[clang] [llvm] [AArch64] Stack probing for dynamic allocas in SelectionDAG (PR #66525)
Eli Friedman via cfe-commits
cfe-commits at lists.llvm.org
Tue Nov 21 15:07:03 PST 2023
================
@@ -0,0 +1,363 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple aarch64-none-eabi < %s -verify-machineinstrs | FileCheck %s
+
+; Dynamically-sized allocation, needs a loop which can handle any size at
+; runtime. The final iteration of the loop will temporarily put SP below the
+; target address, but this doesn't break any of the ABI constraints on the
+; stack, and also doesn't probe below the target SP value.
+define void @dynamic(i64 %size, ptr %out) #0 {
+; CHECK-LABEL: dynamic:
+; CHECK: // %bb.0:
+; CHECK-NEXT: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 16
+; CHECK-NEXT: mov x29, sp
+; CHECK-NEXT: .cfi_def_cfa w29, 16
+; CHECK-NEXT: .cfi_offset w30, -8
+; CHECK-NEXT: .cfi_offset w29, -16
+; CHECK-NEXT: add x9, x0, #15
+; CHECK-NEXT: mov x8, sp
+; CHECK-NEXT: and x9, x9, #0xfffffffffffffff0
+; CHECK-NEXT: sub x8, x8, x9
+; CHECK-NEXT: .LBB0_1: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: sub sp, sp, #1, lsl #12 // =4096
+; CHECK-NEXT: cmp sp, x8
+; CHECK-NEXT: b.le .LBB0_3
+; CHECK-NEXT: // %bb.2: // in Loop: Header=BB0_1 Depth=1
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: b .LBB0_1
+; CHECK-NEXT: .LBB0_3:
+; CHECK-NEXT: mov sp, x8
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: str x8, [x1]
+; CHECK-NEXT: mov sp, x29
+; CHECK-NEXT: .cfi_def_cfa wsp, 16
+; CHECK-NEXT: ldp x29, x30, [sp], #16 // 16-byte Folded Reload
+; CHECK-NEXT: .cfi_def_cfa_offset 0
+; CHECK-NEXT: .cfi_restore w30
+; CHECK-NEXT: .cfi_restore w29
+; CHECK-NEXT: ret
+ %v = alloca i8, i64 %size, align 1
+ store ptr %v, ptr %out, align 8
+ ret void
+}
+
+; This function has a fixed-size stack slot and a dynamic one. The fixed size
+; slot isn't large enough that we would normally probe it, but we need to do so
+; here otherwise the gap between the CSR save and the first probe of the
+; dynamic allocation could be too far apart when the size of the dynamic
+; allocation is close to the guard size.
+define void @dynamic_fixed(i64 %size, ptr %out1, ptr %out2) #0 {
+; CHECK-LABEL: dynamic_fixed:
+; CHECK: // %bb.0:
+; CHECK-NEXT: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 16
+; CHECK-NEXT: mov x29, sp
+; CHECK-NEXT: .cfi_def_cfa w29, 16
+; CHECK-NEXT: .cfi_offset w30, -8
+; CHECK-NEXT: .cfi_offset w29, -16
+; CHECK-NEXT: str xzr, [sp, #-64]!
+; CHECK-NEXT: add x9, x0, #15
+; CHECK-NEXT: mov x8, sp
+; CHECK-NEXT: sub x10, x29, #64
+; CHECK-NEXT: and x9, x9, #0xfffffffffffffff0
+; CHECK-NEXT: str x10, [x1]
+; CHECK-NEXT: sub x8, x8, x9
+; CHECK-NEXT: .LBB1_1: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: sub sp, sp, #1, lsl #12 // =4096
+; CHECK-NEXT: cmp sp, x8
+; CHECK-NEXT: b.le .LBB1_3
+; CHECK-NEXT: // %bb.2: // in Loop: Header=BB1_1 Depth=1
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: b .LBB1_1
+; CHECK-NEXT: .LBB1_3:
+; CHECK-NEXT: mov sp, x8
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: str x8, [x2]
+; CHECK-NEXT: mov sp, x29
+; CHECK-NEXT: .cfi_def_cfa wsp, 16
+; CHECK-NEXT: ldp x29, x30, [sp], #16 // 16-byte Folded Reload
+; CHECK-NEXT: .cfi_def_cfa_offset 0
+; CHECK-NEXT: .cfi_restore w30
+; CHECK-NEXT: .cfi_restore w29
+; CHECK-NEXT: ret
+ %v1 = alloca i8, i64 64, align 1
+ store ptr %v1, ptr %out1, align 8
+ %v2 = alloca i8, i64 %size, align 1
+ store ptr %v2, ptr %out2, align 8
+ ret void
+}
+
+; Dynamic allocation, with an alignment requirement greater than the alignment
+; of SP. Done by ANDing the target SP with a constant to align it down, then
+; doing the loop as normal. Note that we also re-align the stack in the prolog,
+; which isn't actually needed because the only aligned allocations are dynamic,
+; this is done even without stack probing.
