Test source: git
Comments: LLVM PR52971
Source: <stdin>
-- 1. ModuleToFunctionPassAdaptor
-- 1. PassManager<Function> : Skipping NOP
-- 2. SROAPass
----------------------------------------
define i32 @test1(<4 x i32> %x, <4 x i32> %y) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%tmp1 = load i32, ptr %a.tmp1, align 4
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, ptr %a.tmp3, align 4
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct! (syntactically equal)
-- 3. SROAPass
----------------------------------------
define i32 @test1(<4 x i32> %x, <4 x i32> %y) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%tmp1 = load i32, ptr %a.tmp1, align 8
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, ptr %a.tmp3, align 16
%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 doesn't verify! (not unsound)
ERROR: Timeout
-- 4. PassManager<Function> : Skipping NOP
-- 5. PassManager<Function> : Skipping NOP
-- 6. SROAPass
----------------------------------------
define i32 @test2(<4 x i32> %x, <4 x i32> %y) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%tmp1 = load i32, ptr %a.tmp1, align 4
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3.vec = load <2 x i32>, ptr %a.tmp3, align 8
%tmp3 = extractelement <2 x i32> %tmp3.vec, i32 0
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct! (syntactically equal)
-- 7. SROAPass
----------------------------------------
define i32 @test2(<4 x i32> %x, <4 x i32> %y) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%tmp1 = load i32, ptr %a.tmp1, align 8
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3.vec = load <2 x i32>, ptr %a.tmp3, align 16
%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> poison, 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 doesn't verify! (not unsound)
ERROR: Timeout
-- 8. PassManager<Function> : Skipping NOP
-- 9. PassManager<Function> : Skipping NOP
-- 10. SROAPass
----------------------------------------
define i32 @test3(<4 x i32> %x, <4 x i32> %y) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
memset ptr %a.y align 1, i8 0, i32 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
memset ptr %a.tmp1 align 1, i8 255, i32 4
%tmp1 = load i32, ptr %a.tmp1, align 4
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, ptr %a.tmp3, align 4
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct! (syntactically equal)
-- 11. SROAPass
----------------------------------------
define i32 @test3(<4 x i32> %x, <4 x i32> %y) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
memset ptr %a.y align 16, i8 0, i32 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
memset ptr %a.tmp1 align 8, i8 255, i32 4
%tmp1 = load i32, ptr %a.tmp1, align 8
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, ptr %a.tmp3, align 16
%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!
-- 12. PassManager<Function> : Skipping NOP
-- 13. PassManager<Function> : Skipping NOP
-- 14. SROAPass
----------------------------------------
define i32 @test4(<4 x i32> %x, <4 x i32> %y, ptr %z) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
memcpy ptr %a.y align 1, ptr %z align 1, i32 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%z.tmp1 = gep inbounds ptr %z, 16 x i64 0, 4 x i64 2
memcpy ptr %a.tmp1 align 1, ptr %z.tmp1 align 1, i32 4
%tmp1 = load i32, ptr %a.tmp1, align 4
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, ptr %a.tmp3, align 4
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct! (syntactically equal)
-- 15. SROAPass
----------------------------------------
define i32 @test4(<4 x i32> %x, <4 x i32> %y, ptr %z) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
memcpy ptr %a.y align 16, ptr %z align 1, i32 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%z.tmp1 = gep inbounds ptr %z, 16 x i64 0, 4 x i64 2
memcpy ptr %a.tmp1 align 8, ptr %z.tmp1 align 1, i32 4
%tmp1 = load i32, ptr %a.tmp1, align 8
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, ptr %a.tmp3, align 16
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
=>
define i32 @test4(<4 x i32> %x, <4 x i32> %y, ptr %z) {
entry:
%a.sroa.3.16.copyload = load <4 x i32>, ptr %z, align 1
%z.tmp1 = gep inbounds ptr %z, 16 x i64 0, 4 x i64 2
%a.sroa.0.8.copyload = load i32, ptr %z.tmp1, 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!
