Skip to content

[AMDGPU][True16][MC] test update for v_add/sub_f16 in true16 #118926

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Dec 9, 2024
Merged

[AMDGPU][True16][MC] test update for v_add/sub_f16 in true16 #118926

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 2 additions & 4 deletions llvm/lib/Target/AMDGPU/VOP2Instructions.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1818,10 +1818,8 @@ defm V_CVT_PK_RTZ_F16_F32 : VOP2_Real_FULL_with_name_gfx11_gfx12<0x02f,
"V_CVT_PKRTZ_F16_F32", "v_cvt_pk_rtz_f16_f32">;
defm V_PK_FMAC_F16 : VOP2_Real_e32_gfx11_gfx12<0x03c>;

defm V_ADD_F16_t16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x032, "v_add_f16">;
defm V_ADD_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x032, "v_add_f16">;
defm V_SUB_F16_t16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x033, "v_sub_f16">;
defm V_SUB_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x033, "v_sub_f16">;
defm V_ADD_F16 : VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12<0x032, "v_add_f16">;
defm V_SUB_F16 : VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12<0x033, "v_sub_f16">;
defm V_SUBREV_F16_t16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x034, "v_subrev_f16">;
defm V_SUBREV_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x034, "v_subrev_f16">;
defm V_MUL_F16_t16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x035, "v_mul_f16">;
Expand Down
150 changes: 90 additions & 60 deletions llvm/test/MC/AMDGPU/gfx11_asm_vop2.s
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,50 +124,65 @@ v_add_co_ci_u32 v255, vcc, 0xaf123456, v255, vcc
// W64: v_add_co_ci_u32_e32 v255, vcc, 0xaf123456, v255, vcc ; encoding: [0xff,0xfe,0xff,0x41,0x56,0x34,0x12,0xaf]
// W32-ERR: :[[@LINE-2]]:1: error: operands are not valid for this GPU or mode

v_add_f16 v5, v1, v2
// GFX11: v_add_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x64]
v_add_f16 v5.l, v1.l, v2.l
// GFX11: v_add_f16_e32 v5.l, v1.l, v2.l ; encoding: [0x01,0x05,0x0a,0x64]

v_add_f16 v5, v127, v2
// GFX11: v_add_f16_e32 v5, v127, v2 ; encoding: [0x7f,0x05,0x0a,0x64]
v_add_f16 v5.l, v127.l, v2.l
// GFX11: v_add_f16_e32 v5.l, v127.l, v2.l ; encoding: [0x7f,0x05,0x0a,0x64]

v_add_f16 v5, s1, v2
// GFX11: v_add_f16_e32 v5, s1, v2 ; encoding: [0x01,0x04,0x0a,0x64]
v_add_f16 v5.l, s1, v2.l
// GFX11: v_add_f16_e32 v5.l, s1, v2.l ; encoding: [0x01,0x04,0x0a,0x64]

v_add_f16 v5, s105, v2
// GFX11: v_add_f16_e32 v5, s105, v2 ; encoding: [0x69,0x04,0x0a,0x64]
v_add_f16 v5.l, s105, v2.l
// GFX11: v_add_f16_e32 v5.l, s105, v2.l ; encoding: [0x69,0x04,0x0a,0x64]

v_add_f16 v5, vcc_lo, v2
// GFX11: v_add_f16_e32 v5, vcc_lo, v2 ; encoding: [0x6a,0x04,0x0a,0x64]
v_add_f16 v5.l, vcc_lo, v2.l
// GFX11: v_add_f16_e32 v5.l, vcc_lo, v2.l ; encoding: [0x6a,0x04,0x0a,0x64]

v_add_f16 v5, vcc_hi, v2
// GFX11: v_add_f16_e32 v5, vcc_hi, v2 ; encoding: [0x6b,0x04,0x0a,0x64]
v_add_f16 v5.l, vcc_hi, v2.l
// GFX11: v_add_f16_e32 v5.l, vcc_hi, v2.l ; encoding: [0x6b,0x04,0x0a,0x64]

