Skip to content

Commit 7b7f5d5

Browse files
arsenmarsenm
committed
AMDGPU: Cleanup selection patterns for buffer loads
We should just support these for all register types.
1 parent 5021e6d commit 7b7f5d5

File tree

2 files changed

+39
-49
lines changed

2 files changed

+39
-49
lines changed

‎llvm/lib/Target/AMDGPU/BUFInstructions.td

Lines changed: 30 additions & 42 deletions
Original file line numberDiff line numberDiff line change
@@ -1419,27 +1419,21 @@ let OtherPredicates = [HasPackedD16VMem] in {
14191419
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_d16, v4i16, "BUFFER_LOAD_FORMAT_D16_XYZW">;
14201420
} // End HasPackedD16VMem.
14211421

1422-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, f32, "BUFFER_LOAD_DWORD">;
1423-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, i32, "BUFFER_LOAD_DWORD">;
1424-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2i16, "BUFFER_LOAD_DWORD">;
1425-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2f16, "BUFFER_LOAD_DWORD">;
1426-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2bf16, "BUFFER_LOAD_DWORD">;
1427-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2f32, "BUFFER_LOAD_DWORDX2">;
1428-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2i32, "BUFFER_LOAD_DWORDX2">;
1429-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v4i16, "BUFFER_LOAD_DWORDX2">;
1430-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v4f16, "BUFFER_LOAD_DWORDX2">;
1431-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, i64, "BUFFER_LOAD_DWORDX2">;
1432-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, f64, "BUFFER_LOAD_DWORDX2">;
1433-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v4bf16, "BUFFER_LOAD_DWORDX2">;
1434-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v3f32, "BUFFER_LOAD_DWORDX3">;
1435-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v3i32, "BUFFER_LOAD_DWORDX3">;
1436-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v4f32, "BUFFER_LOAD_DWORDX4">;
1437-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v4i32, "BUFFER_LOAD_DWORDX4">;
1438-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2i64, "BUFFER_LOAD_DWORDX4">;
1439-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2f64, "BUFFER_LOAD_DWORDX4">;
1440-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v8i16, "BUFFER_LOAD_DWORDX4">;
1441-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v8f16, "BUFFER_LOAD_DWORDX4">;
1442-
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v8bf16, "BUFFER_LOAD_DWORDX4">;
1422+
foreach vt = Reg32Types.types in {
1423+
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, vt, "BUFFER_LOAD_DWORD">;
1424+
}
1425+
1426+
foreach vt = Reg64Types.types in {
1427+
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, vt, "BUFFER_LOAD_DWORDX2">;
1428+
}
1429+
1430+
foreach vt = Reg96Types.types in {
1431+
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, vt, "BUFFER_LOAD_DWORDX3">;
1432+
}
1433+
1434+
foreach vt = Reg128Types.types in {
1435+
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, vt, "BUFFER_LOAD_DWORDX4">;
1436+
}
14431437

14441438
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_byte, i32, "BUFFER_LOAD_SBYTE">;
14451439
defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_short, i32, "BUFFER_LOAD_SSHORT">;
@@ -1530,27 +1524,21 @@ let OtherPredicates = [HasPackedD16VMem] in {
15301524
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format_d16, v4i16, "BUFFER_STORE_FORMAT_D16_XYZW">;
15311525
} // End HasPackedD16VMem.
15321526

1533-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, f32, "BUFFER_STORE_DWORD">;
1534-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, i32, "BUFFER_STORE_DWORD">;
1535-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2i16, "BUFFER_STORE_DWORD">;
1536-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2f16, "BUFFER_STORE_DWORD">;
1537-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2bf16, "BUFFER_STORE_DWORD">;
1538-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2f32, "BUFFER_STORE_DWORDX2">;
1539-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2i32, "BUFFER_STORE_DWORDX2">;
1540-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, i64, "BUFFER_STORE_DWORDX2">;
1541-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, f64, "BUFFER_STORE_DWORDX2">;
1542-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v4i16, "BUFFER_STORE_DWORDX2">;
1543-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v4f16, "BUFFER_STORE_DWORDX2">;
1544-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v4bf16, "BUFFER_STORE_DWORDX2">;
1545-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v3f32, "BUFFER_STORE_DWORDX3">;
1546-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v3i32, "BUFFER_STORE_DWORDX3">;
1547-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v4f32, "BUFFER_STORE_DWORDX4">;
1548-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v4i32, "BUFFER_STORE_DWORDX4">;
1549-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2i64, "BUFFER_STORE_DWORDX4">;
1550-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2f64, "BUFFER_STORE_DWORDX4">;
1551-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v8f16, "BUFFER_STORE_DWORDX4">;
1552-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v8i16, "BUFFER_STORE_DWORDX4">;
1553-
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v8bf16, "BUFFER_STORE_DWORDX4">;
1527+
foreach vt = Reg32Types.types in {
1528+
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, vt, "BUFFER_STORE_DWORD">;
1529+
}
1530+
1531+
foreach vt = Reg64Types.types in {
1532+
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, vt, "BUFFER_STORE_DWORDX2">;
1533+
}
1534+
1535+
foreach vt = Reg96Types.types in {
1536+
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, vt, "BUFFER_STORE_DWORDX3">;
1537+
}
1538+
1539+
foreach vt = Reg128Types.types in {
1540+
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, vt, "BUFFER_STORE_DWORDX4">;
1541+
}
15541542

