d9a5aa8e2d
Change all the cstval_pred_ty based PatternMatch helpers (things like m_AllOnes and m_Zero) to only allow poison elements inside vector splats, not undef elements. Historically, we used to represent non-demanded elements in vectors using undef. Nowadays, we use poison instead. As such, I believe that support for undef in vector splats is no longer useful. At the same time, while poison splat elements are pretty much always safe to ignore, this is not generally the case for undef elements. We have existing miscompiles in our tests due to this (see the masked-merge-*.ll tests changed here) and it's easy to miss such cases in the future, now that we write tests using poison instead of undef elements. I think overall, keeping support for undef elements no longer makes sense, and we should drop it. Once this is done consistently, I think we may also consider allowing poison in m_APInt by default, as doing that change is much less risky than doing the same with undef. This change involves a substantial amount of test changes. For most tests, I've just replaced undef with poison, as I don't think there is value in retaining both. For some tests (where the distinction between undef and poison is important), I've duplicated tests.
556 lines
21 KiB
LLVM
556 lines
21 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt < %s -passes=instcombine -S | FileCheck %s
|
|
|
|
; Can't get smaller than this.
|
|
|
|
define <2 x i1> @trunc(<2 x i64> %a) {
|
|
; CHECK-LABEL: @trunc(
|
|
; CHECK-NEXT: [[T:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i1>
|
|
; CHECK-NEXT: ret <2 x i1> [[T]]
|
|
;
|
|
%t = trunc <2 x i64> %a to <2 x i1>
|
|
ret <2 x i1> %t
|
|
}
|
|
|
|
; This is trunc.
|
|
|
|
define <2 x i1> @and_cmp_is_trunc(<2 x i64> %a) {
|
|
; CHECK-LABEL: @and_cmp_is_trunc(
|
|
; CHECK-NEXT: [[R:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i1>
|
|
; CHECK-NEXT: ret <2 x i1> [[R]]
|
|
;
|
|
%t = and <2 x i64> %a, <i64 1, i64 1>
|
|
%r = icmp ne <2 x i64> %t, zeroinitializer
|
|
ret <2 x i1> %r
|
|
}
|
|
|
|
; This is trunc.
|
|
|
|
define <2 x i1> @and_cmp_is_trunc_even_with_poison_elt(<2 x i64> %a) {
|
|
; CHECK-LABEL: @and_cmp_is_trunc_even_with_poison_elt(
|
|
; CHECK-NEXT: [[R:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i1>
|
|
; CHECK-NEXT: ret <2 x i1> [[R]]
|
|
;
|
|
%t = and <2 x i64> %a, <i64 poison, i64 1>
|
|
%r = icmp ne <2 x i64> %t, zeroinitializer
|
|
ret <2 x i1> %r
|
|
}
|
|
|
|
; TODO: This could be just 1 instruction (trunc), but our poison matching is incomplete.
|
|
|
|
define <2 x i1> @and_cmp_is_trunc_even_with_poison_elts(<2 x i64> %a) {
|
|
; CHECK-LABEL: @and_cmp_is_trunc_even_with_poison_elts(
|
|
; CHECK-NEXT: [[T:%.*]] = and <2 x i64> [[A:%.*]], <i64 poison, i64 1>
|
|
; CHECK-NEXT: [[R:%.*]] = icmp ne <2 x i64> [[T]], <i64 poison, i64 0>
|
|
; CHECK-NEXT: ret <2 x i1> [[R]]
|
|
;
|
|
%t = and <2 x i64> %a, <i64 poison, i64 1>
|
|
%r = icmp ne <2 x i64> %t, <i64 poison, i64 0>
|
|
ret <2 x i1> %r
|
|
}
|
|
|
|
; The ashr turns into an lshr.
