Please use GitHub pull requests for new patches. Avoid migrating existing patches. Phabricator shutdown timeline
Changeset View
Changeset View
Standalone View
Standalone View
llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll
- This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 4,231 Lines • ▼ Show 20 Lines | ; GFX11-NEXT: s_setpc_b64 s[30:31] | ||||
%result = call i48 @llvm.ssub.sat.i48(i48 %lhs, i48 %rhs) | %result = call i48 @llvm.ssub.sat.i48(i48 %lhs, i48 %rhs) | ||||
ret i48 %result | ret i48 %result | ||||
} | } | ||||
define amdgpu_ps i48 @s_ssubsat_i48(i48 inreg %lhs, i48 inreg %rhs) { | define amdgpu_ps i48 @s_ssubsat_i48(i48 inreg %lhs, i48 inreg %rhs) { | ||||
; GFX6-LABEL: s_ssubsat_i48: | ; GFX6-LABEL: s_ssubsat_i48: | ||||
; GFX6: ; %bb.0: | ; GFX6: ; %bb.0: | ||||
; GFX6-NEXT: s_sub_u32 s4, s0, s2 | ; GFX6-NEXT: s_sub_u32 s4, s0, s2 | ||||
; GFX6-NEXT: s_subb_u32 s5, s1, s3 | ; GFX6-NEXT: s_subb_u32 s3, s1, s3 | ||||
; GFX6-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ; GFX6-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ||||
; GFX6-NEXT: v_mov_b32_e32 v0, s0 | ; GFX6-NEXT: v_mov_b32_e32 v0, s0 | ||||
; GFX6-NEXT: s_bfe_i64 s[6:7], s[4:5], 0x300000 | ; GFX6-NEXT: s_bfe_i64 s[6:7], s[4:5], 0x300000 | ||||
; GFX6-NEXT: v_mov_b32_e32 v1, s1 | ; GFX6-NEXT: v_mov_b32_e32 v1, s1 | ||||
; GFX6-NEXT: s_bfe_i64 s[0:1], s[2:3], 0x300000 | ; GFX6-NEXT: s_bfe_i64 s[0:1], s[2:3], 0x300000 | ||||
; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, s[6:7], v[0:1] | ; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, s[6:7], v[0:1] | ||||
; GFX6-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], 0 | ; GFX6-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], 0 | ||||
; GFX6-NEXT: s_ashr_i32 s3, s7, 31 | ; GFX6-NEXT: s_ashr_i32 s2, s7, 31 | ||||
; GFX6-NEXT: s_ashr_i32 s2, s7, 15 | ; GFX6-NEXT: s_ashr_i32 s5, s7, 15 | ||||
; GFX6-NEXT: s_add_u32 s3, s3, 0xffff8000 | ; GFX6-NEXT: s_add_u32 s2, s2, 0xffff8000 | ||||
; GFX6-NEXT: v_mov_b32_e32 v0, s2 | ; GFX6-NEXT: v_mov_b32_e32 v0, s5 | ||||
; GFX6-NEXT: v_mov_b32_e32 v1, s3 | ; GFX6-NEXT: v_mov_b32_e32 v1, s2 | ||||
; GFX6-NEXT: v_mov_b32_e32 v2, s4 | ; GFX6-NEXT: v_mov_b32_e32 v2, s4 | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s5 | ; GFX6-NEXT: v_mov_b32_e32 v3, s3 | ||||
; GFX6-NEXT: s_xor_b64 vcc, s[0:1], vcc | ; GFX6-NEXT: s_xor_b64 vcc, s[0:1], vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc | ||||
; GFX6-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX6-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX6-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX6-NEXT: ; return to shader part epilog | ; GFX6-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX8-LABEL: s_ssubsat_i48: | ; GFX8-LABEL: s_ssubsat_i48: | ||||
; GFX8: ; %bb.0: | ; GFX8: ; %bb.0: | ||||
; GFX8-NEXT: s_sub_u32 s4, s0, s2 | ; GFX8-NEXT: s_sub_u32 s4, s0, s2 | ||||
; GFX8-NEXT: s_subb_u32 s5, s1, s3 | ; GFX8-NEXT: s_subb_u32 s3, s1, s3 | ||||
; GFX8-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ; GFX8-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ||||
; GFX8-NEXT: v_mov_b32_e32 v0, s0 | ; GFX8-NEXT: v_mov_b32_e32 v0, s0 | ||||
; GFX8-NEXT: s_bfe_i64 s[6:7], s[4:5], 0x300000 | ; GFX8-NEXT: s_bfe_i64 s[6:7], s[4:5], 0x300000 | ||||
; GFX8-NEXT: v_mov_b32_e32 v1, s1 | ; GFX8-NEXT: v_mov_b32_e32 v1, s1 | ||||
; GFX8-NEXT: s_bfe_i64 s[0:1], s[2:3], 0x300000 | ; GFX8-NEXT: s_bfe_i64 s[0:1], s[2:3], 0x300000 | ||||
; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, s[6:7], v[0:1] | ; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, s[6:7], v[0:1] | ||||
; GFX8-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], 0 | ; GFX8-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], 0 | ||||
; GFX8-NEXT: s_ashr_i32 s3, s7, 31 | ; GFX8-NEXT: s_ashr_i32 s2, s7, 31 | ||||
; GFX8-NEXT: s_ashr_i32 s2, s7, 15 | ; GFX8-NEXT: s_ashr_i32 s5, s7, 15 | ||||
; GFX8-NEXT: s_add_u32 s3, s3, 0xffff8000 | ; GFX8-NEXT: s_add_u32 s2, s2, 0xffff8000 | ||||
; GFX8-NEXT: v_mov_b32_e32 v0, s2 | ; GFX8-NEXT: v_mov_b32_e32 v0, s5 | ||||
; GFX8-NEXT: v_mov_b32_e32 v1, s3 | ; GFX8-NEXT: v_mov_b32_e32 v1, s2 | ||||
; GFX8-NEXT: v_mov_b32_e32 v2, s4 | ; GFX8-NEXT: v_mov_b32_e32 v2, s4 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s5 | ; GFX8-NEXT: v_mov_b32_e32 v3, s3 | ||||
; GFX8-NEXT: s_xor_b64 vcc, s[0:1], vcc | ; GFX8-NEXT: s_xor_b64 vcc, s[0:1], vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc | ||||
; GFX8-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX8-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX8-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX8-NEXT: ; return to shader part epilog | ; GFX8-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX9-LABEL: s_ssubsat_i48: | ; GFX9-LABEL: s_ssubsat_i48: | ||||
Show All 21 Lines | |||||
; GFX9-NEXT: ; return to shader part epilog | ; GFX9-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX10-LABEL: s_ssubsat_i48: | ; GFX10-LABEL: s_ssubsat_i48: | ||||
; GFX10: ; %bb.0: | ; GFX10: ; %bb.0: | ||||
; GFX10-NEXT: s_lshl_b64 s[0:1], s[0:1], 16 | ; GFX10-NEXT: s_lshl_b64 s[0:1], s[0:1], 16 | ||||
; GFX10-NEXT: s_lshl_b64 s[2:3], s[2:3], 16 | ; GFX10-NEXT: s_lshl_b64 s[2:3], s[2:3], 16 | ||||
; GFX10-NEXT: s_sub_u32 s4, s0, s2 | ; GFX10-NEXT: s_sub_u32 s4, s0, s2 | ||||
; GFX10-NEXT: s_subb_u32 s5, s1, s3 | ; GFX10-NEXT: s_subb_u32 s5, s1, s3 | ||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s2, s[2:3], 0 | |||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s6, s[4:5], s[0:1] | |||||
; GFX10-NEXT: v_mov_b32_e32 v0, s4 | ; GFX10-NEXT: v_mov_b32_e32 v0, s4 | ||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s0, s[4:5], s[0:1] | |||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s1, s[2:3], 0 | |||||
; GFX10-NEXT: v_mov_b32_e32 v1, s5 | ; GFX10-NEXT: v_mov_b32_e32 v1, s5 | ||||
; GFX10-NEXT: s_ashr_i32 s0, s5, 31 | ; GFX10-NEXT: s_ashr_i32 s2, s5, 31 | ||||
; GFX10-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX10-NEXT: s_add_u32 s3, s2, 0x80000000 | ||||
; GFX10-NEXT: s_xor_b32 s2, s2, s6 | ; GFX10-NEXT: s_xor_b32 s0, s1, s0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s0, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s2, s0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s1, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s3, s0 | ||||
; GFX10-NEXT: v_ashrrev_i64 v[0:1], 16, v[0:1] | ; GFX10-NEXT: v_ashrrev_i64 v[0:1], 16, v[0:1] | ||||
; GFX10-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX10-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX10-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX10-NEXT: ; return to shader part epilog | ; GFX10-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX11-LABEL: s_ssubsat_i48: | ; GFX11-LABEL: s_ssubsat_i48: | ||||