-- 16. PassManager<Function> : Skipping NOP
-- 17. PassManager<Function> : Skipping NOP
-- 18. SROAPass
-- 19. SROAPass
-- 20. PassManager<Function> : Skipping NOP
-- 21. PassManager<Function> : Skipping NOP
-- 22. SROAPass
----------------------------------------
define i32 @test5(<4 x i32> %x, <4 x i32> %y, ptr %z) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
memcpy ptr %a align 1, ptr %a.y align 1, i32 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%z.tmp1 = gep inbounds ptr %z, 16 x i64 0, 4 x i64 2
memcpy ptr %z.tmp1 align 1, ptr %a.tmp1 align 1, i32 4
%tmp1 = load i32, ptr %a.tmp1, align 4
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, ptr %a.tmp3, align 4
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
Transformation seems to be correct! (syntactically equal)
-- 23. SROAPass
----------------------------------------
define i32 @test5(<4 x i32> %x, <4 x i32> %y, ptr %z) {
entry:
%a = alloca i64 32, align 16
store <4 x i32> %x, ptr %a, align 16
%a.y = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1
store <4 x i32> %y, ptr %a.y, align 16
memcpy ptr %a align 16, ptr %a.y align 16, i32 16
%a.tmp1 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 0, 4 x i64 2
%z.tmp1 = gep inbounds ptr %z, 16 x i64 0, 4 x i64 2
memcpy ptr %z.tmp1 align 1, ptr %a.tmp1 align 8, i32 4
%tmp1 = load i32, ptr %a.tmp1, align 8
%a.tmp2 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 3
%tmp2 = load i32, ptr %a.tmp2, align 4
%a.tmp3 = gep inbounds ptr %a, 32 x i64 0, 16 x i64 1, 4 x i64 0
%tmp3 = load i32, ptr %a.tmp3, align 16
%tmp4 = add i32 %tmp1, %tmp2
%tmp5 = add i32 %tmp3, %tmp4
ret i32 %tmp5
}
=>
define i32 @test5(<4 x i32> %x, <4 x i32> %y, ptr %z) {
entry:
%z.tmp1 = gep inbounds ptr %z, 16 x i64 0, 4 x i64 2
%a.sroa.0.8.vec.extract3 = extractelement <4 x i32> %y, i32 2
store i32 %a.sroa.0.8.vec.extract3, ptr %z.tmp1, 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! (not unsound)
ERROR: Timeout
-- 24. PassManager<Function> : Skipping NOP
-- 25. PassManager<Function> : Skipping NOP
-- 26. SROAPass
----------------------------------------
define i64 @test6(<4 x i64> %x, <4 x i64> %y, i64 %n) {
#0:
%tmp = alloca i64 64, align 32
%p0 = gep inbounds ptr %tmp, 64 x i32 0, 1 x i64 0
store <4 x i64> %x, ptr %p0, align 32
%p1 = gep inbounds ptr %tmp, 64 x i32 0, 1 x i64 32
store <4 x i64> %y, ptr %p1, align 32
%addr = gep inbounds ptr %tmp, 64 x i32 0, 1 x i64 0, 8 x i64 %n
%res = load i64, ptr %addr, align 4
ret i64 %res
}
Transformation seems to be correct! (syntactically equal)
-- 27. SROAPass
----------------------------------------
define i64 @test6(<4 x i64> %x, <4 x i64> %y, i64 %n) {
#0:
%tmp = alloca i64 64, align 32
%p0 = gep inbounds ptr %tmp, 64 x i32 0, 1 x i64 0
store <4 x i64> %x, ptr %p0, align 32
%p1 = gep inbounds ptr %tmp, 64 x i32 0, 1 x i64 32
store <4 x i64> %y, ptr %p1, align 32
%addr = gep inbounds ptr %tmp, 64 x i32 0, 1 x i64 0, 8 x i64 %n
%res = load i64, ptr %addr, align 4
ret i64 %res
}
Transformation seems to be correct! (syntactically equal)
-- 28. PassManager<Function> : Skipping NOP
-- 29. PassManager<Function> : Skipping NOP
-- 30. SROAPass
----------------------------------------
define <4 x i32> @test_subvec_store() {
entry:
%a = alloca i64 16, align 16
store <2 x i32> { 0, 0 }, ptr %a, align 8
%a.gep1 = gep ptr %a, 16 x i32 0, 4 x i32 1
store <2 x i32> { 1, 1 }, ptr %a.gep1, align 8
%a.gep2 = gep ptr %a, 16 x i32 0, 4 x i32 2
store <2 x i32> { 2, 2 }, ptr %a.gep2, align 8
%a.gep3 = gep ptr %a, 16 x i32 0, 4 x i32 3
store i32 3, ptr %a.gep3, align 4
%ret = load <4 x i32>, ptr %a, align 16
ret <4 x i32> %ret
}
Transformation seems to be correct! (syntactically equal)
-- 31. SROAPass
----------------------------------------
define <4 x i32> @test_subvec_store() {
entry:
%a = alloca i64 16, align 16
store <2 x i32> { 0, 0 }, ptr %a, align 16
%a.gep1 = gep ptr %a, 16 x i32 0, 4 x i32 1
store <2 x i32> { 1, 1 }, ptr %a.gep1, align 8
%a.gep2 = gep ptr %a, 16 x i32 0, 4 x i32 2
store <2 x i32> { 2, 2 }, ptr %a.gep2, align 8
%a.gep3 = gep ptr %a, 16 x i32 0, 4 x i32 3
store i32 3, ptr %a.gep3, align 4
%ret = load <4 x i32>, ptr %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!
