dd40bf075a
This enables more runtests to be executed on s390x. Doing so
uncovered a two back-end bugs, which are fixed as well:
- The result of cls was always off by one.
- The result of popcnt.i16 has uninitialized high bits.
In addition, I found a bug in the load-op-store.clif test case:
v3 = heap_addr.i64 heap0, v1, 4
v4 = iconst.i64 42
store.i32 v4, v3
This was clearly intended to perform a 32-bit store, but
actually performs a 64-bit store (it seems the type annotation
of the store opcode is ignored, and the type of the operand
is used instead). That bug did not show any noticable symptoms
on little-endian architectures, but broke on big-endian.
87 lines
2.7 KiB
Plaintext
87 lines
2.7 KiB
Plaintext
test run
|
|
|
|
target x86_64
|
|
target s390x
|
|
target aarch64
|
|
|
|
function %fpromote_f32_f64(i64 vmctx, i64, f32) -> f64 {
|
|
gv0 = vmctx
|
|
gv1 = load.i64 notrap aligned gv0+0
|
|
heap0 = static gv1, min 0x10, bound 0x10, offset_guard 0x0, index_type i64
|
|
|
|
block0(v0: i64, v1: i64, v2: f32):
|
|
v3 = heap_addr.i64 heap0, v1, 4
|
|
store.f32 v2, v3
|
|
v4 = load.f32 v3
|
|
v5 = fpromote.f64 v4
|
|
return v5
|
|
}
|
|
|
|
; heap: static, size=0x10, ptr=vmctx+0, bound=vmctx+8
|
|
; run: %fpromote_f32_f64(0, 0x0.0) == 0x0.0
|
|
; run: %fpromote_f32_f64(1, 0x0.1) == 0x0.1
|
|
; run: %fpromote_f32_f64(2, 0x0.2) == 0x0.2
|
|
; run: %fpromote_f32_f64(3, 0x3.2) == 0x3.2
|
|
; run: %fpromote_f32_f64(0xc, 0x3.2) == 0x3.2
|
|
|
|
function %fdemote_test(i64 vmctx, i64, f64) -> f32 {
|
|
gv0 = vmctx
|
|
gv1 = load.i64 notrap aligned gv0+0
|
|
heap0 = static gv1, min 0x10, bound 0x10, offset_guard 0x0, index_type i64
|
|
|
|
block0(v0: i64, v1: i64, v2: f64):
|
|
v3 = heap_addr.i64 heap0, v1, 8
|
|
store.f64 v2, v3
|
|
v4 = load.f64 v3
|
|
v5 = fdemote.f32 v4
|
|
return v5
|
|
}
|
|
|
|
; heap: static, size=0x10, ptr=vmctx+0, bound=vmctx+8
|
|
; run: %fdemote_test(0, 0x0.0) == 0x0.0
|
|
; run: %fdemote_test(1, 0x0.1) == 0x0.1
|
|
; run: %fdemote_test(2, 0x0.2) == 0x0.2
|
|
; run: %fdemote_test(3, 0x3.2) == 0x3.2
|
|
; run: %fdemote_test(0x8, 0x3.2) == 0x3.2
|
|
|
|
function %fvdemote_test(i64 vmctx, i64, f64x2) -> f32x4 {
|
|
gv0 = vmctx
|
|
gv1 = load.i64 notrap aligned gv0+0
|
|
heap0 = static gv1, min 0x20, bound 0x20, offset_guard 0, index_type i64
|
|
|
|
block0(v0: i64, v1: i64, v2: f64x2):
|
|
v3 = heap_addr.i64 heap0, v1, 16
|
|
store.f64x2 v2, v3
|
|
v4 = load.f64x2 v3
|
|
v5 = fvdemote v4
|
|
return v5
|
|
}
|
|
|
|
; heap: static, size=0x20, ptr=vmctx+0, bound=vmctx+8
|
|
; run: %fvdemote_test(0, [0x0.0 0x0.0]) == [0x0.0 0x0.0 0x0.0 0x0.0]
|
|
; run: %fvdemote_test(1, [0x0.1 0x0.2]) == [0x0.1 0x0.2 0x0.0 0x0.0]
|
|
; run: %fvdemote_test(2, [0x2.1 0x1.2]) == [0x2.1 0x1.2 0x0.0 0x0.0]
|
|
; run: %fvdemote_test(8, [0x2.1 0x1.2]) == [0x2.1 0x1.2 0x0.0 0x0.0]
|
|
; run: %fvdemote_test(16, [0x2.1 0x1.2]) == [0x2.1 0x1.2 0x0.0 0x0.0]
|
|
|
|
|
|
function %fvpromote_low_test(i64 vmctx, i64, f32x4) -> f64x2 {
|
|
gv0 = vmctx
|
|
gv1 = load.i64 notrap aligned gv0+0
|
|
heap0 = static gv1, min 0x20, bound 0x20, offset_guard 0, index_type i64
|
|
|
|
block0(v0: i64, v1: i64, v2: f32x4):
|
|
v3 = heap_addr.i64 heap0, v1, 16
|
|
store.f32x4 v2, v3
|
|
v4 = load.f32x4 v3
|
|
v5 = fvpromote_low v4
|
|
return v5
|
|
}
|
|
|
|
; heap: static, size=0x20, ptr=vmctx+0, bound=vmctx+8
|
|
; run: %fvpromote_low_test(0, [0x0.0 0x0.0 0x0.0 0x0.0]) == [0x0.0 0x0.0]
|
|
; run: %fvpromote_low_test(1, [0x0.1 0x0.2 0x0.0 0x0.0]) == [0x0.1 0x0.2]
|
|
; run: %fvpromote_low_test(2, [0x2.1 0x1.2 0x0.0 0x0.0]) == [0x2.1 0x1.2]
|
|
; run: %fvpromote_low_test(5, [0x0.0 0x0.0 0x2.1 0x1.2]) == [0x0.0 0x0.0]
|
|
; run: %fvpromote_low_test(16, [0x0.0 0x0.0 0x2.1 0x1.2]) == [0x0.0 0x0.0]
|