Skip to content

Commit

Permalink
[AMDGPU][True16][MC] test update for v_max_f16/v_min_f16 in true16 (l…
Browse files Browse the repository at this point in the history 
…lvm#119291)

This is a NFC change. Update mc test for v_max/min_f16 in true16 format.

MC source change was done by previous patch and automatically enabled by
t16 pesudo
  • Loading branch information
@broxigarchen
broxigarchen authored Dec 17, 2024
1 parent 5287299 commit 8bbbcad
Show file tree
Hide file tree
Showing 9 changed files with 574 additions and 336 deletions.
6 changes: 2 additions & 4 deletions llvm/lib/Target/AMDGPU/VOP2Instructions.td
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1827,10 +1827,8 @@ defm V_MUL_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x035, "v_mul_f16">;
defm V_FMAC_F16 : VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12<0x036, "v_fmac_f16">;
defm V_LDEXP_F16_t16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x03b, "v_ldexp_f16">;
defm V_LDEXP_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x03b, "v_ldexp_f16">;
defm V_MAX_F16_t16 : VOP2_Real_FULL_t16_gfx11<0x039, "v_max_f16">;
defm V_MAX_F16_fake16 : VOP2_Real_FULL_t16_gfx11<0x039, "v_max_f16">;
defm V_MIN_F16_t16 : VOP2_Real_FULL_t16_gfx11<0x03a, "v_min_f16">;
defm V_MIN_F16_fake16 : VOP2_Real_FULL_t16_gfx11<0x03a, "v_min_f16">;
defm V_MAX_F16 : VOP2_Real_FULL_t16_and_fake16_gfx11<0x039, "v_max_f16">;
defm V_MIN_F16 : VOP2_Real_FULL_t16_and_fake16_gfx11<0x03a, "v_min_f16">;
defm V_FMAMK_F16 : VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<0x037, "v_fmamk_f16">;
defm V_FMAAK_F16 : VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<0x038, "v_fmaak_f16">;

Expand Down
150 changes: 90 additions & 60 deletions llvm/test/MC/AMDGPU/gfx11_asm_vop2.s
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1177,50 +1177,65 @@ v_lshrrev_b32 v5, src_scc, v2
v_lshrrev_b32 v255, 0xaf123456, v255
// GFX11: v_lshrrev_b32_e32 v255, 0xaf123456, v255 ; encoding: [0xff,0xfe,0xff,0x33,0x56,0x34,0x12,0xaf]

v_max_f16 v5, v1, v2
// GFX11: v_max_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x72]
v_max_f16 v5.l, v1.l, v2.l
// GFX11: v_max_f16_e32 v5.l, v1.l, v2.l ; encoding: [0x01,0x05,0x0a,0x72]

v_max_f16 v5, v127, v2
// GFX11: v_max_f16_e32 v5, v127, v2 ; encoding: [0x7f,0x05,0x0a,0x72]
v_max_f16 v5.l, v127.l, v2.l
// GFX11: v_max_f16_e32 v5.l, v127.l, v2.l ; encoding: [0x7f,0x05,0x0a,0x72]

v_max_f16 v5, s1, v2
// GFX11: v_max_f16_e32 v5, s1, v2 ; encoding: [0x01,0x04,0x0a,0x72]
v_max_f16 v5.l, s1, v2.l
// GFX11: v_max_f16_e32 v5.l, s1, v2.l ; encoding: [0x01,0x04,0x0a,0x72]

v_max_f16 v5, s105, v2
// GFX11: v_max_f16_e32 v5, s105, v2 ; encoding: [0x69,0x04,0x0a,0x72]
v_max_f16 v5.l, s105, v2.l
// GFX11: v_max_f16_e32 v5.l, s105, v2.l ; encoding: [0x69,0x04,0x0a,0x72]

v_max_f16 v5, vcc_lo, v2
// GFX11: v_max_f16_e32 v5, vcc_lo, v2 ; encoding: [0x6a,0x04,0x0a,0x72]
v_max_f16 v5.l, vcc_lo, v2.l
// GFX11: v_max_f16_e32 v5.l, vcc_lo, v2.l ; encoding: [0x6a,0x04,0x0a,0x72]

