[llvm] 5e2facb - [AMDGPU] Fix kernel LDS lowering for constants
Stanislav Mekhanoshin via llvm-commits
llvm-commits at lists.llvm.org
Wed May 26 11:34:57 PDT 2021
Author: Stanislav Mekhanoshin
Date: 2021-05-26T11:34:50-07:00
New Revision: 5e2facb922840bfd03eb116eaeead039df021275
URL: https://github.com/llvm/llvm-project/commit/5e2facb922840bfd03eb116eaeead039df021275
DIFF: https://github.com/llvm/llvm-project/commit/5e2facb922840bfd03eb116eaeead039df021275.diff
LOG: [AMDGPU] Fix kernel LDS lowering for constants
There is a trivial but severe bug in the recent code collecting
LDS globals used by kernel. It aborts scan on the first constant
without scanning further uses. That leads to LDS overallocation
with multiple kernels in certain cases.
Differential Revision: https://reviews.llvm.org/D103190
Added:
Modified:
llvm/lib/Target/AMDGPU/Utils/AMDGPULDSUtils.cpp
llvm/test/CodeGen/AMDGPU/ds_read2.ll
llvm/test/CodeGen/AMDGPU/ds_write2.ll
llvm/test/CodeGen/AMDGPU/lower-kernel-lds-constexpr.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPULDSUtils.cpp b/llvm/lib/Target/AMDGPU/Utils/AMDGPULDSUtils.cpp
index 8924072a682fd..8bccf0e5d03e7 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPULDSUtils.cpp
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPULDSUtils.cpp
@@ -81,7 +81,8 @@ bool shouldLowerLDSToStruct(const SmallPtrSetImpl<GlobalValue *> &UsedList,
// variable to be put into a kernel's LDS structure because later
// we will need to replace only this kernel's uses for which we
// need to identify a using function.
- return isUsedOnlyFromFunction(E, F);
+ if (!isUsedOnlyFromFunction(E, F))
+ return false;
}
for (const User *U : E->users()) {
if (Visited.insert(U).second) {
diff --git a/llvm/test/CodeGen/AMDGPU/ds_read2.ll b/llvm/test/CodeGen/AMDGPU/ds_read2.ll
index 26dc97784a229..c21e2ddbf4776 100644
--- a/llvm/test/CodeGen/AMDGPU/ds_read2.ll
+++ b/llvm/test/CodeGen/AMDGPU/ds_read2.ll
@@ -1096,31 +1096,25 @@ define amdgpu_kernel void @sgemm_inner_loop_read2_sequence(float addrspace(1)* %
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
; CI-NEXT: s_lshl_b32 s0, s2, 2
-; CI-NEXT: s_add_i32 s1, s0, 0x8c40
-; CI-NEXT: s_add_i32 s0, s0, 0x8c80
+; CI-NEXT: s_add_i32 s1, s0, 0xc20
+; CI-NEXT: s_addk_i32 s0, 0xc60
+; CI-NEXT: v_mov_b32_e32 v0, s1
; CI-NEXT: v_mov_b32_e32 v2, s0
; CI-NEXT: v_lshlrev_b32_e32 v8, 2, v1
-; CI-NEXT: s_mov_b32 s0, 0x8020
-; CI-NEXT: v_add_i32_e32 v4, vcc, s0, v8
-; CI-NEXT: s_mov_b32 s0, 0x80a0
-; CI-NEXT: v_add_i32_e32 v6, vcc, s0, v8
-; CI-NEXT: v_mov_b32_e32 v0, s1
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: ds_read2_b32 v[0:1], v0 offset1:1
; CI-NEXT: ds_read2_b32 v[2:3], v2 offset1:1
-; CI-NEXT: ds_read2_b32 v[4:5], v4 offset1:1
-; CI-NEXT: ds_read2_b32 v[6:7], v6 offset1:1
-; CI-NEXT: v_add_i32_e32 v8, vcc, 0x8120, v8
+; CI-NEXT: ds_read2_b32 v[4:5], v8 offset1:1
