Skip to content

Commit 64c094b

Browse files
jwanggit86jwanggit86
authored
[AMDGPU][MC] Allow dpp in v_dot2_f32_bf16 for GFX11 and 12 (#142451)
Allowing the dpp operand in v_dot2_f32_bf16 for GFX11 and 12.
1 parent eb83c43 commit 64c094b

File tree

9 files changed

+55
-1
lines changed

9 files changed

+55
-1
lines changed

‎llvm/lib/Target/AMDGPU/VOP3PInstructions.td

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1856,6 +1856,9 @@ multiclass VOP3P_Realtriple<GFXGen Gen, bits<8> op, string backing_ps_name = NAM
18561856
VOP3P_Real_dpp<Gen, op, backing_ps_name, asmName>,
18571857
VOP3P_Real_dpp8<Gen, op, backing_ps_name, asmName>;
18581858

1859+
multiclass VOP3P_Realtriple_gfx11_gfx12<bits<8> op>
1860+
: VOP3P_Realtriple<GFX11Gen, op>, VOP3P_Realtriple<GFX12Gen, op>;
1861+
18591862
//===----------------------------------------------------------------------===//
18601863
// GFX12
18611864
//===----------------------------------------------------------------------===//
@@ -1887,7 +1890,7 @@ multiclass VOP3P_Real_gfx11_gfx12<bits<8> op> :
18871890

18881891
defm V_DOT4_I32_IU8 : VOP3P_Real_gfx11_gfx12<0x16>;
18891892
defm V_DOT8_I32_IU4 : VOP3P_Real_gfx11_gfx12<0x18>;
1890-
defm V_DOT2_F32_BF16 : VOP3P_Real_gfx11_gfx12<0x1a>;
1893+
defm V_DOT2_F32_BF16 : VOP3P_Realtriple_gfx11_gfx12<0x1a>;
18911894

18921895
let AssemblerPredicate = isGFX11Plus in {
18931896
def : AMDGPUMnemonicAlias<"v_dot4_i32_i8", "v_dot4_i32_iu8">;

‎llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp16.s

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5390,3 +5390,15 @@ v_dot2_bf16_bf16_e64_dpp v5.l, -|v1|, -|v2|, -1 row_xmask:0 row_mask:0x1 bank_ma
53905390

53915391
v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
53925392
// GFX11: v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| op_sel:[0,0,0,1] row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x47,0x67,0xd6,0xfa,0xfe,0xf7,0xe3,0xff,0x6f,0x05,0x30]
5393+
5394+
v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
5395+
// GFX11: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
5396+
5397+
v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
5398+
// GFX11: v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
5399+
5400+
v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
5401+
// GFX11: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
5402+
5403+
v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0x00]
5404+
// GFX11: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0x00]

‎llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp8.s

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3630,3 +3630,9 @@ v_dot2_bf16_bf16_e64_dpp v5.l, -|v1|, -|v2|, -1 dpp8:[7,6,5,4,3,2,1,0] fi:1
36303630

36313631
v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| dpp8:[0,0,0,0,0,0,0,0] fi:0
36323632
// GFX11: v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| op_sel:[0,0,0,1] dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x47,0x67,0xd6,0xe9,0xfe,0xf7,0xe3,0xff,0x00,0x00,0x00]
3633+
3634+
v_dot2_f32_bf16_e64_dpp v5, v1, v2, s3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x40,0x1a,0xcc,0xe9,0x04,0x0e,0x18,0x01,0x77,0x39,0x05]
3635+
// GFX11: v_dot2_f32_bf16_e64_dpp v5, v1, v2, s3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x40,0x1a,0xcc,0xe9,0x04,0x0e,0x18,0x01,0x77,0x39,0x05]
3636+
3637+
v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
3638+
// GFX11: v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]

‎llvm/test/MC/AMDGPU/gfx12_asm_vop3_dpp16.s

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5645,6 +5645,18 @@ v_dot2_bf16_bf16_e64_dpp v5.l, v1, v2, v255.h row_mirror
56455645
v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
56465646
// GFX12: v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| op_sel:[0,0,0,1] row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x47,0x67,0xd6,0xfa,0xfe,0xf7,0xe3,0xff,0x6f,0x05,0x30]
56475647

5648+
v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
5649+
// GFX12: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
5650+
5651+
v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
5652+
// GFX12: v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
5653+
5654+
v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
5655+
// GFX12: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
5656+
5657+
v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0x00]
5658+
// GFX12: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0x00]
5659+
56485660
v_minimum_f32 v5, v1, v2 quad_perm:[3,2,1,0]
56495661
// GFX12: v_minimum_f32_e64_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x65,0xd7,0xfa,0x04,0x02,0x00,0x01,0x1b,0x00,0xff]
56505662

‎llvm/test/MC/AMDGPU/gfx12_asm_vop3_dpp8.s

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3902,6 +3902,12 @@ v_dot2_bf16_bf16_e64_dpp v5.l, v1, v2, v255.h dpp8:[7,6,5,4,3,2,1,0]
39023902
v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| dpp8:[0,0,0,0,0,0,0,0] fi:0
39033903
// GFX12: v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| op_sel:[0,0,0,1] dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x47,0x67,0xd6,0xe9,0xfe,0xf7,0xe3,0xff,0x00,0x00,0x00]
39043904

3905+
v_dot2_f32_bf16_e64_dpp v5, v1, v2, s3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x40,0x1a,0xcc,0xe9,0x04,0x0e,0x18,0x01,0x77,0x39,0x05]
3906+
// GFX12: v_dot2_f32_bf16_e64_dpp v5, v1, v2, s3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x40,0x1a,0xcc,0xe9,0x04,0x0e,0x18,0x01,0x77,0x39,0x05]
3907+
3908+
v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
3909+
// GFX12: v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
3910+
39053911
v_minimum_f32 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
39063912
// GFX12: v_minimum_f32_e64_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0x65,0xd7,0xe9,0x04,0x02,0x00,0x01,0x77,0x39,0x05]
39073913

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

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,9 @@
44
# 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]
55
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 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]
8+
0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe
9+
710
# 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]
811
0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
912

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

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,9 @@
44
# 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]
55
0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05
66

7+
# GFX11: v_dot2_f32_bf16_e64_dpp v5, v1, v2, s3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x40,0x1a,0xcc,0xe9,0x04,0x0e,0x18,0x01,0x77,0x39,0x05]
8+
0x05,0x40,0x1a,0xcc,0xe9,0x04,0x0e,0x18,0x01,0x77,0x39,0x05
9+
710
# 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]
811
0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92
912

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

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,12 @@
77
# 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]
88
0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
99

10+
# GFX12: v_dot2_f32_bf16_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: [0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
11+
0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe
12+
13+
# GFX12: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff]
14+
0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
15+
1016
# 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]
1117
0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1
1218

‎llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,9 @@
44
# GFX12: 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]
55
0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05
66

7+
# GFX12: v_dot2_f32_bf16_e64_dpp v5, v1, v2, s3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x40,0x1a,0xcc,0xe9,0x04,0x0e,0x18,0x01,0x77,0x39,0x05]
8+
0x05,0x40,0x1a,0xcc,0xe9,0x04,0x0e,0x18,0x01,0x77,0x39,0x05
9+
710
# GFX12: 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]
811
0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92
912

0 commit comments

Comments
 (0)