v_max_f16 v5, vcc_hi, v2
// GFX11: v_max_f16_e32 v5, vcc_hi, v2 ; encoding: [0x6b,0x04,0x0a,0x72]
v_max_f16 v5.l, vcc_hi, v2.l
// GFX11: v_max_f16_e32 v5.l, vcc_hi, v2.l ; encoding: [0x6b,0x04,0x0a,0x72]

v_max_f16 v5, ttmp15, v2
// GFX11: v_max_f16_e32 v5, ttmp15, v2 ; encoding: [0x7b,0x04,0x0a,0x72]
v_max_f16 v5.l, ttmp15, v2.l
// GFX11: v_max_f16_e32 v5.l, ttmp15, v2.l ; encoding: [0x7b,0x04,0x0a,0x72]

v_max_f16 v5, m0, v2
// GFX11: v_max_f16_e32 v5, m0, v2 ; encoding: [0x7d,0x04,0x0a,0x72]
v_max_f16 v5.l, m0, v2.l
// GFX11: v_max_f16_e32 v5.l, m0, v2.l ; encoding: [0x7d,0x04,0x0a,0x72]

v_max_f16 v5, exec_lo, v2
// GFX11: v_max_f16_e32 v5, exec_lo, v2 ; encoding: [0x7e,0x04,0x0a,0x72]
v_max_f16 v5.l, exec_lo, v2.l
// GFX11: v_max_f16_e32 v5.l, exec_lo, v2.l ; encoding: [0x7e,0x04,0x0a,0x72]

v_max_f16 v5, exec_hi, v2
// GFX11: v_max_f16_e32 v5, exec_hi, v2 ; encoding: [0x7f,0x04,0x0a,0x72]
v_max_f16 v5.l, exec_hi, v2.l
// GFX11: v_max_f16_e32 v5.l, exec_hi, v2.l ; encoding: [0x7f,0x04,0x0a,0x72]

v_max_f16 v5, null, v2
// GFX11: v_max_f16_e32 v5, null, v2 ; encoding: [0x7c,0x04,0x0a,0x72]
v_max_f16 v5.l, null, v2.l
// GFX11: v_max_f16_e32 v5.l, null, v2.l ; encoding: [0x7c,0x04,0x0a,0x72]

v_max_f16 v5, -1, v2
// GFX11: v_max_f16_e32 v5, -1, v2 ; encoding: [0xc1,0x04,0x0a,0x72]
v_max_f16 v5.l, -1, v2.l
// GFX11: v_max_f16_e32 v5.l, -1, v2.l ; encoding: [0xc1,0x04,0x0a,0x72]

v_max_f16 v5, 0.5, v2
// GFX11: v_max_f16_e32 v5, 0.5, v2 ; encoding: [0xf0,0x04,0x0a,0x72]
v_max_f16 v5.l, 0.5, v2.l
// GFX11: v_max_f16_e32 v5.l, 0.5, v2.l ; encoding: [0xf0,0x04,0x0a,0x72]

v_max_f16 v5, src_scc, v2
// GFX11: v_max_f16_e32 v5, src_scc, v2 ; encoding: [0xfd,0x04,0x0a,0x72]
v_max_f16 v5.l, src_scc, v2.l
// GFX11: v_max_f16_e32 v5.l, src_scc, v2.l ; encoding: [0xfd,0x04,0x0a,0x72]

v_max_f16 v127, 0xfe0b, v127
// GFX11: v_max_f16_e32 v127, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x72,0x0b,0xfe,0x00,0x00]
v_max_f16 v127.l, 0xfe0b, v127.l
// GFX11: v_max_f16_e32 v127.l, 0xfe0b, v127.l ; encoding: [0xff,0xfe,0xfe,0x72,0x0b,0xfe,0x00,0x00]

v_max_f16 v5.l, v1.h, v2.l
// GFX11: v_max_f16_e32 v5.l, v1.h, v2.l ; encoding: [0x81,0x05,0x0a,0x72]