; GFX11: ; %bb.0: | ; GFX11: ; %bb.0: | ||||
; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], 16 | ; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], 16 | ||||
; GFX11-NEXT: s_lshl_b64 s[2:3], s[2:3], 16 | ; GFX11-NEXT: s_lshl_b64 s[2:3], s[2:3], 16 | ||||
; GFX11-NEXT: s_sub_u32 s4, s0, s2 | ; GFX11-NEXT: s_sub_u32 s4, s0, s2 | ||||
; GFX11-NEXT: s_subb_u32 s5, s1, s3 | ; GFX11-NEXT: s_subb_u32 s5, s1, s3 | ||||
; GFX11-NEXT: v_cmp_gt_i64_e64 s2, s[2:3], 0 | |||||
; GFX11-NEXT: v_cmp_lt_i64_e64 s6, s[4:5], s[0:1] | |||||
; GFX11-NEXT: v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5 | ; GFX11-NEXT: v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5 | ||||
; GFX11-NEXT: s_ashr_i32 s0, s5, 31 | ; GFX11-NEXT: v_cmp_lt_i64_e64 s0, s[4:5], s[0:1] | ||||
; GFX11-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX11-NEXT: v_cmp_gt_i64_e64 s1, s[2:3], 0 | ||||
; GFX11-NEXT: s_xor_b32 s2, s2, s6 | ; GFX11-NEXT: s_ashr_i32 s2, s5, 31 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s0, s2 | ; GFX11-NEXT: s_add_u32 s3, s2, 0x80000000 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s1, s2 | ; GFX11-NEXT: s_xor_b32 s0, s1, s0 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s2, s0 | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s3, s0 | |||||
; GFX11-NEXT: v_ashrrev_i64 v[0:1], 16, v[0:1] | ; GFX11-NEXT: v_ashrrev_i64 v[0:1], 16, v[0:1] | ||||
; GFX11-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX11-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX11-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX11-NEXT: ; return to shader part epilog | ; GFX11-NEXT: ; return to shader part epilog | ||||
%result = call i48 @llvm.ssub.sat.i48(i48 %lhs, i48 %rhs) | %result = call i48 @llvm.ssub.sat.i48(i48 %lhs, i48 %rhs) | ||||
ret i48 %result | ret i48 %result | ||||
} | } | ||||
define amdgpu_ps <2 x float> @ssubsat_i48_sv(i48 inreg %lhs, i48 %rhs) { | define amdgpu_ps <2 x float> @ssubsat_i48_sv(i48 inreg %lhs, i48 %rhs) { | ||||
; GFX6-LABEL: ssubsat_i48_sv: | ; GFX6-LABEL: ssubsat_i48_sv: | ||||
; GFX6: ; %bb.0: | ; GFX6: ; %bb.0: | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s1 | ; GFX6-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX6-NEXT: v_sub_i32_e32 v2, vcc, s0, v0 | ; GFX6-NEXT: v_sub_i32_e32 v2, vcc, s0, v0 | ||||
; GFX6-NEXT: v_subb_u32_e32 v4, vcc, v3, v1, vcc | ; GFX6-NEXT: v_subb_u32_e32 v4, vcc, v3, v1, vcc | ||||
; GFX6-NEXT: v_bfe_i32 v3, v2, 0, 16 | ; GFX6-NEXT: v_bfe_i32 v3, v2, 0, 16 | ||||
; GFX6-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ; GFX6-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ||||
; GFX6-NEXT: v_bfe_i32 v1, v0, 0, 16 | ; GFX6-NEXT: v_bfe_i32 v1, v0, 0, 16 | ||||
; GFX6-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], v[2:3] | ; GFX6-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], v[2:3] | ||||
; GFX6-NEXT: v_cmp_lt_i64_e64 s[2:3], 0, v[0:1] | ; GFX6-NEXT: v_cmp_lt_i64_e64 s[2:3], 0, v[0:1] | ||||
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ; GFX6-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ||||
; GFX6-NEXT: v_ashrrev_i32_e32 v1, 15, v3 | ; GFX6-NEXT: v_ashrrev_i32_e32 v1, 15, v3 | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, 0xffff8000 | ; GFX6-NEXT: v_add_i32_e32 v3, vcc, 0xffff8000, v0 | ||||
; GFX6-NEXT: v_add_i32_e32 v3, vcc, v0, v3 | |||||
; GFX6-NEXT: s_xor_b64 vcc, s[2:3], s[0:1] | ; GFX6-NEXT: s_xor_b64 vcc, s[2:3], s[0:1] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc | ||||
; GFX6-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ; GFX6-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ||||
; GFX6-NEXT: ; return to shader part epilog | ; GFX6-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX8-LABEL: ssubsat_i48_sv: | ; GFX8-LABEL: ssubsat_i48_sv: | ||||
; GFX8: ; %bb.0: | ; GFX8: ; %bb.0: | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX8-NEXT: v_sub_u32_e32 v2, vcc, s0, v0 | ; GFX8-NEXT: v_sub_u32_e32 v2, vcc, s0, v0 | ||||
; GFX8-NEXT: v_subb_u32_e32 v4, vcc, v3, v1, vcc | ; GFX8-NEXT: v_subb_u32_e32 v4, vcc, v3, v1, vcc | ||||
; GFX8-NEXT: v_bfe_i32 v3, v2, 0, 16 | ; GFX8-NEXT: v_bfe_i32 v3, v2, 0, 16 | ||||
; GFX8-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ; GFX8-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ||||
; GFX8-NEXT: v_bfe_i32 v1, v0, 0, 16 | ; GFX8-NEXT: v_bfe_i32 v1, v0, 0, 16 | ||||
; GFX8-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], v[2:3] | ; GFX8-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], v[2:3] | ||||
; GFX8-NEXT: v_cmp_lt_i64_e64 s[2:3], 0, v[0:1] | ; GFX8-NEXT: v_cmp_lt_i64_e64 s[2:3], 0, v[0:1] | ||||
; GFX8-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ; GFX8-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ||||
; GFX8-NEXT: v_ashrrev_i32_e32 v1, 15, v3 | ; GFX8-NEXT: v_ashrrev_i32_e32 v1, 15, v3 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, 0xffff8000 | ; GFX8-NEXT: v_add_u32_e32 v3, vcc, 0xffff8000, v0 | ||||
; GFX8-NEXT: v_add_u32_e32 v3, vcc, v0, v3 | |||||
; GFX8-NEXT: s_xor_b64 vcc, s[2:3], s[0:1] | ; GFX8-NEXT: s_xor_b64 vcc, s[2:3], s[0:1] | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc | ||||
; GFX8-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ; GFX8-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ||||
; GFX8-NEXT: ; return to shader part epilog | ; GFX8-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX9-LABEL: ssubsat_i48_sv: | ; GFX9-LABEL: ssubsat_i48_sv: | ||||
; GFX9: ; %bb.0: | ; GFX9: ; %bb.0: | ||||
; GFX9-NEXT: v_lshlrev_b64 v[0:1], 16, v[0:1] | ; GFX9-NEXT: v_lshlrev_b64 v[0:1], 16, v[0:1] | ||||
; GFX9-NEXT: s_lshl_b64 s[0:1], s[0:1], 16 | ; GFX9-NEXT: s_lshl_b64 s[0:1], s[0:1], 16 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX9-NEXT: v_sub_co_u32_e32 v2, vcc, s0, v0 | ; GFX9-NEXT: v_sub_co_u32_e32 v2, vcc, s0, v0 | ||||
; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v3, v1, vcc | ; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v3, v1, vcc | ||||
; GFX9-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], v[2:3] | ; GFX9-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], v[2:3] | ||||
; GFX9-NEXT: v_cmp_lt_i64_e64 s[2:3], 0, v[0:1] | ; GFX9-NEXT: v_cmp_lt_i64_e64 s[2:3], 0, v[0:1] | ||||
; GFX9-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ; GFX9-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ||||
; GFX9-NEXT: v_bfrev_b32_e32 v1, 1 | ; GFX9-NEXT: v_add_co_u32_e32 v1, vcc, 0x80000000, v0 | ||||
; GFX9-NEXT: v_add_co_u32_e32 v1, vcc, v0, v1 | |||||
; GFX9-NEXT: s_xor_b64 vcc, s[2:3], s[0:1] | ; GFX9-NEXT: s_xor_b64 vcc, s[2:3], s[0:1] | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc | ||||
; GFX9-NEXT: v_ashrrev_i64 v[0:1], 16, v[0:1] | ; GFX9-NEXT: v_ashrrev_i64 v[0:1], 16, v[0:1] | ||||
; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ||||
; GFX9-NEXT: ; return to shader part epilog | ; GFX9-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX10-LABEL: ssubsat_i48_sv: | ; GFX10-LABEL: ssubsat_i48_sv: | ||||
▲ Show 20 Lines • Show All 42 Lines • ▼ Show 20 Lines | |||||
; GFX6-NEXT: v_subb_u32_e32 v4, vcc, v1, v3, vcc | ; GFX6-NEXT: v_subb_u32_e32 v4, vcc, v1, v3, vcc | ||||
; GFX6-NEXT: v_bfe_i32 v3, v2, 0, 16 | ; GFX6-NEXT: v_bfe_i32 v3, v2, 0, 16 | ||||
; GFX6-NEXT: v_bfe_i32 v1, v0, 0, 16 | ; GFX6-NEXT: v_bfe_i32 v1, v0, 0, 16 | ||||
; GFX6-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ; GFX6-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ||||
; GFX6-NEXT: v_cmp_lt_i64_e64 s[2:3], v[2:3], v[0:1] | ; GFX6-NEXT: v_cmp_lt_i64_e64 s[2:3], v[2:3], v[0:1] | ||||
; GFX6-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], 0 | ; GFX6-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], 0 | ||||
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ; GFX6-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ||||
; GFX6-NEXT: v_ashrrev_i32_e32 v1, 15, v3 | ; GFX6-NEXT: v_ashrrev_i32_e32 v1, 15, v3 | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, 0xffff8000 | ; GFX6-NEXT: v_add_i32_e32 v3, vcc, 0xffff8000, v0 | ||||
; GFX6-NEXT: v_add_i32_e32 v3, vcc, v0, v3 | |||||
; GFX6-NEXT: s_xor_b64 vcc, s[0:1], s[2:3] | ; GFX6-NEXT: s_xor_b64 vcc, s[0:1], s[2:3] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc | ||||
; GFX6-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ; GFX6-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ||||
; GFX6-NEXT: ; return to shader part epilog | ; GFX6-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX8-LABEL: ssubsat_i48_vs: | ; GFX8-LABEL: ssubsat_i48_vs: | ||||
; GFX8: ; %bb.0: | ; GFX8: ; %bb.0: | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX8-NEXT: v_subrev_u32_e32 v2, vcc, s0, v0 | ; GFX8-NEXT: v_subrev_u32_e32 v2, vcc, s0, v0 | ||||
; GFX8-NEXT: v_subb_u32_e32 v4, vcc, v1, v3, vcc | ; GFX8-NEXT: v_subb_u32_e32 v4, vcc, v1, v3, vcc | ||||
; GFX8-NEXT: v_bfe_i32 v3, v2, 0, 16 | ; GFX8-NEXT: v_bfe_i32 v3, v2, 0, 16 | ||||
; GFX8-NEXT: v_bfe_i32 v1, v0, 0, 16 | ; GFX8-NEXT: v_bfe_i32 v1, v0, 0, 16 | ||||
; GFX8-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ; GFX8-NEXT: s_bfe_i64 s[0:1], s[0:1], 0x300000 | ||||
; GFX8-NEXT: v_cmp_lt_i64_e64 s[2:3], v[2:3], v[0:1] | ; GFX8-NEXT: v_cmp_lt_i64_e64 s[2:3], v[2:3], v[0:1] | ||||
; GFX8-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], 0 | ; GFX8-NEXT: v_cmp_gt_i64_e64 s[0:1], s[0:1], 0 | ||||
; GFX8-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ; GFX8-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ||||
; GFX8-NEXT: v_ashrrev_i32_e32 v1, 15, v3 | ; GFX8-NEXT: v_ashrrev_i32_e32 v1, 15, v3 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, 0xffff8000 | ; GFX8-NEXT: v_add_u32_e32 v3, vcc, 0xffff8000, v0 | ||||
; GFX8-NEXT: v_add_u32_e32 v3, vcc, v0, v3 | |||||
; GFX8-NEXT: s_xor_b64 vcc, s[0:1], s[2:3] | ; GFX8-NEXT: s_xor_b64 vcc, s[0:1], s[2:3] | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc | ||||
; GFX8-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ; GFX8-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ||||
; GFX8-NEXT: ; return to shader part epilog | ; GFX8-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX9-LABEL: ssubsat_i48_vs: | ; GFX9-LABEL: ssubsat_i48_vs: | ||||
; GFX9: ; %bb.0: | ; GFX9: ; %bb.0: | ||||
; GFX9-NEXT: v_lshlrev_b64 v[0:1], 16, v[0:1] | ; GFX9-NEXT: v_lshlrev_b64 v[0:1], 16, v[0:1] | ||||
; GFX9-NEXT: s_lshl_b64 s[2:3], s[0:1], 16 | ; GFX9-NEXT: s_lshl_b64 s[2:3], s[0:1], 16 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s3 | ; GFX9-NEXT: v_mov_b32_e32 v3, s3 | ||||
; GFX9-NEXT: v_subrev_co_u32_e32 v2, vcc, s2, v0 | ; GFX9-NEXT: v_subrev_co_u32_e32 v2, vcc, s2, v0 | ||||
; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v3, vcc | ; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v3, vcc | ||||
; GFX9-NEXT: v_cmp_lt_i64_e64 s[0:1], v[2:3], v[0:1] | ; GFX9-NEXT: v_cmp_lt_i64_e64 s[0:1], v[2:3], v[0:1] | ||||
; GFX9-NEXT: v_cmp_gt_i64_e64 s[2:3], s[2:3], 0 | ; GFX9-NEXT: v_cmp_gt_i64_e64 s[2:3], s[2:3], 0 | ||||
; GFX9-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ; GFX9-NEXT: v_ashrrev_i32_e32 v0, 31, v3 | ||||
; GFX9-NEXT: v_bfrev_b32_e32 v1, 1 | ; GFX9-NEXT: v_add_co_u32_e32 v1, vcc, 0x80000000, v0 | ||||
; GFX9-NEXT: v_add_co_u32_e32 v1, vcc, v0, v1 | |||||
; GFX9-NEXT: s_xor_b64 vcc, s[2:3], s[0:1] | ; GFX9-NEXT: s_xor_b64 vcc, s[2:3], s[0:1] | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc | ||||
; GFX9-NEXT: v_ashrrev_i64 v[0:1], 16, v[0:1] | ; GFX9-NEXT: v_ashrrev_i64 v[0:1], 16, v[0:1] | ||||
; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v1 | ||||
; GFX9-NEXT: ; return to shader part epilog | ; GFX9-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX10-LABEL: ssubsat_i48_vs: | ; GFX10-LABEL: ssubsat_i48_vs: | ||||
▲ Show 20 Lines • Show All 172 Lines • ▼ Show 20 Lines | |||||
; GFX9-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX9-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX9-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX9-NEXT: ; return to shader part epilog | ; GFX9-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX10-LABEL: s_ssubsat_i64: | ; GFX10-LABEL: s_ssubsat_i64: | ||||
; GFX10: ; %bb.0: | ; GFX10: ; %bb.0: | ||||
; GFX10-NEXT: s_sub_u32 s4, s0, s2 | ; GFX10-NEXT: s_sub_u32 s4, s0, s2 | ||||
; GFX10-NEXT: s_subb_u32 s5, s1, s3 | ; GFX10-NEXT: s_subb_u32 s5, s1, s3 | ||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s2, s[2:3], 0 | |||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s6, s[4:5], s[0:1] | |||||
; GFX10-NEXT: v_mov_b32_e32 v0, s4 | ; GFX10-NEXT: v_mov_b32_e32 v0, s4 | ||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s0, s[4:5], s[0:1] | |||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s1, s[2:3], 0 | |||||
; GFX10-NEXT: v_mov_b32_e32 v1, s5 | ; GFX10-NEXT: v_mov_b32_e32 v1, s5 | ||||
; GFX10-NEXT: s_ashr_i32 s0, s5, 31 | ; GFX10-NEXT: s_ashr_i32 s2, s5, 31 | ||||
; GFX10-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX10-NEXT: s_add_u32 s3, s2, 0x80000000 | ||||
; GFX10-NEXT: s_xor_b32 s2, s2, s6 | ; GFX10-NEXT: s_xor_b32 s0, s1, s0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s0, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s2, s0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s1, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s3, s0 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX10-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX10-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX10-NEXT: ; return to shader part epilog | ; GFX10-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX11-LABEL: s_ssubsat_i64: | ; GFX11-LABEL: s_ssubsat_i64: | ||||
