Skip to content

[AMDGPU] Regenerate checks in the CodeGen/AMDGPU/fptrunc.f16.ll. NFC. #151575

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 2 commits into from
Jul 31, 2025
Merged

[AMDGPU] Regenerate checks in the CodeGen/AMDGPU/fptrunc.f16.ll. NFC. #151575

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8910788
[AMDGPU] Regenerate checks in the CodeGen/AMDGPU/fptrunc.f16.ll. NFC.
rampitec Jul 31, 2025
d6e4899
Merge branch 'main' into users/rampitec/07-31-_amdgpu_regenerate_chec…
rampitec Jul 31, 2025
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
164 changes: 11 additions & 153 deletions llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1499,60 +1499,13 @@ define amdgpu_kernel void @fptrunc_f64_to_f16_afn(
; GFX1250-SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s8, s2
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s9, s3
; GFX1250-SDAG-FAKE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], null
; GFX1250-SDAG-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s2, v1
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s3, s2, 0x1ff
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s5, s2, 8
; GFX1250-SDAG-FAKE16-NEXT: v_or_b32_e32 v0, s3, v0
; GFX1250-SDAG-FAKE16-NEXT: s_bfe_u32 s3, s2, 0xb0014
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s5, s5, 0xffe
; GFX1250-SDAG-FAKE16-NEXT: s_sub_co_i32 s4, 0x3f1, s3
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX1250-SDAG-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0
; GFX1250-SDAG-FAKE16-NEXT: v_med3_i32 v1, s4, 0, 13
; GFX1250-SDAG-FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s8, v1
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s4, v0
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s4, s5, s4
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s5, s4, 0x1000
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s9, s5, s8
; GFX1250-SDAG-FAKE16-NEXT: s_lshl_b32 s8, s9, s8
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lg_u32 s8, s5
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s5, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_addk_co_i32 s3, 0xfc10
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s5, s9, s5
; GFX1250-SDAG-FAKE16-NEXT: s_lshl_b32 s8, s3, 12
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s8, s4, s8
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lt_i32 s3, 1
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s5, s5, s8
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s8, s5, 7
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_gt_i32 s8, 5
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s9, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_eq_u32 s8, 3
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s8, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s5, s5, 2
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s8, s8, s9
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_add_co_i32 s5, s5, s8
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lt_i32 s3, 31
; GFX1250-SDAG-FAKE16-NEXT: s_movk_i32 s8, 0x7e00
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s5, s5, 0x7c00
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lg_u32 s4, 0
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s4, s8, 0x7c00
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_eq_u32 s3, 0x40f
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s3, s4, s5
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s2, s2, 16
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s4, s0
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s2, s2, 0x8000
; GFX1250-SDAG-FAKE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], null
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s5, s1
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s2, s2, s3
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-SDAG-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-SDAG-FAKE16-NEXT: v_cvt_f32_f64_e32 v0, v[0:1]
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-SDAG-FAKE16-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null
; GFX1250-SDAG-FAKE16-NEXT: s_endpgm
;
Expand Down Expand Up @@ -3538,109 +3491,14 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16_afn(
; GFX1250-SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s8, s2
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s9, s3
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s4, s0
; GFX1250-SDAG-FAKE16-NEXT: buffer_load_b128 v[0:3], off, s[8:11], null
; GFX1250-SDAG-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s2, v3
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s3, s2, 0x1ff
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s5, s2, 8
; GFX1250-SDAG-FAKE16-NEXT: v_or_b32_e32 v2, s3, v2
; GFX1250-SDAG-FAKE16-NEXT: s_bfe_u32 s3, s2, 0xb0014
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s5, s5, 0xffe
; GFX1250-SDAG-FAKE16-NEXT: s_sub_co_i32 s4, 0x3f1, s3
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX1250-SDAG-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v2
; GFX1250-SDAG-FAKE16-NEXT: v_med3_i32 v3, s4, 0, 13
; GFX1250-SDAG-FAKE16-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s8, v3
