Test source: git
Comments: LLVM PR52971
Source: <stdin>
----------------------------------------
define i32 @test1(<4 x i32> %x, <4 x i32> %y) {
%entry:
%a = alloca i64 32, align 16
%a.x = gep inbounds * %a, 32 x i64 0, 16 x i64 0
store <4 x i32> %x, * %a.x, align 16
%a.y = gep inbounds * %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, * %a.y, align 16
%a.tmp1 = gep inbounds * %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%tmp1 = load i32, * %a.tmp1, align 4
%a.tmp2 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, * %a.tmp2, align 4
%a.tmp3 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, * %a.tmp3, align 4
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
=>
define i32 @test1(<4 x i32> %x, <4 x i32> %y) {
%entry:
%a.sroa.0.8.vec.extract = extractelement <4 x i32> %x, i32 2
%a.sroa.2.28.vec.extract = extractelement <4 x i32> %y, i32 3
%a.sroa.2.16.vec.extract = extractelement <4 x i32> %y, i32 0
%tmp4 = add i32 %a.sroa.0.8.vec.extract, %a.sroa.2.28.vec.extract
%tmp5 = add i32 %a.sroa.2.16.vec.extract, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct!
----------------------------------------
define i32 @test2(<4 x i32> %x, <4 x i32> %y) {
%entry:
%a = alloca i64 32, align 16
%a.x = gep inbounds * %a, 32 x i64 0, 16 x i64 0
store <4 x i32> %x, * %a.x, align 16
%a.y = gep inbounds * %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, * %a.y, align 16
%a.tmp1 = gep inbounds * %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%tmp1 = load i32, * %a.tmp1, align 4
%a.tmp2 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, * %a.tmp2, align 4
%a.tmp3 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%a.tmp3.cast = bitcast * %a.tmp3 to *
%tmp3.vec = load <2 x i32>, * %a.tmp3.cast, align 8
%tmp3 = extractelement <2 x i32> %tmp3.vec, i32 0
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
=>
define i32 @test2(<4 x i32> %x, <4 x i32> %y) {
%entry:
%a.sroa.0.8.vec.extract = extractelement <4 x i32> %x, i32 2
%a.sroa.2.28.vec.extract = extractelement <4 x i32> %y, i32 3
%a.sroa.2.16.vec.extract = shufflevector <4 x i32> %y, <4 x i32> undef, 0, 1
%tmp3 = extractelement <2 x i32> %a.sroa.2.16.vec.extract, i32 0
%tmp4 = add i32 %a.sroa.0.8.vec.extract, %a.sroa.2.28.vec.extract
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct!
----------------------------------------
define i32 @test3(<4 x i32> %x, <4 x i32> %y) {
%entry:
%a = alloca i64 32, align 16
%a.x = gep inbounds * %a, 32 x i64 0, 16 x i64 0
store <4 x i32> %x, * %a.x, align 16
%a.y = gep inbounds * %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, * %a.y, align 16
%a.y.cast = bitcast * %a.y to *
memset * %a.y.cast align 1, i8 0, i32 16
%a.tmp1 = gep inbounds * %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%a.tmp1.cast = bitcast * %a.tmp1 to *
memset * %a.tmp1.cast align 1, i8 255, i32 4
%tmp1 = load i32, * %a.tmp1, align 4
%a.tmp2 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, * %a.tmp2, align 4
%a.tmp3 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, * %a.tmp3, align 4
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
=>
define i32 @test3(<4 x i32> %x, <4 x i32> %y) {
%entry:
%a.sroa.0.8.vec.insert = insertelement <4 x i32> %x, i32 4294967295, i32 2
%a.sroa.0.8.vec.extract = extractelement <4 x i32> %a.sroa.0.8.vec.insert, i32 2
%a.sroa.3.28.vec.extract = extractelement <4 x i32> { 0, 0, 0, 0 }, i32 3
%a.sroa.3.16.vec.extract = extractelement <4 x i32> { 0, 0, 0, 0 }, i32 0
%tmp4 = add i32 %a.sroa.0.8.vec.extract, %a.sroa.3.28.vec.extract
%tmp5 = add i32 %a.sroa.3.16.vec.extract, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct!