15551543
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_byte, i32, "BUFFER_STORE_BYTE">;
15561544
defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_short, i32, "BUFFER_STORE_SHORT">;

‎llvm/lib/Target/AMDGPU/SIRegisterInfo.td

Lines changed: 9 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -586,7 +586,9 @@ class RegisterTypes<list<ValueType> reg_types> {
586586

587587
def Reg16Types : RegisterTypes<[i16, f16, bf16]>;
588588
def Reg32Types : RegisterTypes<[i32, f32, v2i16, v2f16, v2bf16, p2, p3, p5, p6]>;
589-
def Reg64Types : RegisterTypes<[i64, f64, v2i32, v2f32, v4i16, v4f16, v4bf16, p0]>;
589+
def Reg64Types : RegisterTypes<[i64, f64, v2i32, v2f32, p0, v4i16, v4f16, v4bf16]>;
590+
def Reg96Types : RegisterTypes<[v3i32, v3f32]>;
591+
def Reg128Types : RegisterTypes<[v4i32, v4f32, v2i64, v2f64, v8i16, v8f16, v8bf16]>;
590592

591593
let HasVGPR = 1 in {
592594
// VOP3 and VINTERP can access 256 lo and 256 hi registers.
@@ -744,7 +746,7 @@ def Pseudo_SReg_32 : SIRegisterClass<"AMDGPU", [i32, f32, i16, f16, bf16, v2i16,
744746
let BaseClassOrder = 10000;
745747
}
746748

747-
def Pseudo_SReg_128 : SIRegisterClass<"AMDGPU", [v4i32, v2i64, v2f64, v8i16, v8f16, v8bf16], 32,
749+
def Pseudo_SReg_128 : SIRegisterClass<"AMDGPU", Reg128Types.types, 32,
748750
(add PRIVATE_RSRC_REG)> {
749751
let isAllocatable = 0;
750752
let CopyCost = -1;
@@ -815,7 +817,7 @@ def SRegOrLds_32 : SIRegisterClass<"AMDGPU", [i32, f32, i16, f16, bf16, v2i16, v
815817
let HasSGPR = 1;
816818
}
817819

818-
def SGPR_64 : SIRegisterClass<"AMDGPU", [v2i32, i64, v2f32, f64, v4i16, v4f16, v4bf16], 32,
820+
def SGPR_64 : SIRegisterClass<"AMDGPU", Reg64Types.types, 32,
819821
(add SGPR_64Regs)> {
820822
let CopyCost = 1;
821823
let AllocationPriority = 1;
@@ -905,8 +907,8 @@ multiclass SRegClass<int numRegs,
905907
}
906908
}
907909

908-
defm "" : SRegClass<3, [v3i32, v3f32], SGPR_96Regs, TTMP_96Regs>;
909-
defm "" : SRegClass<4, [v4i32, v4f32, v2i64, v2f64, v8i16, v8f16, v8bf16], SGPR_128Regs, TTMP_128Regs>;
910+
defm "" : SRegClass<3, Reg96Types.types, SGPR_96Regs, TTMP_96Regs>;
911+
defm "" : SRegClass<4, Reg128Types.types, SGPR_128Regs, TTMP_128Regs>;
910912
defm "" : SRegClass<5, [v5i32, v5f32], SGPR_160Regs, TTMP_160Regs>;
911913
defm "" : SRegClass<6, [v6i32, v6f32, v3i64, v3f64], SGPR_192Regs, TTMP_192Regs>;
912914
defm "" : SRegClass<7, [v7i32, v7f32], SGPR_224Regs, TTMP_224Regs>;
@@ -958,8 +960,8 @@ multiclass VRegClass<int numRegs, list<ValueType> regTypes, dag regList> {
958960

959961
defm VReg_64 : VRegClass<2, [i64, f64, v2i32, v2f32, v4f16, v4bf16, v4i16, p0, p1, p4],
960962
(add VGPR_64)>;
961-
defm VReg_96 : VRegClass<3, [v3i32, v3f32], (add VGPR_96)>;
962-
defm VReg_128 : VRegClass<4, [v4i32, v4f32, v2i64, v2f64, v8i16, v8f16, v8bf16], (add VGPR_128)>;
963+
defm VReg_96 : VRegClass<3, Reg96Types.types, (add VGPR_96)>;
964+
defm VReg_128 : VRegClass<4, Reg128Types.types, (add VGPR_128)>;
963965
defm VReg_160 : VRegClass<5, [v5i32, v5f32], (add VGPR_160)>;
964966

965967
defm VReg_192 : VRegClass<6, [v6i32, v6f32, v3i64, v3f64], (add VGPR_192)>;

0 commit comments

Comments
 (0)