-- 32. PassManager<Function> : Skipping NOP
-- 33. PassManager<Function> : Skipping NOP
-- 34. SROAPass
----------------------------------------
define <4 x i32> @test_subvec_load() {
entry:
%a = alloca i64 16, align 16
store <4 x i32> { 0, 1, 2, 3 }, ptr %a, align 16
%first = load <2 x i32>, ptr %a, align 8
%a.gep1 = gep ptr %a, 16 x i32 0, 4 x i32 1
%second = load <2 x i32>, ptr %a.gep1, align 8
%a.gep2 = gep ptr %a, 16 x i32 0, 4 x i32 2
%third = load <2 x i32>, ptr %a.gep2, 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
}
Transformation seems to be correct! (syntactically equal)
-- 35. SROAPass
----------------------------------------
define <4 x i32> @test_subvec_load() {
entry:
%a = alloca i64 16, align 16
store <4 x i32> { 0, 1, 2, 3 }, ptr %a, align 16
%first = load <2 x i32>, ptr %a, align 16
%a.gep1 = gep ptr %a, 16 x i32 0, 4 x i32 1
%second = load <2 x i32>, ptr %a.gep1, align 8
%a.gep2 = gep ptr %a, 16 x i32 0, 4 x i32 2
%third = load <2 x i32>, ptr %a.gep2, 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> poison, 0, 1
%a.4.vec.extract = shufflevector <4 x i32> { 0, 1, 2, 3 }, <4 x i32> poison, 1, 2
%a.8.vec.extract = shufflevector <4 x i32> { 0, 1, 2, 3 }, <4 x i32> poison, 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!
-- 36. PassManager<Function> : Skipping NOP
-- 37. PassManager<Function> : Skipping NOP
-- 38. SROAPass
----------------------------------------
define <4 x float> @test_subvec_memset() {
entry:
%a = alloca i64 16, align 16
memset ptr %a align 1, i8 0, i32 8
%a.gep1 = gep ptr %a, 16 x i32 0, 4 x i32 1
memset ptr %a.gep1 align 1, i8 1, i32 8
%a.gep2 = gep ptr %a, 16 x i32 0, 4 x i32 2
memset ptr %a.gep2 align 1, i8 3, i32 8
%a.gep3 = gep ptr %a, 16 x i32 0, 4 x i32 3
memset ptr %a.gep3 align 1, i8 7, i32 4
%ret = load <4 x float>, ptr %a, align 16
ret <4 x float> %ret
}
Transformation seems to be correct! (syntactically equal)
-- 39. SROAPass
----------------------------------------
define <4 x float> @test_subvec_memset() {
entry:
%a = alloca i64 16, align 16
memset ptr %a align 16, i8 0, i32 8
%a.gep1 = gep ptr %a, 16 x i32 0, 4 x i32 1
memset ptr %a.gep1 align 4, i8 1, i32 8
%a.gep2 = gep ptr %a, 16 x i32 0, 4 x i32 2
memset ptr %a.gep2 align 8, i8 3, i32 8
%a.gep3 = gep ptr %a, 16 x i32 0, 4 x i32 3
memset ptr %a.gep3 align 4, i8 7, i32 4
%ret = load <4 x float>, ptr %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!