v_add_f16 v5, ttmp15, v2
// GFX11: v_add_f16_e32 v5, ttmp15, v2 ; encoding: [0x7b,0x04,0x0a,0x64]
v_add_f16 v5.l, ttmp15, v2.l
// GFX11: v_add_f16_e32 v5.l, ttmp15, v2.l ; encoding: [0x7b,0x04,0x0a,0x64]

v_add_f16 v5, m0, v2
// GFX11: v_add_f16_e32 v5, m0, v2 ; encoding: [0x7d,0x04,0x0a,0x64]
v_add_f16 v5.l, m0, v2.l
// GFX11: v_add_f16_e32 v5.l, m0, v2.l ; encoding: [0x7d,0x04,0x0a,0x64]

v_add_f16 v5, exec_lo, v2
// GFX11: v_add_f16_e32 v5, exec_lo, v2 ; encoding: [0x7e,0x04,0x0a,0x64]
v_add_f16 v5.l, exec_lo, v2.l
// GFX11: v_add_f16_e32 v5.l, exec_lo, v2.l ; encoding: [0x7e,0x04,0x0a,0x64]

v_add_f16 v5, exec_hi, v2
// GFX11: v_add_f16_e32 v5, exec_hi, v2 ; encoding: [0x7f,0x04,0x0a,0x64]
v_add_f16 v5.l, exec_hi, v2.l
// GFX11: v_add_f16_e32 v5.l, exec_hi, v2.l ; encoding: [0x7f,0x04,0x0a,0x64]

v_add_f16 v5, null, v2
// GFX11: v_add_f16_e32 v5, null, v2 ; encoding: [0x7c,0x04,0x0a,0x64]
v_add_f16 v5.l, null, v2.l
// GFX11: v_add_f16_e32 v5.l, null, v2.l ; encoding: [0x7c,0x04,0x0a,0x64]

v_add_f16 v5, -1, v2
// GFX11: v_add_f16_e32 v5, -1, v2 ; encoding: [0xc1,0x04,0x0a,0x64]
v_add_f16 v5.l, -1, v2.l
// GFX11: v_add_f16_e32 v5.l, -1, v2.l ; encoding: [0xc1,0x04,0x0a,0x64]

v_add_f16 v5, 0.5, v2
// GFX11: v_add_f16_e32 v5, 0.5, v2 ; encoding: [0xf0,0x04,0x0a,0x64]
v_add_f16 v5.l, 0.5, v2.l
// GFX11: v_add_f16_e32 v5.l, 0.5, v2.l ; encoding: [0xf0,0x04,0x0a,0x64]

v_add_f16 v5, src_scc, v2
// GFX11: v_add_f16_e32 v5, src_scc, v2 ; encoding: [0xfd,0x04,0x0a,0x64]
v_add_f16 v5.l, src_scc, v2.l
// GFX11: v_add_f16_e32 v5.l, src_scc, v2.l ; encoding: [0xfd,0x04,0x0a,0x64]

v_add_f16 v127, 0xfe0b, v127
// GFX11: v_add_f16_e32 v127, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x64,0x0b,0xfe,0x00,0x00]
v_add_f16 v127.l, 0xfe0b, v127.l
// GFX11: v_add_f16_e32 v127.l, 0xfe0b, v127.l ; encoding: [0xff,0xfe,0xfe,0x64,0x0b,0xfe,0x00,0x00]

v_add_f16 v5.l, v1.h, v2.l
// GFX11: v_add_f16_e32 v5.l, v1.h, v2.l ; encoding: [0x81,0x05,0x0a,0x64]

v_add_f16 v5.l, v127.h, v2.l
// GFX11: v_add_f16_e32 v5.l, v127.h, v2.l ; encoding: [0xff,0x05,0x0a,0x64]

v_add_f16 v127.l, 0.5, v127.l
// GFX11: v_add_f16_e32 v127.l, 0.5, v127.l ; encoding: [0xf0,0xfe,0xfe,0x64]