|
|
define <2 x i64> @test2(<2 x i64> %a) {
|
|
; CHECK-LABEL: @test2(
|
|
; CHECK-NEXT: [[B:%.*]] = lshr <2 x i64> [[A:%.*]], <i64 1, i64 1>
|
|
; CHECK-NEXT: [[T:%.*]] = and <2 x i64> [[B]], <i64 32767, i64 32767>
|
|
; CHECK-NEXT: ret <2 x i64> [[T]]
|
|
;
|
|
%b = and <2 x i64> %a, <i64 65535, i64 65535>
|
|
%t = ashr <2 x i64> %b, <i64 1, i64 1>
|
|
ret <2 x i64> %t
|
|
}
|
|
|
|
define <2 x i64> @test3(<4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: @test3(
|
|
; CHECK-NEXT: [[AND1:%.*]] = fcmp ord <4 x float> [[A:%.*]], [[B:%.*]]
|
|
; CHECK-NEXT: [[AND:%.*]] = sext <4 x i1> [[AND1]] to <4 x i32>
|
|
; CHECK-NEXT: [[CONV:%.*]] = bitcast <4 x i32> [[AND]] to <2 x i64>
|
|
; CHECK-NEXT: ret <2 x i64> [[CONV]]
|
|
;
|
|
%cmp = fcmp ord <4 x float> %a, zeroinitializer
|
|
%sext = sext <4 x i1> %cmp to <4 x i32>
|
|
%cmp4 = fcmp ord <4 x float> %b, zeroinitializer
|
|
%sext5 = sext <4 x i1> %cmp4 to <4 x i32>
|
|
%and = and <4 x i32> %sext, %sext5
|
|
%conv = bitcast <4 x i32> %and to <2 x i64>
|
|
ret <2 x i64> %conv
|
|
}
|
|
|
|
define <2 x i64> @test4(<4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: @test4(
|
|
; CHECK-NEXT: [[OR1:%.*]] = fcmp uno <4 x float> [[A:%.*]], [[B:%.*]]
|
|
; CHECK-NEXT: [[OR:%.*]] = sext <4 x i1> [[OR1]] to <4 x i32>
|
|
; CHECK-NEXT: [[CONV:%.*]] = bitcast <4 x i32> [[OR]] to <2 x i64>
|
|
; CHECK-NEXT: ret <2 x i64> [[CONV]]
|
|
;
|
|
%cmp = fcmp uno <4 x float> %a, zeroinitializer
|
|
%sext = sext <4 x i1> %cmp to <4 x i32>
|
|
%cmp4 = fcmp uno <4 x float> %b, zeroinitializer
|
|
%sext5 = sext <4 x i1> %cmp4 to <4 x i32>
|
|
%or = or <4 x i32> %sext, %sext5
|
|
%conv = bitcast <4 x i32> %or to <2 x i64>
|
|
ret <2 x i64> %conv
|
|
}
|
|
|
|
; rdar://7434900
|
|
define <2 x i64> @test5(<4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: @test5(
|
|
; CHECK-NEXT: [[CMP:%.*]] = fcmp ult <4 x float> [[A:%.*]], zeroinitializer
|
|
; CHECK-NEXT: [[CMP4:%.*]] = fcmp ult <4 x float> [[B:%.*]], zeroinitializer
|
|
; CHECK-NEXT: [[AND1:%.*]] = and <4 x i1> [[CMP]], [[CMP4]]
|
|
; CHECK-NEXT: [[AND:%.*]] = sext <4 x i1> [[AND1]] to <4 x i32>
|
|
; CHECK-NEXT: [[CONV:%.*]] = bitcast <4 x i32> [[AND]] to <2 x i64>
|
|
; CHECK-NEXT: ret <2 x i64> [[CONV]]
|
|
;
|
|
%cmp = fcmp ult <4 x float> %a, zeroinitializer
|
|
%sext = sext <4 x i1> %cmp to <4 x i32>
|
|
%cmp4 = fcmp ult <4 x float> %b, zeroinitializer
|
|
%sext5 = sext <4 x i1> %cmp4 to <4 x i32>
|
|
%and = and <4 x i32> %sext, %sext5
|
|
%conv = bitcast <4 x i32> %and to <2 x i64>
|
|
ret <2 x i64> %conv
|
|
}
|
|
|
|
define <2 x i64> @test6(<4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: @test6(
|
|
; CHECK-NEXT: [[CMP:%.*]] = fcmp ult <4 x float> [[A:%.*]], zeroinitializer
|
|
; CHECK-NEXT: [[CMP4:%.*]] = fcmp ult <4 x float> [[B:%.*]], zeroinitializer
|
|
; CHECK-NEXT: [[AND1:%.*]] = or <4 x i1> [[CMP]], [[CMP4]]
|
|
; CHECK-NEXT: [[AND:%.*]] = sext <4 x i1> [[AND1]] to <4 x i32>
|
|
; CHECK-NEXT: [[CONV:%.*]] = bitcast <4 x i32> [[AND]] to <2 x i64>
|
|
; CHECK-NEXT: ret <2 x i64> [[CONV]]
|
|
;
|
|
%cmp = fcmp ult <4 x float> %a, zeroinitializer
|
|
%sext = sext <4 x i1> %cmp to <4 x i32>
|
|