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s4, v2
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s4, s5, s4
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s5, s4, 0x1000
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s9, s5, s8
; GFX1250-SDAG-FAKE16-NEXT: s_lshl_b32 s8, s9, s8
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lg_u32 s8, s5
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s5, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_addk_co_i32 s3, 0xfc10
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s5, s9, s5
; GFX1250-SDAG-FAKE16-NEXT: s_lshl_b32 s8, s3, 12
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s8, s4, s8
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lt_i32 s3, 1
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s5, s5, s8
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s8, s5, 7
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_gt_i32 s8, 5
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s9, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_eq_u32 s8, 3
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s8, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s5, s5, 2
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s8, s8, s9
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_add_co_i32 s5, s5, s8
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lt_i32 s3, 31
; GFX1250-SDAG-FAKE16-NEXT: s_movk_i32 s8, 0x7e00
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s5, s5, 0x7c00
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lg_u32 s4, 0
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s4, v1
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s9, s8, 0x7c00
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_eq_u32 s3, 0x40f
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s3, s9, s5
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s5, s4, 0x1ff
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s10, s4, 8
; GFX1250-SDAG-FAKE16-NEXT: v_or_b32_e32 v0, s5, v0
; GFX1250-SDAG-FAKE16-NEXT: s_bfe_u32 s5, s4, 0xb0014
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s10, s10, 0xffe
; GFX1250-SDAG-FAKE16-NEXT: s_sub_co_i32 s9, 0x3f1, s5
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s2, s2, 16
; GFX1250-SDAG-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0
; GFX1250-SDAG-FAKE16-NEXT: v_med3_i32 v1, s9, 0, 13
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s2, s2, 0x8000
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s2, s2, s3
; GFX1250-SDAG-FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s11, v1
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: v_readfirstlane_b32 s9, v0
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s9, s10, s9
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s10, s9, 0x1000
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s12, s10, s11
; GFX1250-SDAG-FAKE16-NEXT: s_lshl_b32 s11, s12, s11
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lg_u32 s11, s10
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s3, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_addk_co_i32 s5, 0xfc10
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s3, s12, s3
; GFX1250-SDAG-FAKE16-NEXT: s_lshl_b32 s10, s5, 12
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s10, s9, s10
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lt_i32 s5, 1
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s3, s3, s10
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s10, s3, 7
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_gt_i32 s10, 5
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s11, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_eq_u32 s10, 3
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s10, 1, 0
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s3, s3, 2
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s10, s10, s11
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_add_co_i32 s3, s3, s10
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lt_i32 s5, 31
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s3, s3, 0x7c00
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_lg_u32 s9, 0
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s8, s8, 0x7c00
; GFX1250-SDAG-FAKE16-NEXT: s_cmp_eq_u32 s5, 0x40f
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s5, s1
; GFX1250-SDAG-FAKE16-NEXT: s_cselect_b32 s3, s8, s3
; GFX1250-SDAG-FAKE16-NEXT: s_lshr_b32 s4, s4, 16
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: s_and_b32 s4, s4, 0x8000
; GFX1250-SDAG-FAKE16-NEXT: s_or_b32 s3, s4, s3
; GFX1250-SDAG-FAKE16-NEXT: s_mov_b32 s4, s0
; GFX1250-SDAG-FAKE16-NEXT: s_pack_ll_b32_b16 s2, s3, s2
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-SDAG-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-SDAG-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-SDAG-FAKE16-NEXT: v_cvt_f32_f64_e32 v2, v[2:3]
; GFX1250-SDAG-FAKE16-NEXT: v_cvt_f32_f64_e32 v0, v[0:1]
; GFX1250-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-SDAG-FAKE16-NEXT: v_cvt_pk_f16_f32 v0, v0, v2
; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null
; GFX1250-SDAG-FAKE16-NEXT: s_endpgm
;
Expand Down
Loading