v_max_f16 v5.l, v127.h, v2.l
// GFX11: v_max_f16_e32 v5.l, v127.h, v2.l ; encoding: [0xff,0x05,0x0a,0x72]

v_max_f16 v127.l, 0.5, v127.l
// GFX11: v_max_f16_e32 v127.l, 0.5, v127.l ; encoding: [0xf0,0xfe,0xfe,0x72]

v_max_f16 v5.h, src_scc, v2.h
// GFX11: v_max_f16_e32 v5.h, src_scc, v2.h ; encoding: [0xfd,0x04,0x0b,0x73]

v_max_f16 v127.h, 0xfe0b, v127.h
// GFX11: v_max_f16_e32 v127.h, 0xfe0b, v127.h ; encoding: [0xff,0xfe,0xff,0x73,0x0b,0xfe,0x00,0x00]

v_max_f32 v5, v1, v2
// GFX11: v_max_f32_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x20]
Expand Down Expand Up @@ -1357,50 +1372,65 @@ v_max_u32 v5, src_scc, v2
v_max_u32 v255, 0xaf123456, v255
// GFX11: v_max_u32_e32 v255, 0xaf123456, v255 ; encoding: [0xff,0xfe,0xff,0x29,0x56,0x34,0x12,0xaf]

v_min_f16 v5, v1, v2
// GFX11: v_min_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x74]
v_min_f16 v5.l, v1.l, v2.l
// GFX11: v_min_f16_e32 v5.l, v1.l, v2.l ; encoding: [0x01,0x05,0x0a,0x74]

v_min_f16 v5.l, v127.l, v2.l
// GFX11: v_min_f16_e32 v5.l, v127.l, v2.l ; encoding: [0x7f,0x05,0x0a,0x74]

v_min_f16 v5.l, s1, v2.l
// GFX11: v_min_f16_e32 v5.l, s1, v2.l ; encoding: [0x01,0x04,0x0a,0x74]

v_min_f16 v5.l, s105, v2.l
// GFX11: v_min_f16_e32 v5.l, s105, v2.l ; encoding: [0x69,0x04,0x0a,0x74]

v_min_f16 v5.l, vcc_lo, v2.l
// GFX11: v_min_f16_e32 v5.l, vcc_lo, v2.l ; encoding: [0x6a,0x04,0x0a,0x74]

v_min_f16 v5.l, vcc_hi, v2.l
// GFX11: v_min_f16_e32 v5.l, vcc_hi, v2.l ; encoding: [0x6b,0x04,0x0a,0x74]

v_min_f16 v5, v127, v2
// GFX11: v_min_f16_e32 v5, v127, v2 ; encoding: [0x7f,0x05,0x0a,0x74]
v_min_f16 v5.l, ttmp15, v2.l
// GFX11: v_min_f16_e32 v5.l, ttmp15, v2.l ; encoding: [0x7b,0x04,0x0a,0x74]

v_min_f16 v5, s1, v2
// GFX11: v_min_f16_e32 v5, s1, v2 ; encoding: [0x01,0x04,0x0a,0x74]
v_min_f16 v5.l, m0, v2.l
// GFX11: v_min_f16_e32 v5.l, m0, v2.l ; encoding: [0x7d,0x04,0x0a,0x74]

v_min_f16 v5, s105, v2
// GFX11: v_min_f16_e32 v5, s105, v2 ; encoding: [0x69,0x04,0x0a,0x74]
v_min_f16 v5.l, exec_lo, v2.l
// GFX11: v_min_f16_e32 v5.l, exec_lo, v2.l ; encoding: [0x7e,0x04,0x0a,0x74]

v_min_f16 v5, vcc_lo, v2
// GFX11: v_min_f16_e32 v5, vcc_lo, v2 ; encoding: [0x6a,0x04,0x0a,0x74]
v_min_f16 v5.l, exec_hi, v2.l
// GFX11: v_min_f16_e32 v5.l, exec_hi, v2.l ; encoding: [0x7f,0x04,0x0a,0x74]