%cmp4 = fcmp ult <4 x float> %b, zeroinitializer
|
|
%sext5 = sext <4 x i1> %cmp4 to <4 x i32>
|
|
%and = or <4 x i32> %sext, %sext5
|
|
%conv = bitcast <4 x i32> %and to <2 x i64>
|
|
ret <2 x i64> %conv
|
|
}
|
|
|
|
define <2 x i64> @test7(<4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: @test7(
|
|
; CHECK-NEXT: [[CMP:%.*]] = fcmp ult <4 x float> [[A:%.*]], zeroinitializer
|
|
; CHECK-NEXT: [[CMP4:%.*]] = fcmp ult <4 x float> [[B:%.*]], zeroinitializer
|
|
; CHECK-NEXT: [[AND1:%.*]] = xor <4 x i1> [[CMP]], [[CMP4]]
|
|
; CHECK-NEXT: [[AND:%.*]] = sext <4 x i1> [[AND1]] to <4 x i32>
|
|
; CHECK-NEXT: [[CONV:%.*]] = bitcast <4 x i32> [[AND]] to <2 x i64>
|
|
; CHECK-NEXT: ret <2 x i64> [[CONV]]
|
|
;
|
|
%cmp = fcmp ult <4 x float> %a, zeroinitializer
|
|
%sext = sext <4 x i1> %cmp to <4 x i32>
|
|
%cmp4 = fcmp ult <4 x float> %b, zeroinitializer
|
|
%sext5 = sext <4 x i1> %cmp4 to <4 x i32>
|
|
%and = xor <4 x i32> %sext, %sext5
|
|
%conv = bitcast <4 x i32> %and to <2 x i64>
|
|
ret <2 x i64> %conv
|
|
}
|
|
|
|
define void @convert(ptr %dst.addr, <2 x i64> %src) {
|
|
; CHECK-LABEL: @convert(
|
|
; CHECK-NEXT: [[VAL:%.*]] = trunc <2 x i64> [[SRC:%.*]] to <2 x i32>
|
|
; CHECK-NEXT: [[ADD:%.*]] = add <2 x i32> [[VAL]], <i32 1, i32 1>
|
|
; CHECK-NEXT: store <2 x i32> [[ADD]], ptr [[DST_ADDR:%.*]], align 8
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
%val = trunc <2 x i64> %src to <2 x i32>
|
|
%add = add <2 x i32> %val, <i32 1, i32 1>
|
|
store <2 x i32> %add, ptr %dst.addr
|
|
ret void
|
|
}
|
|
|
|
define <2 x i65> @foo(<2 x i64> %t) {
|
|
; CHECK-LABEL: @foo(
|
|
; CHECK-NEXT: [[A_MASK:%.*]] = and <2 x i64> [[T:%.*]], <i64 4294967295, i64 4294967295>
|
|
; CHECK-NEXT: [[B:%.*]] = zext nneg <2 x i64> [[A_MASK]] to <2 x i65>
|
|
; CHECK-NEXT: ret <2 x i65> [[B]]
|
|
;
|
|
%a = trunc <2 x i64> %t to <2 x i32>
|
|
%b = zext <2 x i32> %a to <2 x i65>
|
|
ret <2 x i65> %b
|
|
}
|
|
|
|
define <2 x i64> @bar(<2 x i65> %t) {
|
|
; CHECK-LABEL: @bar(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i65> [[T:%.*]] to <2 x i64>
|
|
; CHECK-NEXT: [[B:%.*]] = and <2 x i64> [[TMP1]], <i64 4294967295, i64 4294967295>
|
|
; CHECK-NEXT: ret <2 x i64> [[B]]
|
|
;
|
|
%a = trunc <2 x i65> %t to <2 x i32>
|
|
%b = zext <2 x i32> %a to <2 x i64>
|
|
ret <2 x i64> %b
|
|
}
|
|
|
|
define <2 x i64> @bars(<2 x i65> %t) {
|
|
; CHECK-LABEL: @bars(
|
|
; CHECK-NEXT: [[A:%.*]] = trunc <2 x i65> [[T:%.*]] to <2 x i32>
|
|
; CHECK-NEXT: [[B:%.*]] = sext <2 x i32> [[A]] to <2 x i64>
|
|
; CHECK-NEXT: ret <2 x i64> [[B]]
|
|
;
|
|
%a = trunc <2 x i65> %t to <2 x i32>
|
|
%b = sext <2 x i32> %a to <2 x i64>
|
|
ret <2 x i64> %b
|
|
}
|
|
|
|
define <2 x i64> @quxs(<2 x i64> %t) {
|
|
; CHECK-LABEL: @quxs(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i64> [[T:%.*]], <i64 32, i64 32>
|
|
; CHECK-NEXT: [[B:%.*]] = ashr exact <2 x i64> [[TMP1]], <i64 32, i64 32>
|
|
; CHECK-NEXT: ret <2 x i64> [[B]]
|
|
;
|
|
%a = trunc <2 x i64> %t to <2 x i32>
|
|
%b = sext <2 x i32> %a to <2 x i64>
|
|
ret <2 x i64> %b
|
|
}
|
|
|
|
define <2 x i64> @quxt(<2 x i64> %t) {
|
|
; CHECK-LABEL: @quxt(
|
|
