[llvm] c9c930a - [SelectionDAG] Don't promote the alignment of allocas beyond the stack alignment.
Eli Friedman via llvm-commits
llvm-commits at lists.llvm.org
Mon May 11 17:40:02 PDT 2020
Author: Eli Friedman
Date: 2020-05-11T17:39:00-07:00
New Revision: c9c930ae67c38b93451aa979de723723aec0067d
URL: https://github.com/llvm/llvm-project/commit/c9c930ae67c38b93451aa979de723723aec0067d
DIFF: https://github.com/llvm/llvm-project/commit/c9c930ae67c38b93451aa979de723723aec0067d.diff
LOG: [SelectionDAG] Don't promote the alignment of allocas beyond the stack alignment.
allocas in LLVM IR have a specified alignment. When that alignment is
specified, the alloca has at least that alignment at runtime.
If the specified type of the alloca has a higher preferred alignment,
SelectionDAG currently ignores that specified alignment, and increases
the alignment. It does this even if it would trigger stack realignment.
I don't think this makes sense, so this patch changes that.
I was looking into this for SVE in particular: for SVE, overaligning
vscale'ed types is extra expensive because it requires realigning the
stack multiple times, or using dynamic allocation. (This currently isn't
implemented.)
I updated the expected assembly for a couple tests; in particular, for
arg-copy-elide.ll, the optimization in question does not increase the
alignment the way SelectionDAG normally would. For the rest, I just
increased the specified alignment on the allocas to match what
SelectionDAG was inferring.
Differential Revision: https://reviews.llvm.org/D79532
Added:
Modified:
llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
llvm/test/CodeGen/AArch64/sve-alloca-stackid.ll
llvm/test/CodeGen/AMDGPU/private-element-size.ll
llvm/test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll
llvm/test/CodeGen/ARM/alloc-no-stack-realign.ll
llvm/test/CodeGen/Thumb2/mve-basic.ll
llvm/test/CodeGen/X86/arg-copy-elide.ll
llvm/test/CodeGen/X86/avx2-vbroadcast.ll
llvm/test/CodeGen/X86/avx512-intel-ocl.ll
llvm/test/CodeGen/X86/load-local-v3i129.ll
llvm/test/CodeGen/X86/movtopush.ll
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp b/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
index 3f302d5fa0ca..c4821254f897 100644
--- a/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
@@ -134,8 +134,20 @@ void FunctionLoweringInfo::set(const Function &fn, MachineFunction &mf,
for (const Instruction &I : BB) {
if (const AllocaInst *AI = dyn_cast<AllocaInst>(&I)) {
Type *Ty = AI->getAllocatedType();
+ Align TyPrefAlign = MF->getDataLayout().getPrefTypeAlign(Ty);
+ // The "specified" alignment is the alignment written on the alloca,
+ // or the preferred alignment of the type if none is specified.
+ //
+ // (Unspecified alignment on allocas will be going away soon.)
+ Align SpecifiedAlign = AI->getAlign() ? *AI->getAlign() : TyPrefAlign;
+
+ // If the preferred alignment of the type is higher than the specified
+ // alignment of the alloca, promote the alignment, as long as it doesn't
+ // require realigning the stack.
+ //
+ // FIXME: Do we really want to second-guess the IR in isel?
