Skip to content

Commit 3dfca24

Browse files
rampitecrampitec
authored
[AMDGPU] Fix encoding of VOP3P dpp on GFX11 and GFX12 (llvm#82710)
The bug affects dpp forms of v_dot2_f32_f16. The encoding does not match SP3 and does not set op_sel_hi bits properly.
1 parent e132643 commit 3dfca24

File tree

11 files changed

+25
-22
lines changed

11 files changed

+25
-22
lines changed

‎llvm/lib/Target/AMDGPU/VOP3PInstructions.td‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1353,6 +1353,7 @@ class VOP3P_DPP16<bits<7> op, VOP_DPP_Pseudo ps, int subtarget,
13531353
let AssemblerPredicate = HasDPP16;
13541354
let SubtargetPredicate = HasDPP16;
13551355
let OtherPredicates = ps.OtherPredicates;
1356+
let IsPacked = ps.IsPacked;
13561357
}
13571358

13581359
class VOP3P_DPP8_Base<bits<7> op, VOP_Pseudo ps, string opName = ps.OpName>
@@ -1362,6 +1363,7 @@ class VOP3P_DPP8_Base<bits<7> op, VOP_Pseudo ps, string opName = ps.OpName>
13621363
let SchedRW = ps.SchedRW;
13631364
let Uses = ps.Uses;
13641365
let OtherPredicates = ps.OtherPredicates;
1366+
let IsPacked = ps.IsPacked;
13651367
}
13661368

13671369
//===----------------------------------------------------------------------===//

‎llvm/lib/Target/AMDGPU/VOPInstructions.td‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -818,6 +818,7 @@ class VOP_DPP_Pseudo <string OpName, VOPProfile P, list<dag> pattern=[],
818818
let VALU = 1;
819819
let DPP = 1;
820820
let Size = 8;
821+
let IsPacked = P.IsPacked;
821822

822823
let ReadsModeReg = !or(P.DstVT.isFP, P.Src0VT.isFP);
823824

‎llvm/test/MC/AMDGPU/gfx11-promotions.s‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -337,17 +337,17 @@ v_dot2_f32_f16_e64 v0, v1, v2, v3
337337
//===----------------------------------------------------------------------===//
338338

339339
v_dot2_f32_f16 v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0]
340-
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x00,0x13,0xcc,0xe9,0x04,0x0e,0x04,0x01,0x77,0x39,0x05]
340+
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x40,0x13,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x77,0x39,0x05]
341341

342342
v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0]
343-
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x00,0x13,0xcc,0xe9,0x04,0x0e,0x04,0x01,0x77,0x39,0x05]
343+
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x40,0x13,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x77,0x39,0x05]
344344

345345
//===----------------------------------------------------------------------===//
346346
// VOP3P.DPP16.
347347
//===----------------------------------------------------------------------===//
348348

349349
v_dot2_f32_f16 v0, v1, v2, v3 quad_perm:[1,2,3,0]
350-
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x39,0x00,0xff]
350+
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
351351

352352
v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0]
353-
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x39,0x00,0xff]
353+
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]

‎llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -2,10 +2,10 @@
22
// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -show-encoding %s | FileCheck --check-prefixes=GFX11 %s
33

44
v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[0,0,0] neg_hi:[0,0,0] quad_perm:[2,2,3,1] bound_ctrl:0 fi:1
5-
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x04,0xff]
5+
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
66

77
v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] bank_mask:0xe
8-
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
8+
// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
99

1010
v_fma_mix_f32 v0, v1, v2, v3 op_sel:[0,0,0] row_ror:7 bank_mask:0x1 bound_ctrl:0
1111
// GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x01,0xf1]

‎llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -15,4 +15,4 @@ v_fma_mixlo_f16 v0, abs(v1), -v2, abs(v3) op_sel:[1,0,0] op_sel_hi:[1,0,0] dpp8:
1515
// GFX11: encoding: [0x00,0x0d,0x21,0xcc,0xe9,0x04,0x0e,0x4c,0x01,0x92,0x44,0x92]
1616

1717
v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0]
18-
// GFX11: encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
18+
// GFX11: encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]

‎llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -2,10 +2,10 @@
22
// RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -show-encoding %s | FileCheck --check-prefixes=GFX12 %s
33

44
v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[0,0,0] neg_hi:[0,0,0] quad_perm:[2,2,3,1] bound_ctrl:0 fi:1
5-
// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x04,0xff]
5+
// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
66

77
v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] bank_mask:0xe
8-
// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
8+
// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
99

1010
v_fma_mix_f32 v0, v1, v2, v3 op_sel:[0,0,0] row_ror:7 bank_mask:0x1 bound_ctrl:0
1111
// GFX12: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x01,0xf1]

‎llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -15,7 +15,7 @@ v_fma_mixlo_f16 v0, abs(v1), -v2, abs(v3) op_sel:[1,0,0] op_sel_hi:[1,0,0] dpp8:
1515
// GFX12: encoding: [0x00,0x0d,0x21,0xcc,0xe9,0x04,0x0e,0x4c,0x01,0x92,0x44,0x92]
1616

1717
v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0]
18-
// GFX12: encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
18+
// GFX12: encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
1919

2020
v_dot4_f32_fp8_bf8 v0, v1, v2, v3 dpp8:[0,1,2,3,4,5,6,7]
2121
// GFX12: v_dot4_f32_fp8_bf8_e64_dpp v0, v1, v2, v3 dpp8:[0,1,2,3,4,5,6,7] ; encoding: [0x00,0x40,0x24,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x88,0xc6,0xfa]

‎llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,11 @@
11
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
22
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-WavefrontSize32,+WavefrontSize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
33

4-
# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
5-
0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe
4+
# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
5+
0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe
66

7-
# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff]
8-
0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff
7+
# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff]
8+
0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
99

1010
# GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 bound_ctrl:1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1]
1111
0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1

‎llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,8 +1,8 @@
11
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
22
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-WavefrontSize32,+WavefrontSize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
33

4-
# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
5-
0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05
4+
# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
5+
0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05
66

77
# GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 clamp dpp8:[2,2,2,2,4,4,4,4] fi:1 ; encoding: [0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92]
88
0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92

‎llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,11 @@
11
# RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
22
# RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
33

4-
# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
5-
0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe
4+
# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
5+
0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe
66

7-
# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff]
8-
0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff
7+
# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff]
8+
0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
99

1010
# GFX12: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 bound_ctrl:1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1]
1111
0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1

0 commit comments

Comments
 (0)