+; CI-NEXT: ds_read2_b32 v[6:7], v8 offset0:32 offset1:33
+; CI-NEXT: ds_read2_b32 v[8:9], v8 offset0:64 offset1:65
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: v_add_f32_e32 v0, v0, v1
; CI-NEXT: v_add_f32_e32 v0, v0, v2
; CI-NEXT: v_add_f32_e32 v0, v0, v3
-; CI-NEXT: ds_read2_b32 v[8:9], v8 offset1:1
; CI-NEXT: v_add_f32_e32 v0, v0, v4
; CI-NEXT: v_add_f32_e32 v0, v0, v5
; CI-NEXT: v_add_f32_e32 v0, v0, v6
; CI-NEXT: v_add_f32_e32 v0, v0, v7
-; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: v_add_f32_e32 v0, v0, v8
; CI-NEXT: s_mov_b32 s7, 0xf000
; CI-NEXT: s_mov_b32 s6, -1
@@ -1131,24 +1125,21 @@ define amdgpu_kernel void @sgemm_inner_loop_read2_sequence(float addrspace(1)* %
; GFX9-LABEL: sgemm_inner_loop_read2_sequence:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_lshl_b32 s2, s2, 2
-; GFX9-NEXT: s_add_i32 s3, s2, 0x8c40
-; GFX9-NEXT: v_lshlrev_b32_e32 v8, 2, v1
-; GFX9-NEXT: s_add_i32 s2, s2, 0x8c80
+; GFX9-NEXT: s_add_i32 s3, s2, 0xc20
+; GFX9-NEXT: s_addk_i32 s2, 0xc60
; GFX9-NEXT: v_mov_b32_e32 v0, s3
; GFX9-NEXT: v_mov_b32_e32 v2, s2
-; GFX9-NEXT: v_add_u32_e32 v4, 0x8020, v8
-; GFX9-NEXT: v_add_u32_e32 v6, 0x80a0, v8
+; GFX9-NEXT: v_lshlrev_b32_e32 v8, 2, v1
; GFX9-NEXT: ds_read2_b32 v[0:1], v0 offset1:1
; GFX9-NEXT: ds_read2_b32 v[2:3], v2 offset1:1
-; GFX9-NEXT: ds_read2_b32 v[4:5], v4 offset1:1
-; GFX9-NEXT: ds_read2_b32 v[6:7], v6 offset1:1
-; GFX9-NEXT: v_add_u32_e32 v8, 0x8120, v8
-; GFX9-NEXT: s_waitcnt lgkmcnt(3)
+; GFX9-NEXT: ds_read2_b32 v[4:5], v8 offset1:1
+; GFX9-NEXT: ds_read2_b32 v[6:7], v8 offset0:32 offset1:33
+; GFX9-NEXT: ds_read2_b32 v[8:9], v8 offset0:64 offset1:65
+; GFX9-NEXT: s_waitcnt lgkmcnt(4)
; GFX9-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX9-NEXT: s_waitcnt lgkmcnt(2)
+; GFX9-NEXT: s_waitcnt lgkmcnt(3)
; GFX9-NEXT: v_add_f32_e32 v0, v0, v2
; GFX9-NEXT: v_add_f32_e32 v0, v0, v3
-; GFX9-NEXT: ds_read2_b32 v[8:9], v8 offset1:1
; GFX9-NEXT: s_waitcnt lgkmcnt(2)
; GFX9-NEXT: v_add_f32_e32 v0, v0, v4
; GFX9-NEXT: v_add_f32_e32 v0, v0, v5
@@ -1474,11 +1465,11 @@ define amdgpu_kernel void @read2_v2i32_align1_odd_offset(<2 x i32> addrspace(1)*
; CI: ; %bb.0: ; %entry
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: s_mov_b32 m0, -1
-; CI-NEXT: ds_read_u8 v1, v0 offset:37032
-; CI-NEXT: ds_read_u8 v2, v0 offset:37031
-; CI-NEXT: ds_read_u8 v3, v0 offset:37030
-; CI-NEXT: ds_read_u8 v4, v0 offset:37029
-; CI-NEXT: ds_read_u8 v5, v0 offset:37028
+; CI-NEXT: ds_read_u8 v1, v0 offset:72
+; CI-NEXT: ds_read_u8 v2, v0 offset:71
+; CI-NEXT: ds_read_u8 v3, v0 offset:70
+; CI-NEXT: ds_read_u8 v4, v0 offset:69
+; CI-NEXT: ds_read_u8 v5, v0 offset:68
; CI-NEXT: s_waitcnt lgkmcnt(4)
; CI-NEXT: v_lshlrev_b32_e32 v1, 8, v1
; CI-NEXT: s_waitcnt lgkmcnt(3)
@@ -1489,9 +1480,9 @@ define amdgpu_kernel void @read2_v2i32_align1_odd_offset(<2 x i32> addrspace(1)*