+define void @dynamic_align_64(i64 %size, ptr %out) #0 {
+; CHECK-LABEL: dynamic_align_64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 32
+; CHECK-NEXT: str x19, [sp, #16] // 8-byte Folded Spill
+; CHECK-NEXT: mov x29, sp
+; CHECK-NEXT: .cfi_def_cfa w29, 32
+; CHECK-NEXT: .cfi_offset w19, -16
+; CHECK-NEXT: .cfi_offset w30, -24
+; CHECK-NEXT: .cfi_offset w29, -32
+; CHECK-NEXT: sub x9, sp, #32
+; CHECK-NEXT: and sp, x9, #0xffffffffffffffc0
+; CHECK-NEXT: add x9, x0, #15
+; CHECK-NEXT: mov x8, sp
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: and x9, x9, #0xfffffffffffffff0
+; CHECK-NEXT: mov x19, sp
+; CHECK-NEXT: sub x8, x8, x9
+; CHECK-NEXT: and x8, x8, #0xffffffffffffffc0
+; CHECK-NEXT: .LBB2_1: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: sub sp, sp, #1, lsl #12 // =4096
+; CHECK-NEXT: cmp sp, x8
+; CHECK-NEXT: b.le .LBB2_3
+; CHECK-NEXT: // %bb.2: // in Loop: Header=BB2_1 Depth=1
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: b .LBB2_1
+; CHECK-NEXT: .LBB2_3:
+; CHECK-NEXT: mov sp, x8
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: str x8, [x1]
+; CHECK-NEXT: mov sp, x29
+; CHECK-NEXT: .cfi_def_cfa wsp, 32
+; CHECK-NEXT: ldr x19, [sp, #16] // 8-byte Folded Reload
+; CHECK-NEXT: ldp x29, x30, [sp], #32 // 16-byte Folded Reload
+; CHECK-NEXT: .cfi_def_cfa_offset 0
+; CHECK-NEXT: .cfi_restore w19
+; CHECK-NEXT: .cfi_restore w30
+; CHECK-NEXT: .cfi_restore w29
+; CHECK-NEXT: ret
+ %v = alloca i8, i64 %size, align 64
+ store ptr %v, ptr %out, align 8
+ ret void
+}
+
+; Dynamic allocation, with an alignment greater than the stack guard size. The
+; only difference to the dynamic allocation is the constant used for aligning
+; the target SP, the loop will probe the whole allocation without needing to
+; know about the alignment padding.
+define void @dynamic_align_8192(i64 %size, ptr %out) #0 {
+; CHECK-LABEL: dynamic_align_8192:
+; CHECK: // %bb.0:
+; CHECK-NEXT: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 32
+; CHECK-NEXT: str x19, [sp, #16] // 8-byte Folded Spill
+; CHECK-NEXT: mov x29, sp
+; CHECK-NEXT: .cfi_def_cfa w29, 32
+; CHECK-NEXT: .cfi_offset w19, -16
+; CHECK-NEXT: .cfi_offset w30, -24
+; CHECK-NEXT: .cfi_offset w29, -32
+; CHECK-NEXT: sub x9, sp, #1, lsl #12 // =4096
+; CHECK-NEXT: sub x9, x9, #4064
+; CHECK-NEXT: and x9, x9, #0xffffffffffffe000
+; CHECK-NEXT: .LBB3_1: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: sub sp, sp, #1, lsl #12 // =4096
+; CHECK-NEXT: cmp sp, x9
+; CHECK-NEXT: b.le .LBB3_3
+; CHECK-NEXT: // %bb.2: // in Loop: Header=BB3_1 Depth=1
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: b .LBB3_1
+; CHECK-NEXT: .LBB3_3:
+; CHECK-NEXT: mov sp, x9
+; CHECK-NEXT: add x9, x0, #15
+; CHECK-NEXT: mov x8, sp
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: and x9, x9, #0xfffffffffffffff0
+; CHECK-NEXT: mov x19, sp
+; CHECK-NEXT: sub x8, x8, x9
+; CHECK-NEXT: and x8, x8, #0xffffffffffffe000
+; CHECK-NEXT: .LBB3_4: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: sub sp, sp, #1, lsl #12 // =4096
+; CHECK-NEXT: cmp sp, x8
+; CHECK-NEXT: b.le .LBB3_6
+; CHECK-NEXT: // %bb.5: // in Loop: Header=BB3_4 Depth=1
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: b .LBB3_4
+; CHECK-NEXT: .LBB3_6:
+; CHECK-NEXT: mov sp, x8
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: str x8, [x1]
+; CHECK-NEXT: mov sp, x29
+; CHECK-NEXT: .cfi_def_cfa wsp, 32
+; CHECK-NEXT: ldr x19, [sp, #16] // 8-byte Folded Reload
+; CHECK-NEXT: ldp x29, x30, [sp], #32 // 16-byte Folded Reload
+; CHECK-NEXT: .cfi_def_cfa_offset 0
+; CHECK-NEXT: .cfi_restore w19
+; CHECK-NEXT: .cfi_restore w30
+; CHECK-NEXT: .cfi_restore w29
+; CHECK-NEXT: ret
+ %v = alloca i8, i64 %size, align 8192
+ store ptr %v, ptr %out, align 8
+ ret void
+}
+
+; For 64k guard pages, the only difference is the constant subtracted from SP
+; in the loop.