Align Alignment =
- max(MF->getDataLayout().getPrefTypeAlign(Ty), AI->getAlign());
+ std::max(std::min(TyPrefAlign, StackAlign), SpecifiedAlign);
// Static allocas can be folded into the initial stack frame
// adjustment. For targets that don't realign the stack, don't
diff --git a/llvm/test/CodeGen/AArch64/sve-alloca-stackid.ll b/llvm/test/CodeGen/AArch64/sve-alloca-stackid.ll
index 4aae5bef22e2..915ffcf17b05 100644
--- a/llvm/test/CodeGen/AArch64/sve-alloca-stackid.ll
+++ b/llvm/test/CodeGen/AArch64/sve-alloca-stackid.ll
@@ -15,3 +15,18 @@ define i32 @foo(<vscale x 16 x i8> %val) {
}
declare i32 @bar(<vscale x 16 x i8>* %ptr);
+
+; CHECKCG-LABEL: foo2:
+; CHECKCG: addvl sp, sp, #-2
+
+; CHECKISEL-LABEL: name: foo2
+; CHECKISEL: stack:
+; CHECKISEL: id: 0, name: ptr, type: default, offset: 0, size: 32, alignment: 16,
+; CHECKISEL-NEXT: stack-id: sve-vec
+
+define i32 @foo2(<vscale x 32 x i8> %val) {
+ %ptr = alloca <vscale x 32 x i8>, align 16
+ %res = call i32 @bar2(<vscale x 32 x i8>* %ptr)
+ ret i32 %res
+}
+declare i32 @bar2(<vscale x 32 x i8>* %ptr);
diff --git a/llvm/test/CodeGen/AMDGPU/private-element-size.ll b/llvm/test/CodeGen/AMDGPU/private-element-size.ll
index d5e5ba5202fc..843f554b0513 100644
--- a/llvm/test/CodeGen/AMDGPU/private-element-size.ll
+++ b/llvm/test/CodeGen/AMDGPU/private-element-size.ll
@@ -113,7 +113,7 @@ entry:
%gep.index = getelementptr inbounds i32, i32 addrspace(1)* %index.array, i64 %idxprom
%index.load = load i32, i32 addrspace(1)* %gep.index
%index = and i32 %index.load, 2
- %alloca = alloca [2 x <8 x i32>], align 16, addrspace(5)
+ %alloca = alloca [2 x <8 x i32>], align 32, addrspace(5)
%gep0 = getelementptr inbounds [2 x <8 x i32>], [2 x <8 x i32>] addrspace(5)* %alloca, i32 0, i32 0
%gep1 = getelementptr inbounds [2 x <8 x i32>], [2 x <8 x i32>] addrspace(5)* %alloca, i32 0, i32 1
store <8 x i32> zeroinitializer, <8 x i32> addrspace(5)* %gep0
diff --git a/llvm/test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll b/llvm/test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll
index c8fb1ec72d50..32bb1ad95262 100644
--- a/llvm/test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll
+++ b/llvm/test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll
@@ -51,8 +51,8 @@ define amdgpu_kernel void @kernel_background_evaluate(float addrspace(5)* %kg, <
; GCN-NEXT: BB0_2: ; %shader_eval_surface.exit
; GCN-NEXT: s_endpgm
entry:
- %sd = alloca < 1339 x i32>, align 16, addrspace(5)
- %state = alloca <4 x i32>, align 4, addrspace(5)
+ %sd = alloca < 1339 x i32>, align 8192, addrspace(5)
+ %state = alloca <4 x i32>, align 16, addrspace(5)
%rslt = call i32 @svm_eval_nodes(float addrspace(5)* %kg, <1339 x i32> addrspace(5)* %sd, <4 x i32> addrspace(5)* %state, i32 0, i32 4194304)
%cmp = icmp eq i32 %rslt, 0
br i1 %cmp, label %shader_eval_surface.exit, label %if.then4.i
diff --git a/llvm/test/CodeGen/ARM/alloc-no-stack-realign.ll b/llvm/test/CodeGen/ARM/alloc-no-stack-realign.ll
index 1d759ba21ec7..b6f3e0747568 100644
--- a/llvm/test/CodeGen/ARM/alloc-no-stack-realign.ll
+++ b/llvm/test/CodeGen/ARM/alloc-no-stack-realign.ll
@@ -34,7 +34,7 @@ entry:
; CHECK: vst1.64 {{{d[0-9]+}}, {{d[0-9]+}}}, [r[[R1]]:128]
; CHECK: vst1.32 {{{d[0-9]+}}, {{d[0-9]+}}}, [r0:128]!
