %a22_offset47_load_to_float_broadcast = insertelement <4 x float> undef, float %a22_offset47_load_to_float, i32 0
  %a22_offset47_load_to_float_broadcast49 = shufflevector <4 x float> %a22_offset47_load_to_float_broadcast, <4 x float> undef, <4 x i32> zeroinitializer
  %load_mask50 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask51" = and <4 x i32> %load_mask50, %__mask
  %load_mask52 = load <4 x i32>, <4 x i32>* %internal_mask_memory, align 16
  %"internal_mask&function_mask53" = and <4 x i32> %load_mask52, %__mask
  %new_offsets57 = add <4 x i64> zeroinitializer, %mul__45_to_64
  %RET_load44_ptr2int_2void = inttoptr i64 %RET_load44_ptr2int to i8*, !filename !3, !first_line !4, !first_column !5, !last_line !4, !last_column !6
  %mul__58.elt0 = mul i32 4, 0
  %mul__58 = mul <4 x i32> <i32 4, i32 4, i32 4, i32 4>, <i32 0, i32 1, i32 2, i32 3>
  %mul__58_cast.elt0 = sext i32 %mul__58.elt0 to i64
  %mul__58_cast = sext <4 x i32> %mul__58 to <4 x i64>
  %_trunc59 = trunc <4 x i64> zeroinitializer to <4 x i32>
  %mul__58_cast_trunc.elt0 = trunc i64 %mul__58_cast.elt0 to i32
  %mul__58_cast_trunc = trunc <4 x i64> %mul__58_cast to <4 x i32>
  %_cast60.elt0 = trunc i64 0 to i32
  %_cast60 = trunc <4 x i64> zeroinitializer to <4 x i32>
  %_cast6061.elt0 = mul i32 %_cast60.elt0, 1
  %_cast6061 = mul <4 x i32> %_cast60, <i32 1, i32 1, i32 1, i32 1>
  %mul__58_cast_trunc62.elt0 = add i32 %_cast6061.elt0, %mul__58_cast_trunc.elt0
  %mul__58_cast_trunc62 = add <4 x i32> %_cast6061, %mul__58_cast_trunc
  %scaled_varying63.elt0 = mul i32 1, 0
  %scaled_varying63 = mul <4 x i32> <i32 1, i32 1, i32 1, i32 1>, zeroinitializer
  %"varying+const_offsets64.elt0" = add i32 %scaled_varying63.elt0, %mul__58_cast_trunc62.elt0
  %"varying+const_offsets64" = add <4 x i32> %scaled_varying63, %mul__58_cast_trunc62
  %ptr65 = getelementptr i8, i8* %RET_load44_ptr2int_2void, i32 %"varying+const_offsets64.elt0"
  %ptrcast66 = bitcast i8* %ptr65 to <4 x float>*
  %ptr.i = bitcast <4 x float>* %ptrcast66 to <4 x i32>*
  %val.i = bitcast <4 x float> %a22_offset47_load_to_float_broadcast49 to <4 x i32>
  %floatmask.i.i.i = bitcast <4 x i32> %"internal_mask&function_mask53" to <4 x float>
  %v.i.i.i = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %floatmask.i.i.i) #0
  %v64.i.i.i = zext i32 %v.i.i.i to i64
  br label %pl_loop.i.i

pl_loop.i.i:                                      ; preds = %pl_loopend.i.i, %for_exit27
  %pl_lane.i.i = phi i32 [ 0, %for_exit27 ], [ %pl_nextlane.i.i, %pl_loopend.i.i ]
  %pl_lanemask.i.i = phi i64 [ 1, %for_exit27 ], [ %pl_nextlanemask.i.i, %pl_loopend.i.i ]
  %pl_and.i.i = and i64 %v64.i.i.i, %pl_lanemask.i.i
  %pl_doit.i.i = icmp eq i64 %pl_and.i.i, %pl_lanemask.i.i
  br i1 %pl_doit.i.i, label %pl_dolane.i.i, label %pl_loopend.i.i

pl_dolane.i.i:                                    ; preds = %pl_loop.i.i
  %ptr__id.i.i = getelementptr <4 x i32>, <4 x i32>* %ptr.i, i32 0, i32 %pl_lane.i.i
  %storeval__id.i.i = extractelement <4 x i32> %val.i, i32 %pl_lane.i.i
  store i32 %storeval__id.i.i, i32* %ptr__id.i.i, align 4
  br label %pl_loopend.i.i

pl_loopend.i.i:                                   ; preds = %pl_dolane.i.i, %pl_loop.i.i
  %pl_nextlane.i.i = add nuw nsw i32 %pl_lane.i.i, 1
  %pl_nextlanemask.i.i = mul i64 %pl_lanemask.i.i, 2
  %pl_test.i.i = icmp ne i32 %pl_nextlane.i.i, 4
  br i1 %pl_test.i.i, label %pl_loop.i.i, label %__masked_store_float.exit84
}

attributes #0 = { nounwind readnone }
attributes #1 = { noinline nounwind }
attributes #2 = { nounwind }

!llvm.ident = !{!0}
!llvm.module.flags = !{!1, !2}

!0 = !{!"clang version 13.0.0 (https://github.com/llvm/llvm-project.git f3449ed6073cac58efd9b62d0eb285affa650238)"}
!1 = !{i32 1, !"wchar_size", i32 4}
!2 = !{i32 7, !"PIC Level", i32 2}
!3 = !{!"aossoa-14.ispc"}
!4 = !{i32 19}
!5 = !{i32 5}
!6 = !{i32 22}
!7 = !{i32 23}