Skip to content

Commit b71a6fd

Browse files
broxigarchenbroxigarchen
authored
[AMDGPU][True16][MC] true16 for v_cvt_i32_i16 (#120645)
Support true16 format for v_cvt_i32_i16 in MC
1 parent dc307be commit b71a6fd

28 files changed

+492
-235
lines changed

‎llvm/lib/Target/AMDGPU/VOP1Instructions.td

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1019,7 +1019,7 @@ defm V_SWAP_B16 : VOP1Only_Real_gfx11_gfx12<0x066>;
10191019
defm V_PERMLANE64_B32 : VOP1Only_Real_gfx11_gfx12<0x067>;
10201020
defm V_MOV_B16_t16 : VOP1_Real_FULL_t16_gfx11_gfx12<0x01c, "v_mov_b16">;
10211021
defm V_NOT_B16 : VOP1_Real_FULL_t16_and_fake16_gfx11_gfx12<0x069, "v_not_b16">;
1022-
defm V_CVT_I32_I16_fake16 : VOP1_Real_FULL_t16_gfx11_gfx12<0x06a, "v_cvt_i32_i16">;
1022+
defm V_CVT_I32_I16 : VOP1_Real_FULL_t16_and_fake16_gfx11_gfx12<0x06a, "v_cvt_i32_i16">;
10231023
defm V_CVT_U32_U16_fake16 : VOP1_Real_FULL_t16_gfx11_gfx12<0x06b, "v_cvt_u32_u16">;
10241024

10251025
defm V_CVT_F16_U16 : VOP1_Real_FULL_t16_and_fake16_gfx11_gfx12<0x050, "v_cvt_f16_u16">;

‎llvm/test/MC/AMDGPU/gfx11_asm_vop1.s

Lines changed: 10 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1271,11 +1271,11 @@ v_cvt_i32_f64 v5, src_scc
12711271
v_cvt_i32_f64 v255, 0xaf123456
12721272
// GFX11: v_cvt_i32_f64_e32 v255, 0xaf123456 ; encoding: [0xff,0x06,0xfe,0x7f,0x56,0x34,0x12,0xaf]
12731273

1274-
v_cvt_i32_i16 v5, v1
1275-
// GFX11: v_cvt_i32_i16_e32 v5, v1 ; encoding: [0x01,0xd5,0x0a,0x7e]
1274+
v_cvt_i32_i16 v5, v1.l
1275+
// GFX11: v_cvt_i32_i16_e32 v5, v1.l ; encoding: [0x01,0xd5,0x0a,0x7e]
12761276

1277-
v_cvt_i32_i16 v5, v127
1278-
// GFX11: v_cvt_i32_i16_e32 v5, v127 ; encoding: [0x7f,0xd5,0x0a,0x7e]
1277+
v_cvt_i32_i16 v5, v127.l
1278+
// GFX11: v_cvt_i32_i16_e32 v5, v127.l ; encoding: [0x7f,0xd5,0x0a,0x7e]
12791279

12801280
v_cvt_i32_i16 v5, s1
12811281
// GFX11: v_cvt_i32_i16_e32 v5, s1 ; encoding: [0x01,0xd4,0x0a,0x7e]
@@ -1316,6 +1316,12 @@ v_cvt_i32_i16 v5, src_scc
13161316
v_cvt_i32_i16 v255, 0xfe0b
13171317
// GFX11: v_cvt_i32_i16_e32 v255, 0xfe0b ; encoding: [0xff,0xd4,0xfe,0x7f,0x0b,0xfe,0x00,0x00]
13181318

1319+
v_cvt_i32_i16 v5, v1.h
1320+
// GFX11: v_cvt_i32_i16_e32 v5, v1.h ; encoding: [0x81,0xd5,0x0a,0x7e]
1321+
1322+
v_cvt_i32_i16 v5, v127.h
1323+
// GFX11: v_cvt_i32_i16_e32 v5, v127.h ; encoding: [0xff,0xd5,0x0a,0x7e]
1324+
13191325
v_cvt_nearest_i32_f32 v5, v1
13201326
// GFX11: v_cvt_nearest_i32_f32_e32 v5, v1 ; encoding: [0x01,0x19,0x0a,0x7e]
13211327

‎llvm/test/MC/AMDGPU/gfx11_asm_vop1_dpp16.s

Lines changed: 37 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -926,47 +926,56 @@ v_cvt_i32_f32 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
926926
v_cvt_i32_f32 v255, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
927927
// GFX11: v_cvt_i32_f32_dpp v255, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x10,0xfe,0x7f,0xff,0x6f,0x35,0x30]
928928

929-
v_cvt_i32_i16 v5, v1 quad_perm:[3,2,1,0]
930-
// GFX11: v_cvt_i32_i16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x1b,0x00,0xff]
929+
v_cvt_i32_i16 v5, v1.l quad_perm:[3,2,1,0]
930+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x1b,0x00,0xff]
931931