; CHECK-NEXT: [[A:%.*]] = shl <2 x i64> [[T:%.*]], <i64 32, i64 32>
|
|
; CHECK-NEXT: [[B:%.*]] = ashr exact <2 x i64> [[A]], <i64 32, i64 32>
|
|
; CHECK-NEXT: ret <2 x i64> [[B]]
|
|
;
|
|
%a = shl <2 x i64> %t, <i64 32, i64 32>
|
|
%b = ashr <2 x i64> %a, <i64 32, i64 32>
|
|
ret <2 x i64> %b
|
|
}
|
|
|
|
define <2 x double> @fa(<2 x double> %t) {
|
|
; CHECK-LABEL: @fa(
|
|
; CHECK-NEXT: [[A:%.*]] = fptrunc <2 x double> [[T:%.*]] to <2 x float>
|
|
; CHECK-NEXT: [[B:%.*]] = fpext <2 x float> [[A]] to <2 x double>
|
|
; CHECK-NEXT: ret <2 x double> [[B]]
|
|
;
|
|
%a = fptrunc <2 x double> %t to <2 x float>
|
|
%b = fpext <2 x float> %a to <2 x double>
|
|
ret <2 x double> %b
|
|
}
|
|
|
|
define <2 x double> @fb(<2 x double> %t) {
|
|
; CHECK-LABEL: @fb(
|
|
; CHECK-NEXT: [[A:%.*]] = fptoui <2 x double> [[T:%.*]] to <2 x i64>
|
|
; CHECK-NEXT: [[B:%.*]] = uitofp <2 x i64> [[A]] to <2 x double>
|
|
; CHECK-NEXT: ret <2 x double> [[B]]
|
|
;
|
|
%a = fptoui <2 x double> %t to <2 x i64>
|
|
%b = uitofp <2 x i64> %a to <2 x double>
|
|
ret <2 x double> %b
|
|
}
|
|
|
|
define <2 x double> @fc(<2 x double> %t) {
|
|
; CHECK-LABEL: @fc(
|
|
; CHECK-NEXT: [[A:%.*]] = fptosi <2 x double> [[T:%.*]] to <2 x i64>
|
|
; CHECK-NEXT: [[B:%.*]] = sitofp <2 x i64> [[A]] to <2 x double>
|
|
; CHECK-NEXT: ret <2 x double> [[B]]
|
|
;
|
|
%a = fptosi <2 x double> %t to <2 x i64>
|
|
%b = sitofp <2 x i64> %a to <2 x double>
|
|
ret <2 x double> %b
|
|
}
|
|
|
|
; PR9228
|
|
define <4 x float> @f(i32 %a) {
|
|
; CHECK-LABEL: @f(
|
|
; CHECK-NEXT: ret <4 x float> undef
|
|
;
|
|
%dim = insertelement <4 x i32> undef, i32 %a, i32 0
|
|
%dim30 = insertelement <4 x i32> %dim, i32 %a, i32 1
|
|
%dim31 = insertelement <4 x i32> %dim30, i32 %a, i32 2
|
|
%dim32 = insertelement <4 x i32> %dim31, i32 %a, i32 3
|
|
|
|
%offset_ptr = getelementptr <4 x float>, ptr null, i32 1
|
|
%offset_int = ptrtoint ptr %offset_ptr to i64
|
|
%sizeof32 = trunc i64 %offset_int to i32
|
|
|
|
%smearinsert33 = insertelement <4 x i32> undef, i32 %sizeof32, i32 0
|
|
%smearinsert34 = insertelement <4 x i32> %smearinsert33, i32 %sizeof32, i32 1
|
|
%smearinsert35 = insertelement <4 x i32> %smearinsert34, i32 %sizeof32, i32 2
|
|
%smearinsert36 = insertelement <4 x i32> %smearinsert35, i32 %sizeof32, i32 3
|
|
|
|
%delta_scale = mul <4 x i32> %dim32, %smearinsert36
|
|
%offset_delta = add <4 x i32> zeroinitializer, %delta_scale
|
|
|
|
%offset_varying_delta = add <4 x i32> %offset_delta, undef
|
|
|
|
ret <4 x float> undef
|
|
}
|
|
|
|
define <8 x i32> @pr24458(<8 x float> %n) {
|
|
; CHECK-LABEL: @pr24458(
|
|
; CHECK-NEXT: ret <8 x i32> <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
|
|
;
|
|
%notequal_b_load_.i = fcmp une <8 x float> %n, zeroinitializer
|
|
%equal_a_load72_.i = fcmp ueq <8 x float> %n, zeroinitializer
|
|
%notequal_b_load__to_boolvec.i = sext <8 x i1> %notequal_b_load_.i to <8 x i32>
|
|
%equal_a_load72__to_boolvec.i = sext <8 x i1> %equal_a_load72_.i to <8 x i32>
|
|
%wrong = or <8 x i32> %notequal_b_load__to_boolvec.i, %equal_a_load72__to_boolvec.i
|
|
ret <8 x i32> %wrong
|
|
}
|
|
|
|
; Hoist a trunc to a scalar if we're inserting into an undef vector.