-- 40. PassManager<Function> : Skipping NOP
-- 41. PassManager<Function> : Skipping NOP
-- 42. SROAPass
----------------------------------------
define <4 x float> @test_subvec_memcpy(ptr %x, ptr %y, ptr %z, ptr %f, ptr %out) {
entry:
%a = alloca i64 16, align 16
memcpy ptr %a align 1, ptr %x align 1, i32 8
%a.gep1 = gep ptr %a, 16 x i32 0, 4 x i32 1
memcpy ptr %a.gep1 align 1, ptr %y align 1, i32 8
%a.gep2 = gep ptr %a, 16 x i32 0, 4 x i32 2
memcpy ptr %a.gep2 align 1, ptr %z align 1, i32 8
%a.gep3 = gep ptr %a, 16 x i32 0, 4 x i32 3
memcpy ptr %a.gep3 align 1, ptr %f align 1, i32 4
memcpy ptr %out align 1, ptr %a.gep2 align 1, i32 8
%ret = load <4 x float>, ptr %a, align 16
ret <4 x float> %ret
}
Transformation seems to be correct! (syntactically equal)
-- 43. SROAPass
----------------------------------------
define <4 x float> @test_subvec_memcpy(ptr %x, ptr %y, ptr %z, ptr %f, ptr %out) {
entry:
%a = alloca i64 16, align 16
memcpy ptr %a align 16, ptr %x align 1, i32 8
%a.gep1 = gep ptr %a, 16 x i32 0, 4 x i32 1
memcpy ptr %a.gep1 align 4, ptr %y align 1, i32 8
%a.gep2 = gep ptr %a, 16 x i32 0, 4 x i32 2
memcpy ptr %a.gep2 align 8, ptr %z align 1, i32 8
%a.gep3 = gep ptr %a, 16 x i32 0, 4 x i32 3
memcpy ptr %a.gep3 align 4, ptr %f align 1, i32 4
memcpy ptr %out align 1, ptr %a.gep2 align 8, i32 8
%ret = load <4 x float>, ptr %a, align 16
ret <4 x float> %ret
}
=>
define <4 x float> @test_subvec_memcpy(ptr %x, ptr %y, ptr %z, ptr %f, ptr %out) {
entry:
%a.0.copyload = load <2 x float>, ptr %x, align 1
%a.0.vec.expand = shufflevector <2 x float> %a.0.copyload, <2 x float> poison, 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.copyload = load <2 x float>, ptr %y, align 1
%a.4.vec.expand = shufflevector <2 x float> %a.4.copyload, <2 x float> poison, 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.copyload = load <2 x float>, ptr %z, align 1
%a.8.vec.expand = shufflevector <2 x float> %a.8.copyload, <2 x float> poison, 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.copyload = load float, ptr %f, align 1
%a.12.vec.insert = insertelement <4 x float> %a.8.vecblend, float %a.12.copyload, i32 3
%a.8.vec.extract = shufflevector <4 x float> %a.12.vec.insert, <4 x float> poison, 2, 3
store <2 x float> %a.8.vec.extract, ptr %out, align 1
ret <4 x float> %a.12.vec.insert
}
Transformation doesn't verify! (unsound)
ERROR: Mismatch in memory
Example:
ptr %x = pointer(non-local, block_id=1, offset=0)
ptr %y = pointer(non-local, block_id=1, offset=0)
ptr %z = pointer(non-local, block_id=1, offset=4)
ptr %f = pointer(non-local, block_id=1, offset=0)
ptr %out = pointer(non-local, block_id=1, offset=6)
Source:
ptr %a = null
ptr %a.gep1 = pointer(local, block_id=8, offset=4)
ptr %a.gep2 = pointer(local, block_id=8, offset=8)
ptr %a.gep3 = pointer(local, block_id=8, offset=12)
<4 x float> %ret = < poison, poison, poison, poison >
SOURCE MEMORY STATE
===================
NON-LOCAL BLOCKS:
Block 0 > size: 0 align: 1 alloc type: 0 alive: false address: 0