932-
v_cvt_i32_i16 v5, v1 quad_perm:[0,1,2,3]
933-
// GFX11: v_cvt_i32_i16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0xe4,0x00,0xff]
932+
v_cvt_i32_i16 v5, v1.l quad_perm:[0,1,2,3]
933+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0xe4,0x00,0xff]
934934

935-
v_cvt_i32_i16 v5, v1 row_mirror
936-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x40,0x01,0xff]
935+
v_cvt_i32_i16 v5, v1.l row_mirror
936+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x40,0x01,0xff]
937937

938-
v_cvt_i32_i16 v5, v1 row_half_mirror
939-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x41,0x01,0xff]
938+
v_cvt_i32_i16 v5, v1.l row_half_mirror
939+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x41,0x01,0xff]
940940

941-
v_cvt_i32_i16 v5, v1 row_shl:1
942-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x01,0x01,0xff]
941+
v_cvt_i32_i16 v5, v1.l row_shl:1
942+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x01,0x01,0xff]
943943

944-
v_cvt_i32_i16 v5, v1 row_shl:15
945-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x0f,0x01,0xff]
944+
v_cvt_i32_i16 v5, v1.l row_shl:15
945+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x0f,0x01,0xff]
946946

947-
v_cvt_i32_i16 v5, v1 row_shr:1
948-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x11,0x01,0xff]
947+
v_cvt_i32_i16 v5, v1.l row_shr:1
948+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x11,0x01,0xff]
949949

950-
v_cvt_i32_i16 v5, v1 row_shr:15
951-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x1f,0x01,0xff]
950+
v_cvt_i32_i16 v5, v1.l row_shr:15
951+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x1f,0x01,0xff]
952952

953-
v_cvt_i32_i16 v5, v1 row_ror:1
954-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x21,0x01,0xff]
953+
v_cvt_i32_i16 v5, v1.l row_ror:1
954+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x21,0x01,0xff]
955955

956-
v_cvt_i32_i16 v5, v1 row_ror:15
957-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x2f,0x01,0xff]
956+
v_cvt_i32_i16 v5, v1.l row_ror:15
957+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x2f,0x01,0xff]
958958

959-
v_cvt_i32_i16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
960-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x50,0x01,0xff]
959+
v_cvt_i32_i16 v5, v1.l row_share:0 row_mask:0xf bank_mask:0xf
960+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x50,0x01,0xff]
961961

962-
v_cvt_i32_i16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
963-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x5f,0x01,0x01]
962+
v_cvt_i32_i16 v5, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
963+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x5f,0x01,0x01]
964964

965-
v_cvt_i32_i16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
966-
// GFX11: v_cvt_i32_i16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x60,0x09,0x13]
965+
v_cvt_i32_i16 v5, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1
966+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xd4,0x0a,0x7e,0x01,0x60,0x09,0x13]
967967

968-
v_cvt_i32_i16 v255, v127 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
969-
// GFX11: v_cvt_i32_i16_dpp v255, v127 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xd4,0xfe,0x7f,0x7f,0x6f,0x05,0x30]
968+
v_cvt_i32_i16 v255, v127.l row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1
969+
// GFX11: v_cvt_i32_i16_dpp v255, v127.l row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xd4,0xfe,0x7f,0x7f,0x6f,0x05,0x30]
970+
971+
v_cvt_i32_i16 v5, v127.l row_share:15 row_mask:0x0 bank_mask:0x1
972+
// GFX11: v_cvt_i32_i16_dpp v5, v127.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xd4,0x0a,0x7e,0x7f,0x5f,0x01,0x01]
973+
974+
v_cvt_i32_i16 v5, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
975+
// GFX11: v_cvt_i32_i16_dpp v5, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xd4,0x0a,0x7e,0x81,0x60,0x09,0x13]
976+
977+
v_cvt_i32_i16 v255, v127.h row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
978+
// GFX11: v_cvt_i32_i16_dpp v255, v127.h row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xd4,0xfe,0x7f,0xff,0x6f,0x05,0x30]
970979

971980
v_cvt_nearest_i32_f32 v5, v1 quad_perm:[3,2,1,0]
972981
// GFX11: v_cvt_nearest_i32_f32_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x18,0x0a,0x7e,0x01,0x1b,0x00,0xff]

‎llvm/test/MC/AMDGPU/gfx11_asm_vop1_dpp8.s

Lines changed: 15 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -218,14 +218,23 @@ v_cvt_i32_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
218218
v_cvt_i32_f32 v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
219219
// GFX11: v_cvt_i32_f32_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0x10,0xfe,0x7f,0xff,0x00,0x00,0x00]
220220