|
|
; trunc (inselt undef, X, Index) --> inselt undef, (trunc X), Index
|
|
|
|
define <3 x i16> @trunc_inselt_undef(i32 %x) {
|
|
; CHECK-LABEL: @trunc_inselt_undef(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 [[X:%.*]] to i16
|
|
; CHECK-NEXT: [[TRUNC:%.*]] = insertelement <3 x i16> <i16 undef, i16 poison, i16 undef>, i16 [[TMP1]], i64 1
|
|
; CHECK-NEXT: ret <3 x i16> [[TRUNC]]
|
|
;
|
|
%vec = insertelement <3 x i32> undef, i32 %x, i32 1
|
|
%trunc = trunc <3 x i32> %vec to <3 x i16>
|
|
ret <3 x i16> %trunc
|
|
}
|
|
|
|
; Hoist a trunc to a scalar if we're inserting into an undef vector.
|
|
; trunc (inselt undef, X, Index) --> inselt undef, (trunc X), Index
|
|
|
|
define <2 x float> @fptrunc_inselt_undef(double %x, i32 %index) {
|
|
; CHECK-LABEL: @fptrunc_inselt_undef(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = fptrunc double [[X:%.*]] to float
|
|
; CHECK-NEXT: [[TRUNC:%.*]] = insertelement <2 x float> undef, float [[TMP1]], i32 [[INDEX:%.*]]
|
|
; CHECK-NEXT: ret <2 x float> [[TRUNC]]
|
|
;
|
|
%vec = insertelement <2 x double> <double undef, double undef>, double %x, i32 %index
|
|
%trunc = fptrunc <2 x double> %vec to <2 x float>
|
|
ret <2 x float> %trunc
|
|
}
|
|
|
|
; TODO: Strengthen the backend, so we can have this canonicalization.
|
|
; Insert a scalar int into a constant vector and truncate:
|
|
; trunc (inselt C, X, Index) --> inselt C, (trunc X), Index
|
|
|
|
define <3 x i16> @trunc_inselt1(i32 %x) {
|
|
; CHECK-LABEL: @trunc_inselt1(
|
|
; CHECK-NEXT: [[VEC:%.*]] = insertelement <3 x i32> <i32 3, i32 poison, i32 65536>, i32 [[X:%.*]], i64 1
|
|
; CHECK-NEXT: [[TRUNC:%.*]] = trunc <3 x i32> [[VEC]] to <3 x i16>
|
|
; CHECK-NEXT: ret <3 x i16> [[TRUNC]]
|
|
;
|
|
%vec = insertelement <3 x i32> <i32 3, i32 -2, i32 65536>, i32 %x, i32 1
|
|
%trunc = trunc <3 x i32> %vec to <3 x i16>
|
|
ret <3 x i16> %trunc
|
|
}
|
|
|
|
; TODO: Strengthen the backend, so we can have this canonicalization.