; GFX11: ; %bb.0: | ; GFX11: ; %bb.0: | ||||
; GFX11-NEXT: s_sub_u32 s4, s0, s2 | ; GFX11-NEXT: s_sub_u32 s4, s0, s2 | ||||
; GFX11-NEXT: s_subb_u32 s5, s1, s3 | ; GFX11-NEXT: s_subb_u32 s5, s1, s3 | ||||
; GFX11-NEXT: v_cmp_gt_i64_e64 s2, s[2:3], 0 | |||||
; GFX11-NEXT: v_cmp_lt_i64_e64 s6, s[4:5], s[0:1] | |||||
; GFX11-NEXT: v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5 | ; GFX11-NEXT: v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5 | ||||
; GFX11-NEXT: s_ashr_i32 s0, s5, 31 | ; GFX11-NEXT: v_cmp_lt_i64_e64 s0, s[4:5], s[0:1] | ||||
; GFX11-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX11-NEXT: v_cmp_gt_i64_e64 s1, s[2:3], 0 | ||||
; GFX11-NEXT: s_xor_b32 s2, s2, s6 | ; GFX11-NEXT: s_ashr_i32 s2, s5, 31 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s0, s2 | ; GFX11-NEXT: s_add_u32 s3, s2, 0x80000000 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s1, s2 | ; GFX11-NEXT: s_xor_b32 s0, s1, s0 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s2, s0 | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s3, s0 | |||||
; GFX11-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX11-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX11-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX11-NEXT: ; return to shader part epilog | ; GFX11-NEXT: ; return to shader part epilog | ||||
%result = call i64 @llvm.ssub.sat.i64(i64 %lhs, i64 %rhs) | %result = call i64 @llvm.ssub.sat.i64(i64 %lhs, i64 %rhs) | ||||
ret i64 %result | ret i64 %result | ||||
} | } | ||||
define amdgpu_ps <2 x float> @ssubsat_i64_sv(i64 inreg %lhs, i64 %rhs) { | define amdgpu_ps <2 x float> @ssubsat_i64_sv(i64 inreg %lhs, i64 %rhs) { | ||||
▲ Show 20 Lines • Show All 376 Lines • ▼ Show 20 Lines | |||||
; GFX9-NEXT: v_readfirstlane_b32 s2, v0 | ; GFX9-NEXT: v_readfirstlane_b32 s2, v0 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s3, v1 | ; GFX9-NEXT: v_readfirstlane_b32 s3, v1 | ||||
; GFX9-NEXT: ; return to shader part epilog | ; GFX9-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX10-LABEL: s_ssubsat_v2i64: | ; GFX10-LABEL: s_ssubsat_v2i64: | ||||
; GFX10: ; %bb.0: | ; GFX10: ; %bb.0: | ||||
; GFX10-NEXT: s_sub_u32 s8, s0, s4 | ; GFX10-NEXT: s_sub_u32 s8, s0, s4 | ||||
; GFX10-NEXT: s_subb_u32 s9, s1, s5 | ; GFX10-NEXT: s_subb_u32 s9, s1, s5 | ||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s4, s[4:5], 0 | |||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s10, s[8:9], s[0:1] | |||||
; GFX10-NEXT: s_ashr_i32 s0, s9, 31 | |||||
; GFX10-NEXT: v_mov_b32_e32 v0, s8 | ; GFX10-NEXT: v_mov_b32_e32 v0, s8 | ||||
; GFX10-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX10-NEXT: v_cmp_lt_i64_e64 s0, s[8:9], s[0:1] | ||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s1, s[4:5], 0 | |||||
; GFX10-NEXT: s_ashr_i32 s4, s9, 31 | |||||
; GFX10-NEXT: v_mov_b32_e32 v1, s9 | ; GFX10-NEXT: v_mov_b32_e32 v1, s9 | ||||
; GFX10-NEXT: s_xor_b32 s8, s4, s10 | ; GFX10-NEXT: s_add_u32 s5, s4, 0x80000000 | ||||
; GFX10-NEXT: s_sub_u32 s4, s2, s6 | ; GFX10-NEXT: s_xor_b32 s8, s1, s0 | ||||
; GFX10-NEXT: s_subb_u32 s5, s3, s7 | ; GFX10-NEXT: s_sub_u32 s0, s2, s6 | ||||
; GFX10-NEXT: v_mov_b32_e32 v2, s4 | ; GFX10-NEXT: s_subb_u32 s1, s3, s7 | ||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s2, s[4:5], s[2:3] | ; GFX10-NEXT: v_mov_b32_e32 v2, s0 | ||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s2, s[0:1], s[2:3] | |||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s3, s[6:7], 0 | ; GFX10-NEXT: v_cmp_gt_i64_e64 s3, s[6:7], 0 | ||||
; GFX10-NEXT: v_mov_b32_e32 v3, s5 | ; GFX10-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s0, s8 | ; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s4, s8 | ||||
; GFX10-NEXT: s_ashr_i32 s0, s5, 31 | ; GFX10-NEXT: s_ashr_i32 s4, s1, 31 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s1, s8 | ; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s5, s8 | ||||
; GFX10-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX10-NEXT: s_add_u32 s0, s4, 0x80000000 | ||||
; GFX10-NEXT: s_xor_b32 s2, s3, s2 | ; GFX10-NEXT: s_xor_b32 s1, s3, s2 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v2, v2, s0, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v2, v2, s4, s1 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v3, v3, s1, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v3, v3, s0, s1 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX10-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX10-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s2, v2 | ; GFX10-NEXT: v_readfirstlane_b32 s2, v2 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s3, v3 | ; GFX10-NEXT: v_readfirstlane_b32 s3, v3 | ||||
; GFX10-NEXT: ; return to shader part epilog | ; GFX10-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX11-LABEL: s_ssubsat_v2i64: | ; GFX11-LABEL: s_ssubsat_v2i64: | ||||
; GFX11: ; %bb.0: | ; GFX11: ; %bb.0: | ||||
; GFX11-NEXT: s_sub_u32 s8, s0, s4 | ; GFX11-NEXT: s_sub_u32 s8, s0, s4 | ||||
; GFX11-NEXT: s_subb_u32 s9, s1, s5 | ; GFX11-NEXT: s_subb_u32 s9, s1, s5 | ||||
; GFX11-NEXT: v_cmp_gt_i64_e64 s4, s[4:5], 0 | |||||
; GFX11-NEXT: v_cmp_lt_i64_e64 s10, s[8:9], s[0:1] | |||||
; GFX11-NEXT: s_ashr_i32 s0, s9, 31 | |||||
; GFX11-NEXT: v_dual_mov_b32 v0, s8 :: v_dual_mov_b32 v1, s9 | ; GFX11-NEXT: v_dual_mov_b32 v0, s8 :: v_dual_mov_b32 v1, s9 | ||||
; GFX11-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX11-NEXT: v_cmp_lt_i64_e64 s0, s[8:9], s[0:1] | ||||
; GFX11-NEXT: s_xor_b32 s8, s4, s10 | ; GFX11-NEXT: v_cmp_gt_i64_e64 s1, s[4:5], 0 | ||||
; GFX11-NEXT: s_sub_u32 s4, s2, s6 | ; GFX11-NEXT: s_ashr_i32 s4, s9, 31 | ||||
; GFX11-NEXT: s_subb_u32 s5, s3, s7 | ; GFX11-NEXT: s_add_u32 s5, s4, 0x80000000 | ||||
; GFX11-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5 | ; GFX11-NEXT: s_xor_b32 s8, s1, s0 | ||||
; GFX11-NEXT: v_cmp_lt_i64_e64 s2, s[4:5], s[2:3] | ; GFX11-NEXT: s_sub_u32 s0, s2, s6 | ||||
; GFX11-NEXT: s_subb_u32 s1, s3, s7 | |||||
; GFX11-NEXT: v_dual_mov_b32 v2, s0 :: v_dual_mov_b32 v3, s1 | |||||
; GFX11-NEXT: v_cmp_lt_i64_e64 s2, s[0:1], s[2:3] | |||||
; GFX11-NEXT: v_cmp_gt_i64_e64 s3, s[6:7], 0 | ; GFX11-NEXT: v_cmp_gt_i64_e64 s3, s[6:7], 0 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s0, s8 | ; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s4, s8 | ||||
; GFX11-NEXT: s_ashr_i32 s0, s5, 31 | ; GFX11-NEXT: s_ashr_i32 s4, s1, 31 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s1, s8 | ; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s5, s8 | ||||
; GFX11-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX11-NEXT: s_add_u32 s0, s4, 0x80000000 | ||||