----------------------------------------
define i32 @test4(<4 x i32> %x, <4 x i32> %y, * %z) {
%entry:
%a = alloca i64 32, align 16
%a.x = gep inbounds * %a, 32 x i64 0, 16 x i64 0
store <4 x i32> %x, * %a.x, align 16
%a.y = gep inbounds * %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, * %a.y, align 16
%a.y.cast = bitcast * %a.y to *
%z.cast = bitcast * %z to *
memcpy * %a.y.cast align 1, * %z.cast align 1, i32 16
%a.tmp1 = gep inbounds * %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%a.tmp1.cast = bitcast * %a.tmp1 to *
%z.tmp1 = gep inbounds * %z, 16 x i64 0, 4 x i64 2
%z.tmp1.cast = bitcast * %z.tmp1 to *
memcpy * %a.tmp1.cast align 1, * %z.tmp1.cast align 1, i32 4
%tmp1 = load i32, * %a.tmp1, align 4
%a.tmp2 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, * %a.tmp2, align 4
%a.tmp3 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, * %a.tmp3, align 4
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
=>
define i32 @test4(<4 x i32> %x, <4 x i32> %y, * %z) {
%entry:
%a.sroa.3.16.copyload = load <4 x i32>, * %z, align 1
%a.sroa.0.8.z.tmp1.cast.sroa_idx = gep inbounds * %z, 16 x i64 0, 4 x i64 2
%a.sroa.0.8.copyload = load i32, * %a.sroa.0.8.z.tmp1.cast.sroa_idx, align 1
%a.sroa.0.8.vec.insert = insertelement <4 x i32> %x, i32 %a.sroa.0.8.copyload, i32 2
%a.sroa.0.8.vec.extract = extractelement <4 x i32> %a.sroa.0.8.vec.insert, i32 2
%a.sroa.3.28.vec.extract = extractelement <4 x i32> %a.sroa.3.16.copyload, i32 3
%a.sroa.3.16.vec.extract = extractelement <4 x i32> %a.sroa.3.16.copyload, i32 0
%tmp4 = add i32 %a.sroa.0.8.vec.extract, %a.sroa.3.28.vec.extract
%tmp5 = add i32 %a.sroa.3.16.vec.extract, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct!
----------------------------------------
define i32 @test5(<4 x i32> %x, <4 x i32> %y, * %z) {
%entry:
%a = alloca i64 32, align 16
%a.x = gep inbounds * %a, 32 x i64 0, 16 x i64 0
store <4 x i32> %x, * %a.x, align 16
%a.y = gep inbounds * %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, * %a.y, align 16
%a.y.cast = bitcast * %a.y to *
%a.x.cast = bitcast * %a.x to *
memcpy * %a.x.cast align 1, * %a.y.cast align 1, i32 16
%a.tmp1 = gep inbounds * %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%a.tmp1.cast = bitcast * %a.tmp1 to *
%z.tmp1 = gep inbounds * %z, 16 x i64 0, 4 x i64 2
%z.tmp1.cast = bitcast * %z.tmp1 to *
memcpy * %z.tmp1.cast align 1, * %a.tmp1.cast align 1, i32 4
%tmp1 = load i32, * %a.tmp1, align 4
%a.tmp2 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, * %a.tmp2, align 4
%a.tmp3 = gep inbounds * %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, * %a.tmp3, align 4
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
=>
define i32 @test5(<4 x i32> %x, <4 x i32> %y, * %z) {
%entry:
%a.sroa.0.8.a.sroa.0.8.z.tmp1.cast.sroa_cast.sroa_idx = gep inbounds * %z, 16 x i64 0, 4 x i64 2
%a.sroa.0.8.vec.extract2 = extractelement <4 x i32> %y, i32 2
store i32 %a.sroa.0.8.vec.extract2, * %a.sroa.0.8.a.sroa.0.8.z.tmp1.cast.sroa_cast.sroa_idx, align 1
%a.sroa.0.8.vec.extract = extractelement <4 x i32> %y, i32 2
%a.sroa.4.12.vec.extract = extractelement <4 x i32> %y, i32 3
%a.sroa.4.0.vec.extract = extractelement <4 x i32> %y, i32 0
%tmp4 = add i32 %a.sroa.0.8.vec.extract, %a.sroa.4.12.vec.extract
%tmp5 = add i32 %a.sroa.4.0.vec.extract, %tmp4
ret i32 %tmp5
}
Transformation doesn't verify!