; CHECK: vst1.64 {{{d[0-9]+}}, {{d[0-9]+}}}, [r0:128]
- %retval = alloca <16 x float>, align 16
+ %retval = alloca <16 x float>, align 64
%0 = load <16 x float>, <16 x float>* @T3_retval, align 16
store <16 x float> %0, <16 x float>* %retval
%1 = load <16 x float>, <16 x float>* %retval
@@ -73,7 +73,7 @@ entry:
; CHECK: vst1.64 {{{d[0-9]+}}, {{d[0-9]+}}}, [r0:128]
-%retval = alloca <16 x float>, align 16
+%retval = alloca <16 x float>, align 64
%0 = load <16 x float>, <16 x float>* @T3_retval, align 16
store <16 x float> %0, <16 x float>* %retval
%1 = load <16 x float>, <16 x float>* %retval
diff --git a/llvm/test/CodeGen/Thumb2/mve-basic.ll b/llvm/test/CodeGen/Thumb2/mve-basic.ll
index 2b5ef2f014a7..797295c5b201 100644
--- a/llvm/test/CodeGen/Thumb2/mve-basic.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-basic.ll
@@ -44,18 +44,12 @@ define void @vector_f64_copy(<2 x double>* %from, <2 x double>* %to) {
define arm_aapcs_vfpcc <16 x i8> @stack_slot_handling(<16 x i8> %a) #0 {
; CHECK-LABEL: stack_slot_handling:
; CHECK: @ %bb.0: @ %entry
-; CHECK-NEXT: push {r4, r6, r7, lr}
-; CHECK-NEXT: add r7, sp, #8
; CHECK-NEXT: sub sp, #16
-; CHECK-NEXT: mov r4, sp
-; CHECK-NEXT: bfc r4, #0, #4
-; CHECK-NEXT: mov sp, r4
; CHECK-NEXT: mov r0, sp
; CHECK-NEXT: vstrw.32 q0, [r0]
; CHECK-NEXT: vldrw.u32 q0, [r0]
-; CHECK-NEXT: sub.w r4, r7, #8
-; CHECK-NEXT: mov sp, r4
-; CHECK-NEXT: pop {r4, r6, r7, pc}
+; CHECK-NEXT: add sp, #16
+; CHECK-NEXT: bx lr
entry:
%a.addr = alloca <16 x i8>, align 8
store <16 x i8> %a, <16 x i8>* %a.addr, align 8
diff --git a/llvm/test/CodeGen/X86/arg-copy-elide.ll b/llvm/test/CodeGen/X86/arg-copy-elide.ll
index 603e50ff30a3..705a35615561 100644
--- a/llvm/test/CodeGen/X86/arg-copy-elide.ll
+++ b/llvm/test/CodeGen/X86/arg-copy-elide.ll
@@ -53,22 +53,18 @@ entry:
}
; CHECK-LABEL: _split_i64:
-; CHECK: pushl %ebp
-; CHECK: movl %esp, %ebp
; CHECK: pushl %[[csr2:[^ ]*]]
; CHECK: pushl %[[csr1:[^ ]*]]
-; CHECK: andl $-8, %esp
-; CHECK-DAG: movl 8(%ebp), %[[csr1]]
-; CHECK-DAG: movl 12(%ebp), %[[csr2]]
-; CHECK-DAG: leal 8(%ebp), %[[reg:[^ ]*]]
+; CHECK-DAG: movl 12(%esp), %[[csr1]]
+; CHECK-DAG: movl 16(%esp), %[[csr2]]
+; CHECK-DAG: leal 12(%esp), %[[reg:[^ ]*]]
; CHECK: pushl %[[reg]]
; CHECK: calll _addrof_i64
+; CHECK: addl $4, %esp
; CHECK-DAG: movl %[[csr1]], %eax
; CHECK-DAG: movl %[[csr2]], %edx
-; CHECK: leal -8(%ebp), %esp
; CHECK: popl %[[csr1]]
; CHECK: popl %[[csr2]]
-; CHECK: popl %ebp
; CHECK: retl
define i1 @i1_arg(i1 %x) {
@@ -101,16 +97,13 @@ entry:
}
; CHECK-LABEL: _fastcc_split_i64:
-; CHECK: pushl %ebp
-; CHECK: movl %esp, %ebp
; CHECK-DAG: movl %edx, %[[r1:[^ ]*]]
-; CHECK-DAG: movl 8(%ebp), %[[r2:[^ ]*]]
+; CHECK-DAG: movl 20(%esp), %[[r2:[^ ]*]]
; CHECK-DAG: movl %[[r2]], 4(%esp)
; CHECK-DAG: movl %edx, (%esp)
; CHECK: movl %esp, %[[reg:[^ ]*]]
; CHECK: pushl %[[reg]]
; CHECK: calll _addrof_i64