; GFX11-NEXT: s_xor_b32 s2, s3, s2 | ; GFX11-NEXT: s_xor_b32 s1, s3, s2 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v2, v2, s0, s2 | ; GFX11-NEXT: v_cndmask_b32_e64 v2, v2, s4, s1 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v3, v3, s1, s2 | ; GFX11-NEXT: v_cndmask_b32_e64 v3, v3, s0, s1 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX11-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX11-NEXT: v_readfirstlane_b32 s1, v1 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s2, v2 | ; GFX11-NEXT: v_readfirstlane_b32 s2, v2 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s3, v3 | ; GFX11-NEXT: v_readfirstlane_b32 s3, v3 | ||||
; GFX11-NEXT: ; return to shader part epilog | ; GFX11-NEXT: ; return to shader part epilog | ||||
%result = call <2 x i64> @llvm.ssub.sat.v2i64(<2 x i64> %lhs, <2 x i64> %rhs) | %result = call <2 x i64> @llvm.ssub.sat.v2i64(<2 x i64> %lhs, <2 x i64> %rhs) | ||||
ret <2 x i64> %result | ret <2 x i64> %result | ||||
} | } | ||||
Show All 21 Lines | |||||
; GFX6-NEXT: v_cmp_eq_u64_e64 vcc, s[6:7], 0 | ; GFX6-NEXT: v_cmp_eq_u64_e64 vcc, s[6:7], 0 | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[0:1] | ; GFX6-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[0:1] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX6-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX6-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX6-NEXT: s_ashr_i32 s0, s11, 31 | ; GFX6-NEXT: s_ashr_i32 s0, s11, 31 | ||||
; GFX6-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX6-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX6-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX6-NEXT: s_add_u32 s1, s0, 0x80000000 | ||||
; GFX6-NEXT: v_mov_b32_e32 v1, s0 | ; GFX6-NEXT: v_mov_b32_e32 v1, s0 | ||||
; GFX6-NEXT: v_mov_b32_e32 v2, s0 | ; GFX6-NEXT: v_mov_b32_e32 v2, s8 | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s8 | ; GFX6-NEXT: v_mov_b32_e32 v3, s9 | ||||
; GFX6-NEXT: v_mov_b32_e32 v4, s9 | |||||
; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v0, v3, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v4, v2, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v2, v3, v1, vcc | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s1 | ; GFX6-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX6-NEXT: v_mov_b32_e32 v4, s10 | ; GFX6-NEXT: v_mov_b32_e32 v4, s10 | ||||
; GFX6-NEXT: v_mov_b32_e32 v5, s11 | ; GFX6-NEXT: v_mov_b32_e32 v5, s11 | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v4, v1, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc | ||||
; GFX6-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX6-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX6-NEXT: v_readfirstlane_b32 s1, v2 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s2, v2 | ; GFX6-NEXT: v_readfirstlane_b32 s2, v1 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s3, v3 | ; GFX6-NEXT: v_readfirstlane_b32 s3, v3 | ||||
; GFX6-NEXT: ; return to shader part epilog | ; GFX6-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX8-LABEL: s_ssubsat_i128: | ; GFX8-LABEL: s_ssubsat_i128: | ||||
; GFX8: ; %bb.0: | ; GFX8: ; %bb.0: | ||||
; GFX8-NEXT: s_sub_u32 s8, s0, s4 | ; GFX8-NEXT: s_sub_u32 s8, s0, s4 | ||||
; GFX8-NEXT: s_subb_u32 s9, s1, s5 | ; GFX8-NEXT: s_subb_u32 s9, s1, s5 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ||||
Show All 20 Lines | |||||
; GFX8-NEXT: s_and_b32 s0, 1, s2 | ; GFX8-NEXT: s_and_b32 s0, 1, s2 | ||||
; GFX8-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 | ; GFX8-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX8-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX8-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX8-NEXT: s_ashr_i32 s0, s11, 31 | ; GFX8-NEXT: s_ashr_i32 s0, s11, 31 | ||||
; GFX8-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX8-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX8-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX8-NEXT: s_add_u32 s1, s0, 0x80000000 | ||||
; GFX8-NEXT: v_mov_b32_e32 v1, s0 | ; GFX8-NEXT: v_mov_b32_e32 v1, s0 | ||||
; GFX8-NEXT: v_mov_b32_e32 v2, s0 | ; GFX8-NEXT: v_mov_b32_e32 v2, s8 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s8 | ; GFX8-NEXT: v_mov_b32_e32 v3, s9 | ||||
; GFX8-NEXT: v_mov_b32_e32 v4, s9 | |||||
; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v0, v3, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v2, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v2, v3, v1, vcc | ||||
; GFX8-NEXT: v_mov_b32_e32 v2, s0 | |||||
; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX8-NEXT: v_mov_b32_e32 v4, s10 | ; GFX8-NEXT: v_mov_b32_e32 v4, s10 | ||||
; GFX8-NEXT: v_mov_b32_e32 v5, s11 | ; GFX8-NEXT: v_mov_b32_e32 v5, s11 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v1, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc | ||||
; GFX8-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX8-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX8-NEXT: v_readfirstlane_b32 s1, v2 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s2, v2 | ; GFX8-NEXT: v_readfirstlane_b32 s2, v1 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s3, v3 | ; GFX8-NEXT: v_readfirstlane_b32 s3, v3 | ||||
; GFX8-NEXT: ; return to shader part epilog | ; GFX8-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX9-LABEL: s_ssubsat_i128: | ; GFX9-LABEL: s_ssubsat_i128: | ||||
; GFX9: ; %bb.0: | ; GFX9: ; %bb.0: | ||||
; GFX9-NEXT: s_sub_u32 s8, s0, s4 | ; GFX9-NEXT: s_sub_u32 s8, s0, s4 | ||||
; GFX9-NEXT: s_subb_u32 s9, s1, s5 | ; GFX9-NEXT: s_subb_u32 s9, s1, s5 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ||||
Show All 20 Lines | |||||
; GFX9-NEXT: s_and_b32 s0, 1, s2 | ; GFX9-NEXT: s_and_b32 s0, 1, s2 | ||||
; GFX9-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 | ; GFX9-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX9-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX9-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX9-NEXT: s_ashr_i32 s0, s11, 31 | ; GFX9-NEXT: s_ashr_i32 s0, s11, 31 | ||||
; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX9-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX9-NEXT: s_add_u32 s1, s0, 0x80000000 | ||||
; GFX9-NEXT: v_mov_b32_e32 v1, s0 | ; GFX9-NEXT: v_mov_b32_e32 v1, s0 | ||||
; GFX9-NEXT: v_mov_b32_e32 v2, s0 | ; GFX9-NEXT: v_mov_b32_e32 v2, s8 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s8 | ; GFX9-NEXT: v_mov_b32_e32 v3, s9 | ||||
; GFX9-NEXT: v_mov_b32_e32 v4, s9 | |||||
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v4, v2, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v2, v3, v1, vcc | ||||
; GFX9-NEXT: v_mov_b32_e32 v2, s0 | |||||
; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX9-NEXT: v_mov_b32_e32 v4, s10 | ; GFX9-NEXT: v_mov_b32_e32 v4, s10 | ||||
; GFX9-NEXT: v_mov_b32_e32 v5, s11 | ; GFX9-NEXT: v_mov_b32_e32 v5, s11 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v4, v1, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc | ||||