ERROR: Timeout
----------------------------------------
define i64 @test6(<4 x i64> %x, <4 x i64> %y, i64 %n) {
%0:
%tmp = alloca i64 64, align 32
%p0 = gep inbounds * %tmp, 64 x i32 0, 1 x i64 0
store <4 x i64> %x, * %p0, align 32
%p1 = gep inbounds * %tmp, 64 x i32 0, 1 x i64 32
store <4 x i64> %y, * %p1, align 32
%addr = gep inbounds * %tmp, 64 x i32 0, 1 x i64 0, 8 x i64 %n
%res = load i64, * %addr, align 4
ret i64 %res
}
=>
define i64 @test6(<4 x i64> %x, <4 x i64> %y, i64 %n) {
%0:
%tmp = alloca i64 64, align 32
%p0 = gep inbounds * %tmp, 64 x i32 0, 1 x i64 0
store <4 x i64> %x, * %p0, align 32
%p1 = gep inbounds * %tmp, 64 x i32 0, 1 x i64 32
store <4 x i64> %y, * %p1, align 32
%addr = gep inbounds * %tmp, 64 x i32 0, 1 x i64 0, 8 x i64 %n
%res = load i64, * %addr, align 4
ret i64 %res
}
Transformation doesn't verify!
ERROR: Timeout
----------------------------------------
define <4 x i32> @test_subvec_store() {
%entry:
%a = alloca i64 16, align 16
%a.gep0 = gep * %a, 16 x i32 0, 4 x i32 0
%a.cast0 = bitcast * %a.gep0 to *
store <2 x i32> { 0, 0 }, * %a.cast0, align 8
%a.gep1 = gep * %a, 16 x i32 0, 4 x i32 1
%a.cast1 = bitcast * %a.gep1 to *
store <2 x i32> { 1, 1 }, * %a.cast1, align 8
%a.gep2 = gep * %a, 16 x i32 0, 4 x i32 2
%a.cast2 = bitcast * %a.gep2 to *
store <2 x i32> { 2, 2 }, * %a.cast2, align 8
%a.gep3 = gep * %a, 16 x i32 0, 4 x i32 3
store i32 3, * %a.gep3, align 4
%ret = load <4 x i32>, * %a, align 16
ret <4 x i32> %ret
}
=>
define <4 x i32> @test_subvec_store() {
%entry:
%a.0.vecblend = select <4 x i1> { 1, 1, 0, 0 }, <4 x i32> { 0, 0, undef, undef }, <4 x i32> undef
%a.4.vecblend = select <4 x i1> { 0, 1, 1, 0 }, <4 x i32> { undef, 1, 1, undef }, <4 x i32> %a.0.vecblend
%a.8.vecblend = select <4 x i1> { 0, 0, 1, 1 }, <4 x i32> { undef, undef, 2, 2 }, <4 x i32> %a.4.vecblend
%a.12.vec.insert = insertelement <4 x i32> %a.8.vecblend, i32 3, i32 3
ret <4 x i32> %a.12.vec.insert
}
Transformation seems to be correct!
----------------------------------------
define <4 x i32> @test_subvec_load() {
%entry:
%a = alloca i64 16, align 16
store <4 x i32> { 0, 1, 2, 3 }, * %a, align 16
%a.gep0 = gep * %a, 16 x i32 0, 4 x i32 0
%a.cast0 = bitcast * %a.gep0 to *
%first = load <2 x i32>, * %a.cast0, align 8
%a.gep1 = gep * %a, 16 x i32 0, 4 x i32 1
%a.cast1 = bitcast * %a.gep1 to *
%second = load <2 x i32>, * %a.cast1, align 8
%a.gep2 = gep * %a, 16 x i32 0, 4 x i32 2
%a.cast2 = bitcast * %a.gep2 to *
%third = load <2 x i32>, * %a.cast2, align 8
%tmp = shufflevector <2 x i32> %first, <2 x i32> %second, 0, 2
%ret = shufflevector <2 x i32> %tmp, <2 x i32> %third, 0, 1, 2, 3
ret <4 x i32> %ret
}
=>
define <4 x i32> @test_subvec_load() {
%entry:
%a.0.vec.extract = shufflevector <4 x i32> { 0, 1, 2, 3 }, <4 x i32> undef, 0, 1
%a.4.vec.extract = shufflevector <4 x i32> { 0, 1, 2, 3 }, <4 x i32> undef, 1, 2
%a.8.vec.extract = shufflevector <4 x i32> { 0, 1, 2, 3 }, <4 x i32> undef, 2, 3
%tmp = shufflevector <2 x i32> %a.0.vec.extract, <2 x i32> %a.4.vec.extract, 0, 2
%ret = shufflevector <2 x i32> %tmp, <2 x i32> %a.8.vec.extract, 0, 1, 2, 3
ret <4 x i32> %ret
}
Transformation seems to be correct!