|
|
; Insert a scalar FP into a constant vector and FP truncate:
|
|
; fptrunc (inselt C, X, Index) --> inselt C, (fptrunc X), Index
|
|
|
|
define <2 x float> @fptrunc_inselt1(double %x, i32 %index) {
|
|
; CHECK-LABEL: @fptrunc_inselt1(
|
|
; CHECK-NEXT: [[VEC:%.*]] = insertelement <2 x double> <double undef, double 3.000000e+00>, double [[X:%.*]], i32 [[INDEX:%.*]]
|
|
; CHECK-NEXT: [[TRUNC:%.*]] = fptrunc <2 x double> [[VEC]] to <2 x float>
|
|
; CHECK-NEXT: ret <2 x float> [[TRUNC]]
|
|
;
|
|
%vec = insertelement <2 x double> <double undef, double 3.0>, double %x, i32 %index
|
|
%trunc = fptrunc <2 x double> %vec to <2 x float>
|
|
ret <2 x float> %trunc
|
|
}
|
|
|
|
; TODO: Strengthen the backend, so we can have this canonicalization.
|
|
; Insert a scalar int constant into a vector and truncate:
|
|
; trunc (inselt X, C, Index) --> inselt (trunc X), C', Index
|
|
|
|
define <8 x i16> @trunc_inselt2(<8 x i32> %x, i32 %index) {
|
|
; CHECK-LABEL: @trunc_inselt2(
|
|
; CHECK-NEXT: [[VEC:%.*]] = insertelement <8 x i32> [[X:%.*]], i32 1048576, i32 [[INDEX:%.*]]
|
|
; CHECK-NEXT: [[TRUNC:%.*]] = trunc <8 x i32> [[VEC]] to <8 x i16>
|
|
; CHECK-NEXT: ret <8 x i16> [[TRUNC]]
|
|
;
|
|
%vec = insertelement <8 x i32> %x, i32 1048576, i32 %index
|
|
%trunc = trunc <8 x i32> %vec to <8 x i16>
|
|
ret <8 x i16> %trunc
|
|
}
|
|
|
|
; TODO: Strengthen the backend, so we can have this canonicalization.
|
|
; Insert a scalar FP constant into a vector and FP truncate:
|
|
; fptrunc (inselt X, C, Index) --> inselt (fptrunc X), C', Index
|
|
|
|
define <3 x float> @fptrunc_inselt2(<3 x double> %x) {
|
|
; CHECK-LABEL: @fptrunc_inselt2(
|
|
; CHECK-NEXT: [[VEC:%.*]] = insertelement <3 x double> [[X:%.*]], double 4.000000e+00, i64 2
|
|
; CHECK-NEXT: [[TRUNC:%.*]] = fptrunc <3 x double> [[VEC]] to <3 x float>
|
|
; CHECK-NEXT: ret <3 x float> [[TRUNC]]
|
|
;
|
|
%vec = insertelement <3 x double> %x, double 4.0, i32 2
|
|
%trunc = fptrunc <3 x double> %vec to <3 x float>
|
|
ret <3 x float> %trunc
|
|
}
|
|
|
|
; Converting to a wide type might reduce instruction count,
|
|
; but we can not do that unless the backend can recover from
|
|
; the creation of a potentially illegal op (like a 64-bit vmul).
|
|
; PR40032 - https://bugs.llvm.org/show_bug.cgi?id=40032
|
|
|
|
define <2 x i64> @sext_less_casting_with_wideop(<2 x i64> %x, <2 x i64> %y) {
|
|
; CHECK-LABEL: @sext_less_casting_with_wideop(
|
|
; CHECK-NEXT: [[XNARROW:%.*]] = trunc <2 x i64> [[X:%.*]] to <2 x i32>
|
|
; CHECK-NEXT: [[YNARROW:%.*]] = trunc <2 x i64> [[Y:%.*]] to <2 x i32>
|
|
; CHECK-NEXT: [[MUL:%.*]] = mul <2 x i32> [[XNARROW]], [[YNARROW]]
|
|
; CHECK-NEXT: [[R:%.*]] = sext <2 x i32> [[MUL]] to <2 x i64>
|
|
; CHECK-NEXT: ret <2 x i64> [[R]]
|
|
;
|
|
%xnarrow = trunc <2 x i64> %x to <2 x i32>
|
|
%ynarrow = trunc <2 x i64> %y to <2 x i32>
|
|
%mul = mul <2 x i32> %xnarrow, %ynarrow
|
|
%r = sext <2 x i32> %mul to <2 x i64>
|
|
ret <2 x i64> %r
|
|
}
|
|
|
|
define <2 x i64> @zext_less_casting_with_wideop(<2 x i64> %x, <2 x i64> %y) {
|
|
; CHECK-LABEL: @zext_less_casting_with_wideop(
|
|
; CHECK-NEXT: [[XNARROW:%.*]] = trunc <2 x i64> [[X:%.*]] to <2 x i32>
|
|
; CHECK-NEXT: [[YNARROW:%.*]] = trunc <2 x i64> [[Y:%.*]] to <2 x i32>
|
|
; CHECK-NEXT: [[MUL:%.*]] = mul <2 x i32> [[XNARROW]], [[YNARROW]]
|
|
; CHECK-NEXT: [[R:%.*]] = zext <2 x i32> [[MUL]] to <2 x i64>
|
|
; CHECK-NEXT: ret <2 x i64> [[R]]
|
|
;
|
|
%xnarrow = trunc <2 x i64> %x to <2 x i32>
|
|
%ynarrow = trunc <2 x i64> %y to <2 x i32>
|
|
%mul = mul <2 x i32> %xnarrow, %ynarrow
|
|
%r = zext <2 x i32> %mul to <2 x i64>