+define void @dynamic_64k_guard(i64 %size, ptr %out) #0 "stack-probe-size"="65536" {
+; CHECK-LABEL: dynamic_64k_guard:
+; CHECK: // %bb.0:
+; CHECK-NEXT: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 16
+; CHECK-NEXT: mov x29, sp
+; CHECK-NEXT: .cfi_def_cfa w29, 16
+; CHECK-NEXT: .cfi_offset w30, -8
+; CHECK-NEXT: .cfi_offset w29, -16
+; CHECK-NEXT: add x9, x0, #15
+; CHECK-NEXT: mov x8, sp
+; CHECK-NEXT: and x9, x9, #0xfffffffffffffff0
+; CHECK-NEXT: sub x8, x8, x9
+; CHECK-NEXT: .LBB4_1: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: sub sp, sp, #16, lsl #12 // =65536
+; CHECK-NEXT: cmp sp, x8
+; CHECK-NEXT: b.le .LBB4_3
+; CHECK-NEXT: // %bb.2: // in Loop: Header=BB4_1 Depth=1
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: b .LBB4_1
+; CHECK-NEXT: .LBB4_3:
+; CHECK-NEXT: mov sp, x8
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: str x8, [x1]
+; CHECK-NEXT: mov sp, x29
+; CHECK-NEXT: .cfi_def_cfa wsp, 16
+; CHECK-NEXT: ldp x29, x30, [sp], #16 // 16-byte Folded Reload
+; CHECK-NEXT: .cfi_def_cfa_offset 0
+; CHECK-NEXT: .cfi_restore w30
+; CHECK-NEXT: .cfi_restore w29
+; CHECK-NEXT: ret
+ %v = alloca i8, i64 %size, align 1
+ store ptr %v, ptr %out, align 8
+ ret void
+}
+
+; If a function has variable-sized stack objects, then any function calls which
+; need to pass arguments on the stack must allocate the stack space for them
+; dynamically, to ensure they are at the bottom of the frame. We need to probe
+; that space when it is larger than the unprobed space allowed by the ABI (1024
+; bytes), so this needs a very large number of arguments.
+define void @no_reserved_call_frame(i64 %n) #0 {
+; CHECK-LABEL: no_reserved_call_frame:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 16
+; CHECK-NEXT: mov x29, sp
+; CHECK-NEXT: .cfi_def_cfa w29, 16
+; CHECK-NEXT: .cfi_offset w30, -8
+; CHECK-NEXT: .cfi_offset w29, -16
+; CHECK-NEXT: lsl x9, x0, #2
+; CHECK-NEXT: mov x8, sp
+; CHECK-NEXT: add x9, x9, #15
+; CHECK-NEXT: and x9, x9, #0xfffffffffffffff0
+; CHECK-NEXT: sub x0, x8, x9
+; CHECK-NEXT: .LBB5_1: // %entry
+; CHECK-NEXT: // =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: sub sp, sp, #1, lsl #12 // =4096
+; CHECK-NEXT: cmp sp, x0
+; CHECK-NEXT: b.le .LBB5_3
+; CHECK-NEXT: // %bb.2: // %entry
+; CHECK-NEXT: // in Loop: Header=BB5_1 Depth=1
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: b .LBB5_1
+; CHECK-NEXT: .LBB5_3: // %entry
+; CHECK-NEXT: mov sp, x0
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: sub sp, sp, #1104
+; CHECK-NEXT: str xzr, [sp]
+; CHECK-NEXT: bl callee_stack_args
+; CHECK-NEXT: add sp, sp, #1104
+; CHECK-NEXT: mov sp, x29
+; CHECK-NEXT: .cfi_def_cfa wsp, 16
+; CHECK-NEXT: ldp x29, x30, [sp], #16 // 16-byte Folded Reload
+; CHECK-NEXT: .cfi_def_cfa_offset 0
+; CHECK-NEXT: .cfi_restore w30
+; CHECK-NEXT: .cfi_restore w29
+; CHECK-NEXT: ret
+entry:
+ %v = alloca i32, i64 %n
+ call void @callee_stack_args(ptr %v, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef) #2
----------------
efriedma-quic wrote:
Instead of listing out all these arguments, you can use something like `call void @callee_stack_args([100 x i64] undef)`, which should produce similar code.
https://github.com/llvm/llvm-project/pull/66525
More information about the cfe-commits
mailing list