; CI-NEXT: v_or_b32_e32 v3, v3, v4
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; CI-NEXT: v_or_b32_e32 v1, v1, v3
-; CI-NEXT: ds_read_u8 v2, v0 offset:37027
-; CI-NEXT: ds_read_u8 v3, v0 offset:37026
-; CI-NEXT: ds_read_u8 v0, v0 offset:37025
+; CI-NEXT: ds_read_u8 v2, v0 offset:67
+; CI-NEXT: ds_read_u8 v3, v0 offset:66
+; CI-NEXT: ds_read_u8 v0, v0 offset:65
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
@@ -1508,14 +1499,14 @@ define amdgpu_kernel void @read2_v2i32_align1_odd_offset(<2 x i32> addrspace(1)*
; GFX9-ALIGNED-LABEL: read2_v2i32_align1_odd_offset:
; GFX9-ALIGNED: ; %bb.0: ; %entry
; GFX9-ALIGNED-NEXT: v_mov_b32_e32 v2, 0
-; GFX9-ALIGNED-NEXT: ds_read_u8 v0, v2 offset:37025
-; GFX9-ALIGNED-NEXT: ds_read_u8 v3, v2 offset:37026
-; GFX9-ALIGNED-NEXT: ds_read_u8 v4, v2 offset:37027
-; GFX9-ALIGNED-NEXT: ds_read_u8 v5, v2 offset:37028
-; GFX9-ALIGNED-NEXT: ds_read_u8 v1, v2 offset:37029
-; GFX9-ALIGNED-NEXT: ds_read_u8 v6, v2 offset:37030
-; GFX9-ALIGNED-NEXT: ds_read_u8 v7, v2 offset:37031
-; GFX9-ALIGNED-NEXT: ds_read_u8 v8, v2 offset:37032
+; GFX9-ALIGNED-NEXT: ds_read_u8 v0, v2 offset:65
+; GFX9-ALIGNED-NEXT: ds_read_u8 v3, v2 offset:66
+; GFX9-ALIGNED-NEXT: ds_read_u8 v4, v2 offset:67
+; GFX9-ALIGNED-NEXT: ds_read_u8 v5, v2 offset:68
+; GFX9-ALIGNED-NEXT: ds_read_u8 v1, v2 offset:69
+; GFX9-ALIGNED-NEXT: ds_read_u8 v6, v2 offset:70
+; GFX9-ALIGNED-NEXT: ds_read_u8 v7, v2 offset:71
+; GFX9-ALIGNED-NEXT: ds_read_u8 v8, v2 offset:72
; GFX9-ALIGNED-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-ALIGNED-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-ALIGNED-NEXT: v_lshlrev_b32_e32 v3, 8, v3
@@ -1533,7 +1524,7 @@ define amdgpu_kernel void @read2_v2i32_align1_odd_offset(<2 x i32> addrspace(1)*
;
; GFX9-UNALIGNED-LABEL: read2_v2i32_align1_odd_offset:
; GFX9-UNALIGNED: ; %bb.0: ; %entry
-; GFX9-UNALIGNED-NEXT: v_mov_b32_e32 v0, 0x90a1
+; GFX9-UNALIGNED-NEXT: v_mov_b32_e32 v0, 0x41
; GFX9-UNALIGNED-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-UNALIGNED-NEXT: ds_read2_b32 v[0:1], v0 offset1:1
; GFX9-UNALIGNED-NEXT: v_mov_b32_e32 v2, 0
diff --git a/llvm/test/CodeGen/AMDGPU/ds_write2.ll b/llvm/test/CodeGen/AMDGPU/ds_write2.ll
index 3375de99e29f3..b680383e8ca17 100644
--- a/llvm/test/CodeGen/AMDGPU/ds_write2.ll
+++ b/llvm/test/CodeGen/AMDGPU/ds_write2.ll
@@ -889,9 +889,9 @@ define amdgpu_kernel void @write2_sgemm_sequence(float addrspace(1)* %C, i32 %ld
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
; CI-NEXT: s_lshl_b32 s2, s2, 2
-; CI-NEXT: s_add_i32 s3, s2, 0x8c40
+; CI-NEXT: s_add_i32 s3, s2, 0xc20
; CI-NEXT: v_mov_b32_e32 v0, s3
-; CI-NEXT: s_add_i32 s2, s2, 0x8c80
+; CI-NEXT: s_addk_i32 s2, 0xc60
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_load_dword s0, s[0:1], 0x0
; CI-NEXT: s_mov_b32 m0, -1