; GFX9-NEXT: v_readfirstlane_b32 s0, v0 | ; GFX9-NEXT: v_readfirstlane_b32 s0, v0 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s1, v1 | ; GFX9-NEXT: v_readfirstlane_b32 s1, v2 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s2, v2 | ; GFX9-NEXT: v_readfirstlane_b32 s2, v1 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s3, v3 | ; GFX9-NEXT: v_readfirstlane_b32 s3, v3 | ||||
; GFX9-NEXT: ; return to shader part epilog | ; GFX9-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX10-LABEL: s_ssubsat_i128: | ; GFX10-LABEL: s_ssubsat_i128: | ||||
; GFX10: ; %bb.0: | ; GFX10: ; %bb.0: | ||||
; GFX10-NEXT: s_sub_u32 s8, s0, s4 | ; GFX10-NEXT: s_sub_u32 s8, s0, s4 | ||||
; GFX10-NEXT: s_subb_u32 s9, s1, s5 | ; GFX10-NEXT: s_subb_u32 s9, s1, s5 | ||||
; GFX10-NEXT: s_subb_u32 s10, s2, s6 | ; GFX10-NEXT: s_subb_u32 s10, s2, s6 | ||||
; GFX10-NEXT: v_cmp_lt_u64_e64 s0, s[8:9], s[0:1] | ; GFX10-NEXT: v_cmp_lt_u64_e64 s0, s[8:9], s[0:1] | ||||
; GFX10-NEXT: s_subb_u32 s11, s3, s7 | ; GFX10-NEXT: s_subb_u32 s11, s3, s7 | ||||
; GFX10-NEXT: s_cmp_eq_u64 s[10:11], s[2:3] | ; GFX10-NEXT: s_cmp_eq_u64 s[10:11], s[2:3] | ||||
; GFX10-NEXT: s_cselect_b32 s12, 1, 0 | ; GFX10-NEXT: s_cselect_b32 s12, 1, 0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0 | ; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0 | ||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s0, s[10:11], s[2:3] | ; GFX10-NEXT: v_cmp_lt_i64_e64 s0, s[10:11], s[2:3] | ||||
; GFX10-NEXT: v_cmp_gt_u64_e64 s2, s[4:5], 0 | ; GFX10-NEXT: v_cmp_gt_u64_e64 s2, s[4:5], 0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0 | ; GFX10-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0 | ||||
; GFX10-NEXT: s_and_b32 s0, 1, s12 | ; GFX10-NEXT: s_and_b32 s0, 1, s12 | ||||
; GFX10-NEXT: s_cmp_eq_u64 s[6:7], 0 | ; GFX10-NEXT: s_cmp_eq_u64 s[6:7], 0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, s2 | ||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s2, s[6:7], 0 | ; GFX10-NEXT: v_cmp_gt_i64_e64 s2, s[6:7], 0 | ||||
; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s0 | ; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s0 | ||||
; GFX10-NEXT: s_cselect_b32 s1, 1, 0 | ; GFX10-NEXT: s_cselect_b32 s1, 1, 0 | ||||
; GFX10-NEXT: s_ashr_i32 s0, s11, 31 | ; GFX10-NEXT: s_ashr_i32 s0, s11, 31 | ||||
; GFX10-NEXT: s_and_b32 s1, 1, s1 | ; GFX10-NEXT: s_and_b32 s1, 1, s1 | ||||
; GFX10-NEXT: s_mov_b32 s3, s0 | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v3, 0, 1, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v3, 0, 1, s2 | ||||
; GFX10-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo | ||||
; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s1 | ; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s1 | ||||
; GFX10-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX10-NEXT: s_add_u32 s1, s0, 0x80000000 | ||||
; GFX10-NEXT: s_mov_b32 s2, s0 | |||||
; GFX10-NEXT: v_cndmask_b32_e32 v1, v3, v2, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e32 v1, v3, v2, vcc_lo | ||||
; GFX10-NEXT: v_mov_b32_e32 v2, s9 | ; GFX10-NEXT: v_mov_b32_e32 v2, s9 | ||||
; GFX10-NEXT: v_mov_b32_e32 v3, s11 | ; GFX10-NEXT: v_mov_b32_e32 v3, s11 | ||||
; GFX10-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX10-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX10-NEXT: v_mov_b32_e32 v1, s8 | ; GFX10-NEXT: v_mov_b32_e32 v1, s8 | ||||
; GFX10-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX10-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX10-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 | ; GFX10-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 | ||||
; GFX10-NEXT: v_mov_b32_e32 v0, s10 | ; GFX10-NEXT: v_mov_b32_e32 v0, s10 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s0, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s0, vcc_lo | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v2, v2, s3, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e64 v2, v2, s0, vcc_lo | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s2, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s0, vcc_lo | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v3, v3, s1, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e64 v3, v3, s1, vcc_lo | ||||
; GFX10-NEXT: v_readfirstlane_b32 s0, v1 | ; GFX10-NEXT: v_readfirstlane_b32 s0, v1 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s1, v2 | ; GFX10-NEXT: v_readfirstlane_b32 s1, v2 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s2, v0 | ; GFX10-NEXT: v_readfirstlane_b32 s2, v0 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s3, v3 | ; GFX10-NEXT: v_readfirstlane_b32 s3, v3 | ||||
; GFX10-NEXT: ; return to shader part epilog | ; GFX10-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX11-LABEL: s_ssubsat_i128: | ; GFX11-LABEL: s_ssubsat_i128: | ||||
Show All 12 Lines | |||||
; GFX11-NEXT: s_and_b32 s0, 1, s12 | ; GFX11-NEXT: s_and_b32 s0, 1, s12 | ||||
; GFX11-NEXT: s_cmp_eq_u64 s[6:7], 0 | ; GFX11-NEXT: s_cmp_eq_u64 s[6:7], 0 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, s2 | ; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, s2 | ||||
; GFX11-NEXT: v_cmp_gt_i64_e64 s2, s[6:7], 0 | ; GFX11-NEXT: v_cmp_gt_i64_e64 s2, s[6:7], 0 | ||||
; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s0 | ; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s0 | ||||
; GFX11-NEXT: s_cselect_b32 s1, 1, 0 | ; GFX11-NEXT: s_cselect_b32 s1, 1, 0 | ||||
; GFX11-NEXT: s_ashr_i32 s0, s11, 31 | ; GFX11-NEXT: s_ashr_i32 s0, s11, 31 | ||||
; GFX11-NEXT: s_and_b32 s1, 1, s1 | ; GFX11-NEXT: s_and_b32 s1, 1, s1 | ||||
; GFX11-NEXT: s_mov_b32 s3, s0 | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s2 | ; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s2 | ||||
; GFX11-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo | ; GFX11-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo | ||||
; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s1 | ; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s1 | ||||
; GFX11-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX11-NEXT: s_add_u32 s1, s0, 0x80000000 | ||||
; GFX11-NEXT: s_mov_b32 s2, s0 | |||||
; GFX11-NEXT: v_dual_cndmask_b32 v1, v3, v2 :: v_dual_mov_b32 v2, s9 | ; GFX11-NEXT: v_dual_cndmask_b32 v1, v3, v2 :: v_dual_mov_b32 v2, s9 | ||||
; GFX11-NEXT: v_mov_b32_e32 v3, s11 | ; GFX11-NEXT: v_mov_b32_e32 v3, s11 | ||||
; GFX11-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX11-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX11-NEXT: v_dual_mov_b32 v1, s8 :: v_dual_and_b32 v0, 1, v0 | ; GFX11-NEXT: v_dual_mov_b32 v1, s8 :: v_dual_and_b32 v0, 1, v0 | ||||
; GFX11-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 | ; GFX11-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 | ||||