-; CHECK: popl %ebp
; CHECK: retl
diff --git a/llvm/test/CodeGen/X86/avx2-vbroadcast.ll b/llvm/test/CodeGen/X86/avx2-vbroadcast.ll
index da77ee5be7fc..edae4f094c9b 100644
--- a/llvm/test/CodeGen/X86/avx2-vbroadcast.ll
+++ b/llvm/test/CodeGen/X86/avx2-vbroadcast.ll
@@ -1164,9 +1164,9 @@ define void @isel_crash_32b(i8* %cV_R.addr) {
; X64-NEXT: vzeroupper
; X64-NEXT: retq
eintry:
- %__a.addr.i = alloca <4 x i64>, align 16
- %__b.addr.i = alloca <4 x i64>, align 16
- %vCr = alloca <4 x i64>, align 16
+ %__a.addr.i = alloca <4 x i64>, align 32
+ %__b.addr.i = alloca <4 x i64>, align 32
+ %vCr = alloca <4 x i64>, align 32
store <4 x i64> zeroinitializer, <4 x i64>* %vCr, align 16
%tmp = load <4 x i64>, <4 x i64>* %vCr, align 16
%tmp2 = load i8, i8* %cV_R.addr, align 4
@@ -1255,9 +1255,9 @@ define void @isel_crash_16w(i16* %cV_R.addr) {
; X64-NEXT: vzeroupper
; X64-NEXT: retq
eintry:
- %__a.addr.i = alloca <4 x i64>, align 16
- %__b.addr.i = alloca <4 x i64>, align 16
- %vCr = alloca <4 x i64>, align 16
+ %__a.addr.i = alloca <4 x i64>, align 32
+ %__b.addr.i = alloca <4 x i64>, align 32
+ %vCr = alloca <4 x i64>, align 32
store <4 x i64> zeroinitializer, <4 x i64>* %vCr, align 16
%tmp = load <4 x i64>, <4 x i64>* %vCr, align 16
%tmp2 = load i16, i16* %cV_R.addr, align 4
@@ -1346,9 +1346,9 @@ define void @isel_crash_8d(i32* %cV_R.addr) {
; X64-NEXT: vzeroupper
; X64-NEXT: retq
eintry:
- %__a.addr.i = alloca <4 x i64>, align 16
- %__b.addr.i = alloca <4 x i64>, align 16
- %vCr = alloca <4 x i64>, align 16
+ %__a.addr.i = alloca <4 x i64>, align 32
+ %__b.addr.i = alloca <4 x i64>, align 32
+ %vCr = alloca <4 x i64>, align 32
store <4 x i64> zeroinitializer, <4 x i64>* %vCr, align 16
%tmp = load <4 x i64>, <4 x i64>* %vCr, align 16
%tmp2 = load i32, i32* %cV_R.addr, align 4
@@ -1436,9 +1436,9 @@ define void @isel_crash_4q(i64* %cV_R.addr) {
; X64-NEXT: vzeroupper
; X64-NEXT: retq
eintry:
- %__a.addr.i = alloca <4 x i64>, align 16
- %__b.addr.i = alloca <4 x i64>, align 16
- %vCr = alloca <4 x i64>, align 16
+ %__a.addr.i = alloca <4 x i64>, align 32
+ %__b.addr.i = alloca <4 x i64>, align 32
+ %vCr = alloca <4 x i64>, align 32
store <4 x i64> zeroinitializer, <4 x i64>* %vCr, align 16
%tmp = load <4 x i64>, <4 x i64>* %vCr, align 16
%tmp2 = load i64, i64* %cV_R.addr, align 4
diff --git a/llvm/test/CodeGen/X86/avx512-intel-ocl.ll b/llvm/test/CodeGen/X86/avx512-intel-ocl.ll
index 232b598c6109..b92854125298 100644
--- a/llvm/test/CodeGen/X86/avx512-intel-ocl.ll
+++ b/llvm/test/CodeGen/X86/avx512-intel-ocl.ll
@@ -77,7 +77,7 @@ define <16 x float> @testf16_inp(<16 x float> %a, <16 x float> %b) nounwind {
; X64-NEXT: popq %r13
; X64-NEXT: popq %rbp
; X64-NEXT: retq
- %y = alloca <16 x float>, align 16
+ %y = alloca <16 x float>, align 64
%x = fadd <16 x float> %a, %b
%1 = call intel_ocl_bicc <16 x float> @func_float16_ptr(<16 x float> %x, <16 x float>* %y)