@@ -902,20 +902,17 @@ define amdgpu_kernel void @write2_sgemm_sequence(float addrspace(1)* %C, i32 %ld
; CI-NEXT: v_mov_b32_e32 v0, s2
; CI-NEXT: ds_write2_b32 v0, v2, v3 offset1:1
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v1
-; CI-NEXT: v_add_i32_e32 v1, vcc, 0x8020, v0
-; CI-NEXT: ds_write2_b32 v1, v2, v3 offset1:1
-; CI-NEXT: v_add_i32_e32 v1, vcc, 0x80a0, v0
-; CI-NEXT: v_add_i32_e32 v0, vcc, 0x8120, v0
-; CI-NEXT: ds_write2_b32 v1, v2, v3 offset1:1
; CI-NEXT: ds_write2_b32 v0, v2, v3 offset1:1
+; CI-NEXT: ds_write2_b32 v0, v2, v3 offset0:32 offset1:33
+; CI-NEXT: ds_write2_b32 v0, v2, v3 offset0:64 offset1:65
; CI-NEXT: s_endpgm
;
; GFX9-LABEL: write2_sgemm_sequence:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NEXT: s_lshl_b32 s2, s2, 2
-; GFX9-NEXT: s_add_i32 s3, s2, 0x8c40
-; GFX9-NEXT: s_add_i32 s2, s2, 0x8c80
+; GFX9-NEXT: s_add_i32 s3, s2, 0xc20
+; GFX9-NEXT: s_addk_i32 s2, 0xc60
; GFX9-NEXT: v_mov_b32_e32 v0, s3
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_load_dword s0, s[0:1], 0x0
@@ -926,12 +923,9 @@ define amdgpu_kernel void @write2_sgemm_sequence(float addrspace(1)* %C, i32 %ld
; GFX9-NEXT: ds_write2_b32 v0, v3, v4 offset1:1
; GFX9-NEXT: ds_write2_b32 v2, v3, v4 offset1:1
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v1
-; GFX9-NEXT: v_add_u32_e32 v1, 0x8020, v0
-; GFX9-NEXT: ds_write2_b32 v1, v3, v4 offset1:1
-; GFX9-NEXT: v_add_u32_e32 v1, 0x80a0, v0
-; GFX9-NEXT: v_add_u32_e32 v0, 0x8120, v0
-; GFX9-NEXT: ds_write2_b32 v1, v3, v4 offset1:1
; GFX9-NEXT: ds_write2_b32 v0, v3, v4 offset1:1
+; GFX9-NEXT: ds_write2_b32 v0, v3, v4 offset0:32 offset1:33
+; GFX9-NEXT: ds_write2_b32 v0, v3, v4 offset0:64 offset1:65
; GFX9-NEXT: s_endpgm
%x.i = tail call i32 @llvm.amdgcn.workgroup.id.x() #1
%y.i = tail call i32 @llvm.amdgcn.workitem.id.y() #1
@@ -1032,37 +1026,37 @@ define amdgpu_kernel void @write2_v2i32_align1_odd_offset() {
; CI-NEXT: v_mov_b32_e32 v0, 0x7b
; CI-NEXT: v_mov_b32_e32 v1, 0
; CI-NEXT: s_mov_b32 m0, -1
-; CI-NEXT: ds_write_b8 v1, v0 offset:37025
+; CI-NEXT: ds_write_b8 v1, v0 offset:65
; CI-NEXT: v_mov_b32_e32 v0, 1
-; CI-NEXT: ds_write_b8 v1, v0 offset:37030
+; CI-NEXT: ds_write_b8 v1, v0 offset:70
; CI-NEXT: v_mov_b32_e32 v0, 0xc8
-; CI-NEXT: ds_write_b8 v1, v0 offset:37029
-; CI-NEXT: ds_write_b8 v1, v1 offset:37028
-; CI-NEXT: ds_write_b8 v1, v1 offset:37027
-; CI-NEXT: ds_write_b8 v1, v1 offset:37026
-; CI-NEXT: ds_write_b8 v1, v1 offset:37032
-; CI-NEXT: ds_write_b8 v1, v1 offset:37031
+; CI-NEXT: ds_write_b8 v1, v0 offset:69
+; CI-NEXT: ds_write_b8 v1, v1 offset:68
+; CI-NEXT: ds_write_b8 v1, v1 offset:67
+; CI-NEXT: ds_write_b8 v1, v1 offset:66
+; CI-NEXT: ds_write_b8 v1, v1 offset:72
+; CI-NEXT: ds_write_b8 v1, v1 offset:71
; CI-NEXT: s_endpgm
;
; GFX9-ALIGNED-LABEL: write2_v2i32_align1_odd_offset:
; GFX9-ALIGNED: ; %bb.0: ; %entry