v_add_f16 v5.h, src_scc, v2.h
// GFX11: v_add_f16_e32 v5.h, src_scc, v2.h ; encoding: [0xfd,0x04,0x0b,0x65]

v_add_f16 v127.h, 0xfe0b, v127.h
// GFX11: v_add_f16_e32 v127.h, 0xfe0b, v127.h ; encoding: [0xff,0xfe,0xff,0x65,0x0b,0xfe,0x00,0x00]

v_add_f32 v5, v1, v2
// GFX11: v_add_f32_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x06]
Expand Down Expand Up @@ -2074,50 +2089,65 @@ v_sub_co_ci_u32 v255, vcc, 0xaf123456, v255, vcc
// W64: v_sub_co_ci_u32_e32 v255, vcc, 0xaf123456, v255, vcc ; encoding: [0xff,0xfe,0xff,0x43,0x56,0x34,0x12,0xaf]
// W32-ERR: :[[@LINE-2]]:1: error: operands are not valid for this GPU or mode

v_sub_f16 v5, v1, v2
// GFX11: v_sub_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x66]
v_sub_f16 v5.l, v1.l, v2.l
// GFX11: v_sub_f16_e32 v5.l, v1.l, v2.l ; encoding: [0x01,0x05,0x0a,0x66]

v_sub_f16 v5.l, v127.l, v2.l
// GFX11: v_sub_f16_e32 v5.l, v127.l, v2.l ; encoding: [0x7f,0x05,0x0a,0x66]

v_sub_f16 v5.l, s1, v2.l
// GFX11: v_sub_f16_e32 v5.l, s1, v2.l ; encoding: [0x01,0x04,0x0a,0x66]

v_sub_f16 v5.l, s105, v2.l
// GFX11: v_sub_f16_e32 v5.l, s105, v2.l ; encoding: [0x69,0x04,0x0a,0x66]

v_sub_f16 v5.l, vcc_lo, v2.l
// GFX11: v_sub_f16_e32 v5.l, vcc_lo, v2.l ; encoding: [0x6a,0x04,0x0a,0x66]

v_sub_f16 v5.l, vcc_hi, v2.l
// GFX11: v_sub_f16_e32 v5.l, vcc_hi, v2.l ; encoding: [0x6b,0x04,0x0a,0x66]

v_sub_f16 v5, v127, v2
// GFX11: v_sub_f16_e32 v5, v127, v2 ; encoding: [0x7f,0x05,0x0a,0x66]
v_sub_f16 v5.l, ttmp15, v2.l
// GFX11: v_sub_f16_e32 v5.l, ttmp15, v2.l ; encoding: [0x7b,0x04,0x0a,0x66]

v_sub_f16 v5, s1, v2
// GFX11: v_sub_f16_e32 v5, s1, v2 ; encoding: [0x01,0x04,0x0a,0x66]
v_sub_f16 v5.l, m0, v2.l
// GFX11: v_sub_f16_e32 v5.l, m0, v2.l ; encoding: [0x7d,0x04,0x0a,0x66]

v_sub_f16 v5, s105, v2
// GFX11: v_sub_f16_e32 v5, s105, v2 ; encoding: [0x69,0x04,0x0a,0x66]
v_sub_f16 v5.l, exec_lo, v2.l
// GFX11: v_sub_f16_e32 v5.l, exec_lo, v2.l ; encoding: [0x7e,0x04,0x0a,0x66]

v_sub_f16 v5, vcc_lo, v2
// GFX11: v_sub_f16_e32 v5, vcc_lo, v2 ; encoding: [0x6a,0x04,0x0a,0x66]
v_sub_f16 v5.l, exec_hi, v2.l
// GFX11: v_sub_f16_e32 v5.l, exec_hi, v2.l ; encoding: [0x7f,0x04,0x0a,0x66]

