T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

aarch64: Implement I128 Loads and Stores

Browse Source
This commit is contained in:
Afonso Bordado
2021-06-14 08:48:34 +01:00
parent 3d56728b86
commit 1c05e06bd5
4 changed files with 450 additions and 93 deletions
Download Patch File Download Diff File Expand all files Collapse all files

164
cranelift/filetests/filetests/isa/aarch64/stack.clif
View File

@@ -276,3 +276,167 @@ block0(v0: b1):
return v0, v137
}
function %i128_stack_store(i128) {
ss0 = explicit_slot 16
block0(v0: i128):
stack_store.i128 v0, ss0
return
}
; TODO: Codegen improvement opportunities: This should be just a stp x0, x1, [sp, #-16]
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: sub sp, sp, #16
; nextln: mov x2, sp
; nextln: stp x0, x1, [x2]
; nextln: add sp, sp, #16
; nextln: ldp fp, lr, [sp], #16
function %i128_stack_store_slot_offset(i128) {
ss0 = explicit_slot 16, offset 16
block0(v0: i128):
stack_store.i128 v0, ss0
return
}
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: sub sp, sp, #16
; nextln: mov x2, sp
; nextln: stp x0, x1, [x2]
; nextln: add sp, sp, #16
; nextln: ldp fp, lr, [sp], #16
function %i128_stack_store_inst_offset(i128) {
ss0 = explicit_slot 16
ss1 = explicit_slot 16
block0(v0: i128):
stack_store.i128 v0, ss1+16
return
}
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: sub sp, sp, #32
; nextln: add x2, sp, #32
; nextln: stp x0, x1, [x2]
; nextln: add sp, sp, #32
; nextln: ldp fp, lr, [sp], #16
; nextln: ret
function %i128_stack_store_big(i128) {
ss0 = explicit_slot 100000
ss1 = explicit_slot 8
block0(v0: i128):
stack_store.i128 v0, ss0
return
}
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: movz w16, #34480
; nextln: movk w16, #1, LSL #16
; nextln: sub sp, sp, x16, UXTX
; nextln: mov x2, sp
; nextln: stp x0, x1, [x2]
; nextln: movz w16, #34480
; nextln: movk w16, #1, LSL #16
; nextln: add sp, sp, x16, UXTX
; nextln: ldp fp, lr, [sp], #16
; nextln: ret
function %i128_stack_load() -> i128 {
ss0 = explicit_slot 16
block0:
v0 = stack_load.i128 ss0
return v0
}
; TODO: Codegen improvement opportunities: This should be just a ldp x0, x1, [sp, #-16]
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: sub sp, sp, #16
; nextln: mov x0, sp
; nextln: ldp x1, x0, [x0]
; nextln: mov x2, x0
; nextln: mov x0, x1
; nextln: mov x1, x2
; nextln: add sp, sp, #16
; nextln: ldp fp, lr, [sp], #16
; nextln: ret
function %i128_stack_load_slot_offset() -> i128 {
ss0 = explicit_slot 16, offset 16
block0:
v0 = stack_load.i128 ss0
return v0
}
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: sub sp, sp, #16
; nextln: mov x0, sp
; nextln: ldp x1, x0, [x0]
; nextln: mov x2, x0
; nextln: mov x0, x1
; nextln: mov x1, x2
; nextln: add sp, sp, #16
; nextln: ldp fp, lr, [sp], #16
; nextln: ret
function %i128_stack_load_inst_offset() -> i128 {
ss0 = explicit_slot 16
ss1 = explicit_slot 16
block0:
v0 = stack_load.i128 ss1+16
return v0
}
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: sub sp, sp, #32
; nextln: add x0, sp, #32
; nextln: ldp x1, x0, [x0]
; nextln: mov x2, x0
; nextln: mov x0, x1
; nextln: mov x1, x2
; nextln: add sp, sp, #32
; nextln: ldp fp, lr, [sp], #16
; nextln: ret
function %i128_stack_load_big() -> i128 {
ss0 = explicit_slot 100000
ss1 = explicit_slot 8
block0:
v0 = stack_load.i128 ss0
return v0
}
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: movz w16, #34480
; nextln: movk w16, #1, LSL #16
; nextln: sub sp, sp, x16, UXTX
; nextln: mov x0, sp
; nextln: ldp x1, x0, [x0]
; nextln: mov x2, x0
; nextln: mov x0, x1
; nextln: mov x1, x2
; nextln: movz w16, #34480
; nextln: movk w16, #1, LSL #16
; nextln: add sp, sp, x16, UXTX
; nextln: ldp fp, lr, [sp], #16
; nextln: ret