; GFX9-ALIGNED-NEXT: v_mov_b32_e32 v0, 0x7b
; GFX9-ALIGNED-NEXT: v_mov_b32_e32 v1, 0
-; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v0 offset:37025
+; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v0 offset:65
; GFX9-ALIGNED-NEXT: v_mov_b32_e32 v0, 1
-; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v0 offset:37030
+; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v0 offset:70
; GFX9-ALIGNED-NEXT: v_mov_b32_e32 v0, 0xc8
-; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v0 offset:37029
-; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:37028
-; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:37027
-; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:37026
-; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:37032
-; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:37031
+; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v0 offset:69
+; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:68
+; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:67
+; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:66
+; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:72
+; GFX9-ALIGNED-NEXT: ds_write_b8 v1, v1 offset:71
; GFX9-ALIGNED-NEXT: s_endpgm
;
; GFX9-UNALIGNED-LABEL: write2_v2i32_align1_odd_offset:
; GFX9-UNALIGNED: ; %bb.0: ; %entry
-; GFX9-UNALIGNED-NEXT: v_mov_b32_e32 v0, 0x90a1
+; GFX9-UNALIGNED-NEXT: v_mov_b32_e32 v0, 0x41
; GFX9-UNALIGNED-NEXT: v_mov_b32_e32 v1, 0x7b
; GFX9-UNALIGNED-NEXT: v_mov_b32_e32 v2, 0x1c8
; GFX9-UNALIGNED-NEXT: ds_write2_b32 v0, v1, v2 offset1:1
diff --git a/llvm/test/CodeGen/AMDGPU/lower-kernel-lds-constexpr.ll b/llvm/test/CodeGen/AMDGPU/lower-kernel-lds-constexpr.ll
index ac7aa00c5a32c..7ca6eaad61b26 100644
--- a/llvm/test/CodeGen/AMDGPU/lower-kernel-lds-constexpr.ll
+++ b/llvm/test/CodeGen/AMDGPU/lower-kernel-lds-constexpr.ll
@@ -2,6 +2,7 @@
; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s | FileCheck %s
; CHECK: %llvm.amdgcn.kernel.k2.lds.t = type { i32 }
+; CHECK-NOT: %llvm.amdgcn.kernel.k4.lds.t
@lds.1 = internal unnamed_addr addrspace(3) global [2 x i8] undef, align 1
@@ -66,3 +67,16 @@ define amdgpu_kernel void @k3(i64 %x) {
store i64 2, i64* %ptr2, align 1
ret void
}
+
+; @lds.1 is used from constant expressions in
diff erent kernels.
+; Make sure we do not create a structure for it as we cannot handle it yet.
+define amdgpu_kernel void @k4(i64 %x) {
+; CHECK-LABEL: @k4(
+; CHECK-NEXT: %ptr = getelementptr inbounds i8, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([2 x i8], [2 x i8] addrspace(3)* @lds.1, i32 0, i32 0) to i8*), i64 %x
+; CHECK-NEXT: store i8 1, i8* %ptr, align 1
+; CHECK-NEXT: ret void
+;
+ %ptr = getelementptr inbounds i8, i8* addrspacecast ([2 x i8] addrspace(3)* @lds.1 to i8*), i64 %x
+ store i8 1, i8 addrspace(0)* %ptr, align 1
+ ret void
+}
More information about the llvm-commits
mailing list