v_sub_f16 v5, vcc_hi, v2
// GFX11: v_sub_f16_e32 v5, vcc_hi, v2 ; encoding: [0x6b,0x04,0x0a,0x66]
v_sub_f16 v5.l, null, v2.l
// GFX11: v_sub_f16_e32 v5.l, null, v2.l ; encoding: [0x7c,0x04,0x0a,0x66]

v_sub_f16 v5, ttmp15, v2
// GFX11: v_sub_f16_e32 v5, ttmp15, v2 ; encoding: [0x7b,0x04,0x0a,0x66]
v_sub_f16 v5.l, -1, v2.l
// GFX11: v_sub_f16_e32 v5.l, -1, v2.l ; encoding: [0xc1,0x04,0x0a,0x66]

v_sub_f16 v5, m0, v2
// GFX11: v_sub_f16_e32 v5, m0, v2 ; encoding: [0x7d,0x04,0x0a,0x66]
v_sub_f16 v5.l, 0.5, v2.l
// GFX11: v_sub_f16_e32 v5.l, 0.5, v2.l ; encoding: [0xf0,0x04,0x0a,0x66]

v_sub_f16 v5, exec_lo, v2
// GFX11: v_sub_f16_e32 v5, exec_lo, v2 ; encoding: [0x7e,0x04,0x0a,0x66]
v_sub_f16 v5.l, src_scc, v2.l
// GFX11: v_sub_f16_e32 v5.l, src_scc, v2.l ; encoding: [0xfd,0x04,0x0a,0x66]

v_sub_f16 v5, exec_hi, v2
// GFX11: v_sub_f16_e32 v5, exec_hi, v2 ; encoding: [0x7f,0x04,0x0a,0x66]
v_sub_f16 v127.l, 0xfe0b, v127.l
// GFX11: v_sub_f16_e32 v127.l, 0xfe0b, v127.l ; encoding: [0xff,0xfe,0xfe,0x66,0x0b,0xfe,0x00,0x00]

v_sub_f16 v5, null, v2
// GFX11: v_sub_f16_e32 v5, null, v2 ; encoding: [0x7c,0x04,0x0a,0x66]
v_sub_f16 v5.l, v1.h, v2.l
// GFX11: v_sub_f16_e32 v5.l, v1.h, v2.l ; encoding: [0x81,0x05,0x0a,0x66]

v_sub_f16 v5, -1, v2
// GFX11: v_sub_f16_e32 v5, -1, v2 ; encoding: [0xc1,0x04,0x0a,0x66]
v_sub_f16 v5.l, v127.h, v2.l
// GFX11: v_sub_f16_e32 v5.l, v127.h, v2.l ; encoding: [0xff,0x05,0x0a,0x66]

v_sub_f16 v5, 0.5, v2
// GFX11: v_sub_f16_e32 v5, 0.5, v2 ; encoding: [0xf0,0x04,0x0a,0x66]
v_sub_f16 v127.l, 0.5, v127.l
// GFX11: v_sub_f16_e32 v127.l, 0.5, v127.l ; encoding: [0xf0,0xfe,0xfe,0x66]

v_sub_f16 v5, src_scc, v2
// GFX11: v_sub_f16_e32 v5, src_scc, v2 ; encoding: [0xfd,0x04,0x0a,0x66]
v_sub_f16 v5.h, src_scc, v2.h
// GFX11: v_sub_f16_e32 v5.h, src_scc, v2.h ; encoding: [0xfd,0x04,0x0b,0x67]

v_sub_f16 v127, 0xfe0b, v127
// GFX11: v_sub_f16_e32 v127, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x66,0x0b,0xfe,0x00,0x00]
v_sub_f16 v127.h, 0xfe0b, v127.h
// GFX11: v_sub_f16_e32 v127.h, 0xfe0b, v127.h ; encoding: [0xff,0xfe,0xff,0x67,0x0b,0xfe,0x00,0x00]

v_sub_f32 v5, v1, v2
// GFX11: v_sub_f32_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x08]
Expand Down
Loading
Loading