91
cranelift/filetests/filetests/runtests/i128-load-store.clif Normal file
View File

@@ -0,0 +1,91 @@
test run
target x86_64 machinst
target aarch64
function %i128_stack_store_load(i64, i64) -> b1 {
ss0 = explicit_slot 16
block0(v0: i64,v1: i64):
v2 = iconcat v0, v1
stack_store.i128 v2, ss0
v3 = stack_load.i128 ss0
v4 = icmp.i128 eq v2, v3
return v4
}
; run: %i128_stack_store_load(0, 0) == true
; run: %i128_stack_store_load(-1, -1) == true
; run: %i128_stack_store_load(-1, 0) == true
; run: %i128_stack_store_load(0, -1) == true
; run: %i128_stack_store_load(0x01234567_89ABCDEF, 0xFEDCBA98_76543210) == true
; run: %i128_stack_store_load(0x06060606_06060606, 0xA00A00A0_0A00A00A) == true
; run: %i128_stack_store_load(0xC0FFEEEE_DECAFFFF, 0xDECAFFFF_C0FFEEEE) == true
function %i128_stack_store_load_offset(i64, i64) -> b1 {
ss0 = explicit_slot 16, offset 16
block0(v0: i64,v1: i64):
v2 = iconcat v0, v1
stack_store.i128 v2, ss0
v3 = stack_load.i128 ss0
v4 = icmp.i128 eq v2, v3
return v4
}
; run: %i128_stack_store_load_offset(0, 0) == true
; run: %i128_stack_store_load_offset(-1, -1) == true
; run: %i128_stack_store_load_offset(-1, 0) == true
; run: %i128_stack_store_load_offset(0, -1) == true
; run: %i128_stack_store_load_offset(0x01234567_89ABCDEF, 0xFEDCBA98_76543210) == true
; run: %i128_stack_store_load_offset(0x06060606_06060606, 0xA00A00A0_0A00A00A) == true
; run: %i128_stack_store_load_offset(0xC0FFEEEE_DECAFFFF, 0xDECAFFFF_C0FFEEEE) == true
function %i128_stack_store_load_inst_offset(i64, i64) -> b1 {
ss0 = explicit_slot 16
ss1 = explicit_slot 16
ss2 = explicit_slot 16
block0(v0: i64,v1: i64):
v2 = iconcat v0, v1
stack_store.i128 v2, ss1+16
v3 = stack_load.i128 ss1+16
v4 = icmp.i128 eq v2, v3
return v4
}
; run: %i128_stack_store_load_inst_offset(0, 0) == true
; run: %i128_stack_store_load_inst_offset(-1, -1) == true
; run: %i128_stack_store_load_inst_offset(-1, 0) == true
; run: %i128_stack_store_load_inst_offset(0, -1) == true
; run: %i128_stack_store_load_inst_offset(0x01234567_89ABCDEF, 0xFEDCBA98_76543210) == true
; run: %i128_stack_store_load_inst_offset(0x06060606_06060606, 0xA00A00A0_0A00A00A) == true
; run: %i128_stack_store_load_inst_offset(0xC0FFEEEE_DECAFFFF, 0xDECAFFFF_C0FFEEEE) == true
; Some arches (aarch64) try to encode the offset into the load/store instructions
; test that we spill if the offset is too large and doesn't fit in the instruction
function %i128_stack_store_load_big_offset(i64, i64) -> b1 {
ss0 = explicit_slot 100000
ss1 = explicit_slot 8
block0(v0: i64,v1: i64):
v2 = iconcat v0, v1
stack_store.i128 v2, ss0
v3 = stack_load.i128 ss0
v4 = icmp.i128 eq v2, v3
return v4
}
; run: %i128_stack_store_load_big_offset(0, 0) == true
; run: %i128_stack_store_load_big_offset(-1, -1) == true
; run: %i128_stack_store_load_big_offset(-1, 0) == true
; run: %i128_stack_store_load_big_offset(0, -1) == true
; run: %i128_stack_store_load_big_offset(0x01234567_89ABCDEF, 0xFEDCBA98_76543210) == true
; run: %i128_stack_store_load_big_offset(0x06060606_06060606, 0xA00A00A0_0A00A00A) == true
; run: %i128_stack_store_load_big_offset(0xC0FFEEEE_DECAFFFF, 0xDECAFFFF_C0FFEEEE) == true