----------------------------------------
define <4 x float> @test_subvec_memset() {
%entry:
%a = alloca i64 16, align 16
%a.gep0 = gep * %a, 16 x i32 0, 4 x i32 0
%a.cast0 = bitcast * %a.gep0 to *
memset * %a.cast0 align 1, i8 0, i32 8
%a.gep1 = gep * %a, 16 x i32 0, 4 x i32 1
%a.cast1 = bitcast * %a.gep1 to *
memset * %a.cast1 align 1, i8 1, i32 8
%a.gep2 = gep * %a, 16 x i32 0, 4 x i32 2
%a.cast2 = bitcast * %a.gep2 to *
memset * %a.cast2 align 1, i8 3, i32 8
%a.gep3 = gep * %a, 16 x i32 0, 4 x i32 3
%a.cast3 = bitcast * %a.gep3 to *
memset * %a.cast3 align 1, i8 7, i32 4
%ret = load <4 x float>, * %a, align 16
ret <4 x float> %ret
}
=>
define <4 x float> @test_subvec_memset() {
%entry:
%a.0.vecblend = select <4 x i1> { 1, 1, 0, 0 }, <4 x float> { 0.000000, 0.000000, undef, undef }, <4 x float> undef
%a.4.vecblend = select <4 x i1> { 0, 1, 1, 0 }, <4 x float> { undef, 0.000000, 0.000000, undef }, <4 x float> %a.0.vecblend
%a.8.vecblend = select <4 x i1> { 0, 0, 1, 1 }, <4 x float> { undef, undef, 0.000000, 0.000000 }, <4 x float> %a.4.vecblend
%a.12.vec.insert = insertelement <4 x float> %a.8.vecblend, float 0.000000, i32 3
ret <4 x float> %a.12.vec.insert
}
Transformation seems to be correct!
----------------------------------------
define <4 x float> @test_subvec_memcpy(* %x, * %y, * %z, * %f, * %out) {
%entry:
%a = alloca i64 16, align 16
%a.gep0 = gep * %a, 16 x i32 0, 4 x i32 0
%a.cast0 = bitcast * %a.gep0 to *
memcpy * %a.cast0 align 1, * %x align 1, i32 8
%a.gep1 = gep * %a, 16 x i32 0, 4 x i32 1
%a.cast1 = bitcast * %a.gep1 to *
memcpy * %a.cast1 align 1, * %y align 1, i32 8
%a.gep2 = gep * %a, 16 x i32 0, 4 x i32 2
%a.cast2 = bitcast * %a.gep2 to *
memcpy * %a.cast2 align 1, * %z align 1, i32 8
%a.gep3 = gep * %a, 16 x i32 0, 4 x i32 3
%a.cast3 = bitcast * %a.gep3 to *
memcpy * %a.cast3 align 1, * %f align 1, i32 4
memcpy * %out align 1, * %a.cast2 align 1, i32 8
%ret = load <4 x float>, * %a, align 16
ret <4 x float> %ret
}
=>
define <4 x float> @test_subvec_memcpy(* %x, * %y, * %z, * %f, * %out) {
%entry:
%a.0.x.sroa_cast = bitcast * %x to *
%a.0.copyload = load <2 x float>, * %a.0.x.sroa_cast, align 1
%a.0.vec.expand = shufflevector <2 x float> %a.0.copyload, <2 x float> undef, 0, 1, 4294967295, 4294967295
%a.0.vecblend = select <4 x i1> { 1, 1, 0, 0 }, <4 x float> %a.0.vec.expand, <4 x float> undef
%a.4.y.sroa_cast = bitcast * %y to *
%a.4.copyload = load <2 x float>, * %a.4.y.sroa_cast, align 1
%a.4.vec.expand = shufflevector <2 x float> %a.4.copyload, <2 x float> undef, 4294967295, 0, 1, 4294967295
%a.4.vecblend = select <4 x i1> { 0, 1, 1, 0 }, <4 x float> %a.4.vec.expand, <4 x float> %a.0.vecblend
%a.8.z.sroa_cast = bitcast * %z to *
%a.8.copyload = load <2 x float>, * %a.8.z.sroa_cast, align 1
%a.8.vec.expand = shufflevector <2 x float> %a.8.copyload, <2 x float> undef, 4294967295, 4294967295, 0, 1
%a.8.vecblend = select <4 x i1> { 0, 0, 1, 1 }, <4 x float> %a.8.vec.expand, <4 x float> %a.4.vecblend
%a.12.f.sroa_cast = bitcast * %f to *
%a.12.copyload = load float, * %a.12.f.sroa_cast, align 1
%a.12.vec.insert = insertelement <4 x float> %a.8.vecblend, float %a.12.copyload, i32 3
%a.8.out.sroa_cast = bitcast * %out to *
%a.8.vec.extract = shufflevector <4 x float> %a.12.vec.insert, <4 x float> undef, 2, 3
store <2 x float> %a.8.vec.extract, * %a.8.out.sroa_cast, align 1
ret <4 x float> %a.12.vec.insert
}
Transformation doesn't verify!