|
|
ret <2 x i64> %r
|
|
}
|
|
|
|
define <4 x float> @sitofp_shuf(<4 x i32> %x) {
|
|
; CHECK-LABEL: @sitofp_shuf(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = sitofp <4 x i32> [[X:%.*]] to <4 x float>
|
|
; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 poison>
|
|
; CHECK-NEXT: ret <4 x float> [[R]]
|
|
;
|
|
%s = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 undef>
|
|
%r = sitofp <4 x i32> %s to <4 x float>
|
|
ret <4 x float> %r
|
|
}
|
|
|
|
define <3 x half> @uitofp_shuf(<3 x i16> %x) {
|
|
; CHECK-LABEL: @uitofp_shuf(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = uitofp <3 x i16> [[X:%.*]] to <3 x half>
|
|
; CHECK-NEXT: [[R:%.*]] = shufflevector <3 x half> [[TMP1]], <3 x half> poison, <3 x i32> <i32 2, i32 poison, i32 0>
|
|
; CHECK-NEXT: ret <3 x half> [[R]]
|
|
;
|
|
%s = shufflevector <3 x i16> %x, <3 x i16> poison, <3 x i32> <i32 2, i32 undef, i32 0>
|
|
%r = uitofp <3 x i16> %s to <3 x half>
|
|
ret <3 x half> %r
|
|
}
|
|
|
|
define <4 x i64> @fptosi_shuf(<4 x double> %x) {
|
|
; CHECK-LABEL: @fptosi_shuf(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = fptosi <4 x double> [[X:%.*]] to <4 x i64>
|
|
; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i64> [[TMP1]], <4 x i64> poison, <4 x i32> <i32 poison, i32 2, i32 3, i32 poison>
|
|
; CHECK-NEXT: ret <4 x i64> [[R]]
|
|
;
|
|
%s = shufflevector <4 x double> %x, <4 x double> poison, <4 x i32> <i32 undef, i32 2, i32 3, i32 undef>
|
|
%r = fptosi <4 x double> %s to <4 x i64>
|
|
ret <4 x i64> %r
|
|
}
|
|
|
|
define <2 x i32> @fptoui_shuf(<2 x float> %x) {
|
|
; CHECK-LABEL: @fptoui_shuf(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = fptoui <2 x float> [[X:%.*]] to <2 x i32>
|
|
; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> <i32 1, i32 1>
|
|
; CHECK-NEXT: ret <2 x i32> [[R]]
|
|
;
|
|
%s = shufflevector <2 x float> %x, <2 x float> poison, <2 x i32> <i32 1, i32 1>
|
|
%r = fptoui <2 x float> %s to <2 x i32>
|
|
ret <2 x i32> %r
|
|
}
|
|
|
|
; negative test
|
|
; TODO: Should we reduce the width of the shuffle?
|
|
|
|
define <4 x half> @narrowing_sitofp_shuf(<4 x i32> %x) {
|
|
; CHECK-LABEL: @narrowing_sitofp_shuf(
|
|
; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 poison>
|
|
; CHECK-NEXT: [[R:%.*]] = sitofp <4 x i32> [[S]] to <4 x half>
|
|
; CHECK-NEXT: ret <4 x half> [[R]]
|
|
;
|
|
%s = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 undef>
|
|
%r = sitofp <4 x i32> %s to <4 x half>
|
|
ret <4 x half> %r
|
|
}
|
|
|
|
; negative test
|
|
|
|
define <4 x double> @widening_uitofp_shuf(<4 x i32> %x) {
|
|
; CHECK-LABEL: @widening_uitofp_shuf(
|
|
; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 poison>
|
|
; CHECK-NEXT: [[R:%.*]] = uitofp <4 x i32> [[S]] to <4 x double>
|
|
; CHECK-NEXT: ret <4 x double> [[R]]
|
|
;
|
|
%s = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 undef>
|
|
%r = uitofp <4 x i32> %s to <4 x double>
|
|
ret <4 x double> %r
|
|
}
|
|
|
|
; negative test
|
|
|
|
define <3 x i64> @fptosi_narrowing_shuf(<4 x double> %x) {
|
|
; CHECK-LABEL: @fptosi_narrowing_shuf(
|
|
; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x double> [[X:%.*]], <4 x double> poison, <3 x i32> <i32 poison, i32 2, i32 3>
|
|
; CHECK-NEXT: [[R:%.*]] = fptosi <3 x double> [[S]] to <3 x i64>
|
|
; CHECK-NEXT: ret <3 x i64> [[R]]
|
|
;
|
|
%s = shufflevector <4 x double> %x, <4 x double> poison, <3 x i32> <i32 undef, i32 2, i32 3>
|
|
%r = fptosi <3 x double> %s to <3 x i64>
|
|
ret <3 x i64> %r
|
|
}
|
|
|
|
; negative test
|
|
; TODO: Should we reduce the width of the cast?