v_min_f16 v5, vcc_hi, v2
// GFX11: v_min_f16_e32 v5, vcc_hi, v2 ; encoding: [0x6b,0x04,0x0a,0x74]
v_min_f16 v5.l, null, v2.l
// GFX11: v_min_f16_e32 v5.l, null, v2.l ; encoding: [0x7c,0x04,0x0a,0x74]

v_min_f16 v5, ttmp15, v2
// GFX11: v_min_f16_e32 v5, ttmp15, v2 ; encoding: [0x7b,0x04,0x0a,0x74]
v_min_f16 v5.l, -1, v2.l
// GFX11: v_min_f16_e32 v5.l, -1, v2.l ; encoding: [0xc1,0x04,0x0a,0x74]

v_min_f16 v5, m0, v2
// GFX11: v_min_f16_e32 v5, m0, v2 ; encoding: [0x7d,0x04,0x0a,0x74]
v_min_f16 v5.l, 0.5, v2.l
// GFX11: v_min_f16_e32 v5.l, 0.5, v2.l ; encoding: [0xf0,0x04,0x0a,0x74]

v_min_f16 v5, exec_lo, v2
// GFX11: v_min_f16_e32 v5, exec_lo, v2 ; encoding: [0x7e,0x04,0x0a,0x74]
v_min_f16 v5.l, src_scc, v2.l
// GFX11: v_min_f16_e32 v5.l, src_scc, v2.l ; encoding: [0xfd,0x04,0x0a,0x74]

v_min_f16 v5, exec_hi, v2
// GFX11: v_min_f16_e32 v5, exec_hi, v2 ; encoding: [0x7f,0x04,0x0a,0x74]
v_min_f16 v127.l, 0xfe0b, v127.l
// GFX11: v_min_f16_e32 v127.l, 0xfe0b, v127.l ; encoding: [0xff,0xfe,0xfe,0x74,0x0b,0xfe,0x00,0x00]

v_min_f16 v5, null, v2
// GFX11: v_min_f16_e32 v5, null, v2 ; encoding: [0x7c,0x04,0x0a,0x74]
v_min_f16 v5.l, v1.h, v2.l
// GFX11: v_min_f16_e32 v5.l, v1.h, v2.l ; encoding: [0x81,0x05,0x0a,0x74]

v_min_f16 v5, -1, v2
// GFX11: v_min_f16_e32 v5, -1, v2 ; encoding: [0xc1,0x04,0x0a,0x74]
v_min_f16 v5.l, v127.h, v2.l
// GFX11: v_min_f16_e32 v5.l, v127.h, v2.l ; encoding: [0xff,0x05,0x0a,0x74]

v_min_f16 v5, 0.5, v2
// GFX11: v_min_f16_e32 v5, 0.5, v2 ; encoding: [0xf0,0x04,0x0a,0x74]
v_min_f16 v127.l, 0.5, v127.l
// GFX11: v_min_f16_e32 v127.l, 0.5, v127.l ; encoding: [0xf0,0xfe,0xfe,0x74]

v_min_f16 v5, src_scc, v2
// GFX11: v_min_f16_e32 v5, src_scc, v2 ; encoding: [0xfd,0x04,0x0a,0x74]
v_min_f16 v5.h, src_scc, v2.h
// GFX11: v_min_f16_e32 v5.h, src_scc, v2.h ; encoding: [0xfd,0x04,0x0b,0x75]

v_min_f16 v127, 0xfe0b, v127
// GFX11: v_min_f16_e32 v127, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x74,0x0b,0xfe,0x00,0x00]
v_min_f16 v127.h, 0xfe0b, v127.h
// GFX11: v_min_f16_e32 v127.h, 0xfe0b, v127.h ; encoding: [0xff,0xfe,0xff,0x75,0x0b,0xfe,0x00,0x00]

v_min_f32 v5, v1, v2
// GFX11: v_min_f32_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x1e]
Expand Down
Loading

0 comments on commit 8bbbcad

Please sign in to comment.