221-
v_cvt_i32_i16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
222-
// GFX11: v_cvt_i32_i16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xd4,0x0a,0x7e,0x01,0x77,0x39,0x05]
221+
v_cvt_i32_i16 v5, v1.l dpp8:[7,6,5,4,3,2,1,0]
222+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xd4,0x0a,0x7e,0x01,0x77,0x39,0x05]
223223

224-
v_cvt_i32_i16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
225-
// GFX11: v_cvt_i32_i16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xd4,0x0a,0x7e,0x01,0x77,0x39,0x05]
224+
v_cvt_i32_i16 v5, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1
225+
// GFX11: v_cvt_i32_i16_dpp v5, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xd4,0x0a,0x7e,0x01,0x77,0x39,0x05]
226226

227-
v_cvt_i32_i16 v255, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
228-
// GFX11: v_cvt_i32_i16_dpp v255, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xd4,0xfe,0x7f,0x7f,0x00,0x00,0x00]
227+
v_cvt_i32_i16 v255, v127.l dpp8:[0,0,0,0,0,0,0,0]
228+
// GFX11: v_cvt_i32_i16_dpp v255, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xd4,0xfe,0x7f,0x7f,0x00,0x00,0x00]
229+
230+
v_cvt_i32_i16 v5, v127.l dpp8:[7,6,5,4,3,2,1,0]
231+
// GFX11: v_cvt_i32_i16_dpp v5, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xd4,0x0a,0x7e,0x7f,0x77,0x39,0x05]
232+
233+
v_cvt_i32_i16 v5, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
234+
// GFX11: v_cvt_i32_i16_dpp v5, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xd4,0x0a,0x7e,0x81,0x77,0x39,0x05]
235+
236+
v_cvt_i32_i16 v255, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
237+
// GFX11: v_cvt_i32_i16_dpp v255, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xd4,0xfe,0x7f,0xff,0x00,0x00,0x00]
229238

230239
v_cvt_nearest_i32_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
231240
// GFX11: v_cvt_nearest_i32_f32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x18,0x0a,0x7e,0x01,0x77,0x39,0x05]

‎llvm/test/MC/AMDGPU/gfx11_asm_vop1_t16_err.s

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -236,6 +236,24 @@ v_cvt_i32_i16_e32 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
236236
v_cvt_i32_i16_e32 v5, v199 quad_perm:[3,2,1,0]
237237
// GFX11: :[[@LINE-1]]:23: error: invalid operand for instruction
238238

239+
v_cvt_i32_i16_e32 v5.h, v199.h
240+
// GFX11: :[[@LINE-1]]:19: error: invalid operand for instruction
241+
242+
v_cvt_i32_i16_e32 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
243+
// GFX11: :[[@LINE-1]]:19: error: invalid operand for instruction
244+
245+
v_cvt_i32_i16_e32 v5.h, v199.h quad_perm:[3,2,1,0]
246+
// GFX11: :[[@LINE-1]]:19: error: invalid operand for instruction
247+
248+
v_cvt_i32_i16_e32 v5.l, v199.l
249+
// GFX11: :[[@LINE-1]]:19: error: invalid operand for instruction
250+
251+
v_cvt_i32_i16_e32 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
252+
// GFX11: :[[@LINE-1]]:19: error: invalid operand for instruction
253+
254+
v_cvt_i32_i16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
255+
// GFX11: :[[@LINE-1]]:19: error: invalid operand for instruction
256+
239257
v_cvt_norm_i16_f16_e32 v128.h, 0xfe0b
240258
// GFX11: :[[@LINE-1]]:24: error: invalid operand for instruction
241259

‎llvm/test/MC/AMDGPU/gfx11_asm_vop1_t16_promote.s

Lines changed: 15 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -662,14 +662,23 @@ v_cvt_i16_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
662662
v_cvt_i16_f16 v5.l, v199.l quad_perm:[3,2,1,0]
663663
// GFX11: v_cvt_i16_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd3,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
664664

665-
v_cvt_i32_i16 v5, v199
666-
// GFX11: v_cvt_i32_i16_e64 v5, v199 ; encoding: [0x05,0x00,0xea,0xd5,0xc7,0x01,0x00,0x00]
665+
v_cvt_i32_i16 v5, v199.h
666+
// GFX11: v_cvt_i32_i16_e64 v5, v199.h op_sel:[1,0] ; encoding: [0x05,0x08,0xea,0xd5,0xc7,0x01,0x00,0x00]
667667

668-
v_cvt_i32_i16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
669-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xea,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
668+
v_cvt_i32_i16 v5, v199.h dpp8:[7,6,5,4,3,2,1,0]
669+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v199.h op_sel:[1,0] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x08,0xea,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
670670