ERROR: Mismatch in memory
Example:
* %x = pointer(non-local, block_id=4, offset=0)
* %y = pointer(non-local, block_id=4, offset=37)
* %z = pointer(non-local, block_id=4, offset=0)
* %f = pointer(non-local, block_id=2, offset=0)
* %out = pointer(non-local, block_id=2, offset=0)
Source:
* %a = pointer(local, block_id=8, offset=0)
* %a.gep0 = pointer(local, block_id=8, offset=0)
* %a.cast0 = pointer(local, block_id=8, offset=0)
* %a.gep1 = pointer(local, block_id=8, offset=4)
* %a.cast1 = pointer(local, block_id=8, offset=4)
* %a.gep2 = pointer(local, block_id=8, offset=8)
* %a.cast2 = pointer(local, block_id=8, offset=8)
* %a.gep3 = pointer(local, block_id=8, offset=12)
* %a.cast3 = pointer(local, block_id=8, offset=12)
<4 x float> %ret = < poison, #x00000000 (+0.0), poison, poison >
SOURCE MEMORY STATE
===================
NON-LOCAL BLOCKS:
Block 0 > size: 0 align: 64 alloc type: 0
Block 1 > align: 2 alloc type: 0
Block 2 > size: 128 align: 2 alloc type: 0
Block 3 > align: 2 alloc type: 0
Block 4 > size: 64 align: 16 alloc type: 0
Block 5 > align: 4 alloc type: 0
LOCAL BLOCKS:
Block 8 > size: 16 align: 16 alloc type: 0
Target:
* %a.0.x.sroa_cast = pointer(non-local, block_id=4, offset=0)
<2 x float> %a.0.copyload = < poison, #x00000000 (+0.0) >
<4 x float> %a.0.vec.expand = < poison, #x00000000 (+0.0), undef, undef >
<4 x float> %a.0.vecblend = < poison, #x00000000 (+0.0), #x00000000 (+0.0) [based on undef value], #x00000000 (+0.0) >
* %a.4.y.sroa_cast = pointer(non-local, block_id=4, offset=37)
<2 x float> %a.4.copyload = < #x00000000 (+0.0), #x00000000 (+0.0) >
<4 x float> %a.4.vec.expand = < undef, #x00000000 (+0.0), #x00000000 (+0.0), undef >
<4 x float> %a.4.vecblend = < poison, #x00000000 (+0.0), #x00000000 (+0.0), #x00000000 (+0.0) >
* %a.8.z.sroa_cast = pointer(non-local, block_id=4, offset=0)
<2 x float> %a.8.copyload = < poison, #x00000000 (+0.0) >
<4 x float> %a.8.vec.expand = < undef, undef, poison, #x00000000 (+0.0) >
<4 x float> %a.8.vecblend = < poison, #x00000000 (+0.0), poison, #x00000000 (+0.0) >
* %a.12.f.sroa_cast = pointer(non-local, block_id=2, offset=0)
float %a.12.copyload = poison
<4 x float> %a.12.vec.insert = < poison, #x00000000 (+0.0), poison, poison >
* %a.8.out.sroa_cast = pointer(non-local, block_id=2, offset=0)
<2 x float> %a.8.vec.extract = < poison, poison >
TARGET MEMORY STATE
===================
LOCAL BLOCKS:
Block 8 > size: 0 align: 64 alloc type: 0
Mismatch in pointer(non-local, block_id=2, offset=0)
Source value: #x02
Target value: poison
------------------- SMT STATS -------------------
Num queries: 32
Num invalid: 0
Num skips: 0
Num trivial: 34 (51.5%)
Num timeout: 3 (9.4%)
Num errors: 0 (0.0%)
Num SAT: 11 (34.4%)
Num UNSAT: 18 (56.2%)
+ : 'RUN: at line 1' + /home/nlopes/alive2/scripts/opt-alive.sh -sroa -S + /home/nlopes/llvm/build/bin/FileCheck /home/nlopes/llvm/llvm/test/Transforms/SROA/vector-promotion.ll Alive2: Transform doesn't verify; aborting! FileCheck error: '<stdin>' is empty. FileCheck command line: /home/nlopes/llvm/build/bin/FileCheck /home/nlopes/llvm/llvm/test/Transforms/SROA/vector-promotion.ll