%2 = load <16 x float>, <16 x float>* %y, align 16
@@ -158,7 +158,7 @@ define <16 x float> @testf16_regs(<16 x float> %a, <16 x float> %b) nounwind {
; X64-NEXT: popq %r13
; X64-NEXT: popq %rbp
; X64-NEXT: retq
- %y = alloca <16 x float>, align 16
+ %y = alloca <16 x float>, align 64
%x = fadd <16 x float> %a, %b
%1 = call intel_ocl_bicc <16 x float> @func_float16_ptr(<16 x float> %x, <16 x float>* %y)
%2 = load <16 x float>, <16 x float>* %y, align 16
diff --git a/llvm/test/CodeGen/X86/load-local-v3i129.ll b/llvm/test/CodeGen/X86/load-local-v3i129.ll
index 00b53ef9740b..8484f6624690 100644
--- a/llvm/test/CodeGen/X86/load-local-v3i129.ll
+++ b/llvm/test/CodeGen/X86/load-local-v3i129.ll
@@ -4,26 +4,20 @@
define void @_start() {
; CHECK-LABEL: _start:
; CHECK: # %bb.0: # %Entry
-; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %rax
; CHECK-NEXT: .cfi_def_cfa_offset 16
-; CHECK-NEXT: .cfi_offset %rbp, -16
-; CHECK-NEXT: movq %rsp, %rbp
-; CHECK-NEXT: .cfi_def_cfa_register %rbp
-; CHECK-NEXT: andq $-128, %rsp
-; CHECK-NEXT: subq $256, %rsp # imm = 0x100
-; CHECK-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; CHECK-NEXT: movq {{[0-9]+}}(%rsp), %rcx
+; CHECK-NEXT: movq -{{[0-9]+}}(%rsp), %rax
+; CHECK-NEXT: movq -{{[0-9]+}}(%rsp), %rcx
; CHECK-NEXT: shrdq $2, %rcx, %rax
; CHECK-NEXT: shrq $2, %rcx
; CHECK-NEXT: leaq 1(,%rax,4), %rdx
-; CHECK-NEXT: movq %rdx, {{[0-9]+}}(%rsp)
+; CHECK-NEXT: movq %rdx, -{{[0-9]+}}(%rsp)
; CHECK-NEXT: shrdq $62, %rcx, %rax
-; CHECK-NEXT: movq %rax, {{[0-9]+}}(%rsp)
-; CHECK-NEXT: orq $-2, {{[0-9]+}}(%rsp)
-; CHECK-NEXT: movq $-1, {{[0-9]+}}(%rsp)
-; CHECK-NEXT: movq %rbp, %rsp
-; CHECK-NEXT: popq %rbp
-; CHECK-NEXT: .cfi_def_cfa %rsp, 8
+; CHECK-NEXT: movq %rax, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT: orq $-2, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT: movq $-1, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT: popq %rax
+; CHECK-NEXT: .cfi_def_cfa_offset 8
; CHECK-NEXT: retq
Entry:
%y = alloca <3 x i129>, align 4
diff --git a/llvm/test/CodeGen/X86/movtopush.ll b/llvm/test/CodeGen/X86/movtopush.ll
index 7c067beb58ad..9f301a22a481 100644
--- a/llvm/test/CodeGen/X86/movtopush.ll
+++ b/llvm/test/CodeGen/X86/movtopush.ll
@@ -246,7 +246,7 @@ define void @test9() optsize {
entry:
%p = alloca i32, align 4
%q = alloca i32, align 4
- %s = alloca %struct.s, align 4
+ %s = alloca %struct.s, align 8
call void @good(i32 1, i32 2, i32 3, i32 4)
%pv = ptrtoint i32* %p to i32
%qv = ptrtoint i32* %q to i32
@@ -407,7 +407,7 @@ declare void @B_func(%struct.B* sret, %struct.B*, i32)
define void @test14(%struct.A* %a) {
entry:
%ref.tmp = alloca %struct.B, align 1
- %agg.tmp = alloca i64, align 4
+ %agg.tmp = alloca i64, align 8
%tmpcast = bitcast i64* %agg.tmp to %struct.A*
%tmp = alloca %struct.B, align 1
%0 = bitcast %struct.A* %a to i64*
More information about the llvm-commits
mailing list