|
|
|
|
define <3 x i32> @fptoui_widening_shuf(<2 x float> %x) {
|
|
; CHECK-LABEL: @fptoui_widening_shuf(
|
|
; CHECK-NEXT: [[S:%.*]] = shufflevector <2 x float> [[X:%.*]], <2 x float> poison, <3 x i32> <i32 1, i32 1, i32 0>
|
|
; CHECK-NEXT: [[R:%.*]] = fptoui <3 x float> [[S]] to <3 x i32>
|
|
; CHECK-NEXT: ret <3 x i32> [[R]]
|
|
;
|
|
%s = shufflevector <2 x float> %x, <2 x float> poison, <3 x i32> <i32 1, i32 1, i32 0>
|
|
%r = fptoui <3 x float> %s to <3 x i32>
|
|
ret <3 x i32> %r
|
|
}
|
|
|
|
; negative test
|
|
; TODO: Should we reduce the width of the cast?
|
|
|
|
define <4 x half> @narrowing_sitofp_widening_shuf(<2 x i32> %x) {
|
|
; CHECK-LABEL: @narrowing_sitofp_widening_shuf(
|
|
; CHECK-NEXT: [[S:%.*]] = shufflevector <2 x i32> [[X:%.*]], <2 x i32> poison, <4 x i32> <i32 1, i32 0, i32 0, i32 poison>
|
|
; CHECK-NEXT: [[R:%.*]] = sitofp <4 x i32> [[S]] to <4 x half>
|
|
; CHECK-NEXT: ret <4 x half> [[R]]
|
|
;
|
|
%s = shufflevector <2 x i32> %x, <2 x i32> poison, <4 x i32> <i32 1, i32 0, i32 0, i32 undef>
|
|
%r = sitofp <4 x i32> %s to <4 x half>
|
|
ret <4 x half> %r
|
|
}
|
|
|
|
declare void @use(<4 x i32>)
|
|
|
|
; negative test
|
|
|
|
define <4 x float> @sitofp_shuf_extra_use(<4 x i32> %x) {
|
|
; CHECK-LABEL: @sitofp_shuf_extra_use(
|
|
; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 poison>
|
|
; CHECK-NEXT: call void @use(<4 x i32> [[S]])
|
|
; CHECK-NEXT: [[R:%.*]] = sitofp <4 x i32> [[S]] to <4 x float>
|
|
; CHECK-NEXT: ret <4 x float> [[R]]
|
|
;
|
|
%s = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 undef>
|
|
call void @use(<4 x i32> %s)
|
|
%r = sitofp <4 x i32> %s to <4 x float>
|
|
ret <4 x float> %r
|
|
}
|
|
|
|
; negative test
|
|
; TODO: Allow scalable vectors?
|
|
|
|
define <vscale x 4 x float> @sitofp_shuf_scalable(<vscale x 4 x i32> %x) {
|
|
; CHECK-LABEL: @sitofp_shuf_scalable(
|
|
; CHECK-NEXT: [[S:%.*]] = shufflevector <vscale x 4 x i32> [[X:%.*]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
|
|
; CHECK-NEXT: [[R:%.*]] = sitofp <vscale x 4 x i32> [[S]] to <vscale x 4 x float>
|
|
; CHECK-NEXT: ret <vscale x 4 x float> [[R]]
|
|
;
|
|
%s = shufflevector <vscale x 4 x i32> %x, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
|
|
%r = sitofp <vscale x 4 x i32> %s to <vscale x 4 x float>
|
|
ret <vscale x 4 x float> %r
|
|
}
|