Block 1 > size: 15 align: 1 alloc type: 0 alive: true address: 6
Contents:
#x06: pointer(non-local, block_id=0, offset=-128), byte offset=4
#x03: #bppp0p0p0
#x07: #bpp0ppppp
#x00: #bpp0ppppp
#x02: #bp010pppp
#x08: #b0ppppppp
#x01: #bpppp0pp0
#x04: #bpp0ppp00
#x05: #bpp0001pp
else: poison
Block 2 > size: 0 align: 4 alloc type: 0 alive: true address: 4
Contents:
#x06: #bpp0ppppp
else: poison
Block 3 > size: 0 align: 1 alloc type: 0 alive: true address: 1
Contents:
#x06: #bppp0pppp
else: poison
Block 4 > size: 0 align: 2 alloc type: 0 alive: true address: 4
Contents:
#x06: #bppp0pppp
else: poison
Block 5 > size: 0 align: 1 alloc type: 0 alive: true address: 4
Contents:
#x06: #bppp0pppp
else: poison
LOCAL BLOCKS:
Block 8 > size: 16 align: 16 alloc type: 1 alive: true
Target:
<2 x float> %a.0.copyload = < poison, poison >
<4 x float> %a.0.vec.expand = < poison, poison, poison, poison >
<4 x float> %a.0.vecblend = < poison, poison, #x00000000 (+0.0) [based on undef value], #x00000000 (+0.0) [based on undef value] >
<2 x float> %a.4.copyload = < poison, poison >
<4 x float> %a.4.vec.expand = < poison, poison, poison, poison >
<4 x float> %a.4.vecblend = < poison, poison, poison, #x00000000 (+0.0) >
<2 x float> %a.8.copyload = < poison, poison >
<4 x float> %a.8.vec.expand = < poison, poison, poison, poison >
<4 x float> %a.8.vecblend = < poison, poison, poison, poison >
float %a.12.copyload = poison
<4 x float> %a.12.vec.insert = < poison, poison, poison, poison >
<2 x float> %a.8.vec.extract = < poison, poison >
Mismatch in pointer(non-local, block_id=1, offset=8)
Source value: pointer(non-local, block_id=0, offset=-128), byte offset=4
Target value: poison
Pass: SROAPass
Command line: '/home/nlopes/llvm/build/bin/opt' '-load=/home/nlopes/alive2/build/tv/tv.so' '-load-pass-plugin=/home/nlopes/alive2/build/tv/tv.so' '-tv-exit-on-error' '-passes=sroa<preserve-cfg>' '-S' '-tv-smt-to=20000' '-tv-report-dir=/home/nlopes/alive2/build/logs' '-tv-smt-stats'
Wrote bitcode to: "/home/nlopes/alive2/build/logs/in_lSgM2qwj_cF5k.bc"
------------------- SMT STATS -------------------
Num queries: 92
Num invalid: 0
Num skips: 0
Num trivial: 47 (33.8%)
Num timeout: 3 (3.3%)
Num errors: 0 (0.0%)
Num SAT: 53 (57.6%)
Num UNSAT: 36 (39.1%)
Alive2: Transform doesn't verify; aborting!
RUN: at line 2: /home/nlopes/alive2/build/opt-alive.sh < /bitbucket/nlopes/llvm/llvm/test/Transforms/SROA/vector-promotion.ll -passes='sroa<preserve-cfg>' -S | /bitbucket/nlopes/llvm/build/bin/FileCheck /bitbucket/nlopes/llvm/llvm/test/Transforms/SROA/vector-promotion.ll --check-prefixes=CHECK,CHECK-PRESERVE-CFG + /home/nlopes/alive2/build/opt-alive.sh '-passes=sroa<preserve-cfg>' -S + /bitbucket/nlopes/llvm/build/bin/FileCheck /bitbucket/nlopes/llvm/llvm/test/Transforms/SROA/vector-promotion.ll --check-prefixes=CHECK,CHECK-PRESERVE-CFG FileCheck error: '<stdin>' is empty. FileCheck command line: /bitbucket/nlopes/llvm/build/bin/FileCheck /bitbucket/nlopes/llvm/llvm/test/Transforms/SROA/vector-promotion.ll --check-prefixes=CHECK,CHECK-PRESERVE-CFG