671-
v_cvt_i32_i16 v5, v199 quad_perm:[3,2,1,0]
672-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
671+
v_cvt_i32_i16 v5, v199.h quad_perm:[3,2,1,0]
672+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v199.h op_sel:[1,0] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x08,0xea,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
673+
674+
v_cvt_i32_i16 v5, v199.l
675+
// GFX11: v_cvt_i32_i16_e64 v5, v199.l ; encoding: [0x05,0x00,0xea,0xd5,0xc7,0x01,0x00,0x00]
676+
677+
v_cvt_i32_i16 v5, v199.l dpp8:[7,6,5,4,3,2,1,0]
678+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xea,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
679+
680+
v_cvt_i32_i16 v5, v199.l quad_perm:[3,2,1,0]
681+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
673682

674683
v_cvt_norm_i16_f16 v128.h, 0xfe0b
675684
// GFX11: v_cvt_norm_i16_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xe3,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]

‎llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp16_from_vop1.s

Lines changed: 31 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -967,47 +967,50 @@ v_cvt_i32_f32_e64_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1
967967
v_cvt_i32_f32_e64_dpp v255, -|v255| clamp row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
968968
// GFX11: v_cvt_i32_f32_e64_dpp v255, -|v255| clamp row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0x88,0xd5,0xfa,0x00,0x00,0x20,0xff,0x6f,0x05,0x30]
969969

970-
v_cvt_i32_i16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
971-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
970+
v_cvt_i32_i16_e64_dpp v5, v1.l quad_perm:[3,2,1,0]
971+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
972972

973-
v_cvt_i32_i16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
974-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
973+
v_cvt_i32_i16_e64_dpp v5, v1.l quad_perm:[0,1,2,3]
974+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
975975

976-
v_cvt_i32_i16_e64_dpp v5, v1 row_mirror
977-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
976+
v_cvt_i32_i16_e64_dpp v5, v1.l row_mirror
977+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
978978

979-
v_cvt_i32_i16_e64_dpp v5, v1 row_half_mirror
980-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
979+
v_cvt_i32_i16_e64_dpp v5, v1.l row_half_mirror
980+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
981981

982-
v_cvt_i32_i16_e64_dpp v5, v1 row_shl:1
983-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
982+
v_cvt_i32_i16_e64_dpp v5, v1.l row_shl:1
983+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
984984

985-
v_cvt_i32_i16_e64_dpp v5, v1 row_shl:15
986-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
985+
v_cvt_i32_i16_e64_dpp v5, v1.l row_shl:15
986+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
987987

988-
v_cvt_i32_i16_e64_dpp v5, v1 row_shr:1
989-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
988+
v_cvt_i32_i16_e64_dpp v5, v1.l row_shr:1
989+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
990990

991-
v_cvt_i32_i16_e64_dpp v5, v1 row_shr:15
992-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
991+
v_cvt_i32_i16_e64_dpp v5, v1.l row_shr:15
992+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
993993

994-
v_cvt_i32_i16_e64_dpp v5, v1 row_ror:1
995-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
994+
v_cvt_i32_i16_e64_dpp v5, v1.l row_ror:1
995+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
996996

997-
v_cvt_i32_i16_e64_dpp v5, v1 row_ror:15
998-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
997+
v_cvt_i32_i16_e64_dpp v5, v1.l row_ror:15
998+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
999999

1000-
v_cvt_i32_i16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
1001-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
1000+
v_cvt_i32_i16_e64_dpp v5, v1.l row_share:0 row_mask:0xf bank_mask:0xf
1001+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
10021002

1003-
v_cvt_i32_i16_e64_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
1004-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x5f,0x01,0x01]
1003+
v_cvt_i32_i16_e64_dpp v5, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
1004+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x5f,0x01,0x01]
10051005

1006-
v_cvt_i32_i16_e64_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
1007-
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x60,0x09,0x13]
1006+
v_cvt_i32_i16_e64_dpp v5, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
1007+
// GFX11: v_cvt_i32_i16_e64_dpp v5, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0x01,0x60,0x09,0x13]
10081008

1009-
v_cvt_i32_i16_e64_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
1010-
// GFX11: v_cvt_i32_i16_e64_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0xff,0x6f,0x05,0x30]
1009+
v_cvt_i32_i16_e64_dpp v255, v255.l row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
1010+
// GFX11: v_cvt_i32_i16_e64_dpp v255, v255.l row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x00,0xea,0xd5,0xfa,0x00,0x00,0x00,0xff,0x6f,0x05,0x30]
1011+
1012+
v_cvt_i32_i16_e64_dpp v255, v255.h row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
1013+
// GFX11: [0xff,0x08,0xea,0xd5,0xfa,0x00,0x00,0x00,0xff,0x6f,0x05,0x30]
10111014

10121015
v_cvt_nearest_i32_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
10131016
// GFX11: v_cvt_nearest_i32_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x8c,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]

0 commit comments

Comments
 (0)