; GFX11-NEXT: v_mov_b32_e32 v0, s10 | ; GFX11-NEXT: v_mov_b32_e32 v0, s10 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s0, vcc_lo | ; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s0, vcc_lo | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v2, v2, s3, vcc_lo | ; GFX11-NEXT: v_cndmask_b32_e64 v2, v2, s0, vcc_lo | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s2, vcc_lo | ; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s0, vcc_lo | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v3, v3, s1, vcc_lo | ; GFX11-NEXT: v_cndmask_b32_e64 v3, v3, s1, vcc_lo | ||||
; GFX11-NEXT: v_readfirstlane_b32 s0, v1 | ; GFX11-NEXT: v_readfirstlane_b32 s0, v1 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s1, v2 | ; GFX11-NEXT: v_readfirstlane_b32 s1, v2 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s2, v0 | ; GFX11-NEXT: v_readfirstlane_b32 s2, v0 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s3, v3 | ; GFX11-NEXT: v_readfirstlane_b32 s3, v3 | ||||
; GFX11-NEXT: ; return to shader part epilog | ; GFX11-NEXT: ; return to shader part epilog | ||||
%result = call i128 @llvm.ssub.sat.i128(i128 %lhs, i128 %rhs) | %result = call i128 @llvm.ssub.sat.i128(i128 %lhs, i128 %rhs) | ||||
ret i128 %result | ret i128 %result | ||||
▲ Show 20 Lines • Show All 337 Lines • ▼ Show 20 Lines | |||||
; GFX6-LABEL: v_ssubsat_v2i128: | ; GFX6-LABEL: v_ssubsat_v2i128: | ||||
; GFX6: ; %bb.0: | ; GFX6: ; %bb.0: | ||||
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | ; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | ||||
; GFX6-NEXT: v_sub_i32_e32 v16, vcc, v0, v8 | ; GFX6-NEXT: v_sub_i32_e32 v16, vcc, v0, v8 | ||||
; GFX6-NEXT: v_subb_u32_e32 v17, vcc, v1, v9, vcc | ; GFX6-NEXT: v_subb_u32_e32 v17, vcc, v1, v9, vcc | ||||
; GFX6-NEXT: v_subb_u32_e32 v18, vcc, v2, v10, vcc | ; GFX6-NEXT: v_subb_u32_e32 v18, vcc, v2, v10, vcc | ||||
; GFX6-NEXT: v_subb_u32_e32 v19, vcc, v3, v11, vcc | ; GFX6-NEXT: v_subb_u32_e32 v19, vcc, v3, v11, vcc | ||||
; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, v[16:17], v[0:1] | ; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, v[16:17], v[0:1] | ||||
; GFX6-NEXT: v_bfrev_b32_e32 v20, 1 | |||||
; GFX6-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc | ; GFX6-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc | ||||
; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, v[18:19], v[2:3] | ; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, v[18:19], v[2:3] | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ; GFX6-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ||||
; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[18:19], v[2:3] | ; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[18:19], v[2:3] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc | ||||
; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[8:9] | ; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[8:9] | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ; GFX6-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ||||
; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[10:11] | ; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[10:11] | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc | ; GFX6-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc | ||||
; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[10:11] | ; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[10:11] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX6-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX6-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX6-NEXT: v_ashrrev_i32_e32 v2, 31, v19 | ; GFX6-NEXT: v_ashrrev_i32_e32 v2, 31, v19 | ||||
; GFX6-NEXT: v_add_i32_e32 v3, vcc, v2, v20 | ; GFX6-NEXT: v_bfrev_b32_e32 v1, 1 | ||||
; GFX6-NEXT: v_add_i32_e32 v3, vcc, v2, v1 | |||||
; GFX6-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX6-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v0, v16, v2, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v0, v16, v2, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v17, v2, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v17, v2, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v2, v18, v2, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v2, v18, v2, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v3, v19, v3, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v3, v19, v3, vcc | ||||
; GFX6-NEXT: v_sub_i32_e32 v8, vcc, v4, v12 | ; GFX6-NEXT: v_sub_i32_e32 v8, vcc, v4, v12 | ||||
; GFX6-NEXT: v_subb_u32_e32 v9, vcc, v5, v13, vcc | ; GFX6-NEXT: v_subb_u32_e32 v9, vcc, v5, v13, vcc | ||||
; GFX6-NEXT: v_subb_u32_e32 v10, vcc, v6, v14, vcc | ; GFX6-NEXT: v_subb_u32_e32 v10, vcc, v6, v14, vcc | ||||
; GFX6-NEXT: v_subb_u32_e32 v11, vcc, v7, v15, vcc | ; GFX6-NEXT: v_subb_u32_e32 v11, vcc, v7, v15, vcc | ||||
; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, v[8:9], v[4:5] | ; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, v[8:9], v[4:5] | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc | ; GFX6-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc | ||||
; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, v[10:11], v[6:7] | ; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, v[10:11], v[6:7] | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ; GFX6-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ||||
; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[10:11], v[6:7] | ; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[10:11], v[6:7] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc | ||||
; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[12:13] | ; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[12:13] | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ; GFX6-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ||||
; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[14:15] | ; GFX6-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[14:15] | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc | ; GFX6-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc | ||||
; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[14:15] | ; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[14:15] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v5, v6, v5, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v5, v6, v5, vcc | ||||
; GFX6-NEXT: v_xor_b32_e32 v4, v5, v4 | ; GFX6-NEXT: v_xor_b32_e32 v4, v5, v4 | ||||
; GFX6-NEXT: v_ashrrev_i32_e32 v6, 31, v11 | ; GFX6-NEXT: v_ashrrev_i32_e32 v6, 31, v11 | ||||
; GFX6-NEXT: v_add_i32_e32 v7, vcc, v6, v20 | ; GFX6-NEXT: v_add_i32_e32 v7, vcc, 0x80000000, v6 | ||||
; GFX6-NEXT: v_and_b32_e32 v4, 1, v4 | ; GFX6-NEXT: v_and_b32_e32 v4, 1, v4 | ||||
; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4 | ; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4 | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v4, v8, v6, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v4, v8, v6, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v5, v9, v6, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v5, v9, v6, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v6, v10, v6, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v6, v10, v6, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v7, v11, v7, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v7, v11, v7, vcc | ||||
; GFX6-NEXT: s_setpc_b64 s[30:31] | ; GFX6-NEXT: s_setpc_b64 s[30:31] | ||||
; | ; | ||||
; GFX8-LABEL: v_ssubsat_v2i128: | ; GFX8-LABEL: v_ssubsat_v2i128: | ||||
; GFX8: ; %bb.0: | ; GFX8: ; %bb.0: | ||||
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | ||||
; GFX8-NEXT: v_sub_u32_e32 v16, vcc, v0, v8 | ; GFX8-NEXT: v_sub_u32_e32 v16, vcc, v0, v8 | ||||
; GFX8-NEXT: v_subb_u32_e32 v17, vcc, v1, v9, vcc | ; GFX8-NEXT: v_subb_u32_e32 v17, vcc, v1, v9, vcc | ||||
; GFX8-NEXT: v_subb_u32_e32 v18, vcc, v2, v10, vcc | ; GFX8-NEXT: v_subb_u32_e32 v18, vcc, v2, v10, vcc | ||||
; GFX8-NEXT: v_subb_u32_e32 v19, vcc, v3, v11, vcc | ; GFX8-NEXT: v_subb_u32_e32 v19, vcc, v3, v11, vcc | ||||
; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, v[16:17], v[0:1] | ; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, v[16:17], v[0:1] | ||||
; GFX8-NEXT: v_bfrev_b32_e32 v20, 1 | |||||
; GFX8-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc | ; GFX8-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc | ||||
; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, v[18:19], v[2:3] | ; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, v[18:19], v[2:3] | ||||
; GFX8-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ; GFX8-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ||||
; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[18:19], v[2:3] | ; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[18:19], v[2:3] | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc | ||||
; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[8:9] | ; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[8:9] | ||||
; GFX8-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ; GFX8-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ||||
; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[10:11] | ; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[10:11] | ||||
; GFX8-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc | ; GFX8-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc | ||||
; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[10:11] | ; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[10:11] | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX8-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX8-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX8-NEXT: v_ashrrev_i32_e32 v2, 31, v19 | ; GFX8-NEXT: v_ashrrev_i32_e32 v2, 31, v19 | ||||
; GFX8-NEXT: v_add_u32_e32 v3, vcc, v2, v20 | ; GFX8-NEXT: v_bfrev_b32_e32 v1, 1 | ||||
; GFX8-NEXT: v_add_u32_e32 v3, vcc, v2, v1 | |||||
; GFX8-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX8-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v0, v16, v2, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v0, v16, v2, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v17, v2, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v17, v2, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v2, v18, v2, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v2, v18, v2, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v3, v19, v3, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v3, v19, v3, vcc | ||||
; GFX8-NEXT: v_sub_u32_e32 v8, vcc, v4, v12 | ; GFX8-NEXT: v_sub_u32_e32 v8, vcc, v4, v12 | ||||
; GFX8-NEXT: v_subb_u32_e32 v9, vcc, v5, v13, vcc | ; GFX8-NEXT: v_subb_u32_e32 v9, vcc, v5, v13, vcc | ||||
; GFX8-NEXT: v_subb_u32_e32 v10, vcc, v6, v14, vcc | ; GFX8-NEXT: v_subb_u32_e32 v10, vcc, v6, v14, vcc | ||||
; GFX8-NEXT: v_subb_u32_e32 v11, vcc, v7, v15, vcc | ; GFX8-NEXT: v_subb_u32_e32 v11, vcc, v7, v15, vcc | ||||
; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, v[8:9], v[4:5] | ; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, v[8:9], v[4:5] | ||||
; GFX8-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc | ; GFX8-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc | ||||
; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, v[10:11], v[6:7] | ; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, v[10:11], v[6:7] | ||||
; GFX8-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ; GFX8-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ||||
; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[10:11], v[6:7] | ; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[10:11], v[6:7] | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc | ||||
; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[12:13] | ; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[12:13] | ||||
; GFX8-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ; GFX8-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ||||
; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[14:15] | ; GFX8-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[14:15] | ||||
; GFX8-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc | ; GFX8-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc | ||||
; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[14:15] | ; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[14:15] | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v5, v6, v5, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v5, v6, v5, vcc | ||||
; GFX8-NEXT: v_xor_b32_e32 v4, v5, v4 | ; GFX8-NEXT: v_xor_b32_e32 v4, v5, v4 | ||||
; GFX8-NEXT: v_ashrrev_i32_e32 v6, 31, v11 | ; GFX8-NEXT: v_ashrrev_i32_e32 v6, 31, v11 | ||||
; GFX8-NEXT: v_add_u32_e32 v7, vcc, v6, v20 | ; GFX8-NEXT: v_add_u32_e32 v7, vcc, 0x80000000, v6 | ||||
; GFX8-NEXT: v_and_b32_e32 v4, 1, v4 | ; GFX8-NEXT: v_and_b32_e32 v4, 1, v4 | ||||
; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4 | ; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v4, v8, v6, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v4, v8, v6, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v5, v9, v6, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v5, v9, v6, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v6, v10, v6, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v6, v10, v6, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v7, v11, v7, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v7, v11, v7, vcc | ||||
; GFX8-NEXT: s_setpc_b64 s[30:31] | ; GFX8-NEXT: s_setpc_b64 s[30:31] | ||||
; | ; | ||||
; GFX9-LABEL: v_ssubsat_v2i128: | ; GFX9-LABEL: v_ssubsat_v2i128: | ||||
; GFX9: ; %bb.0: | ; GFX9: ; %bb.0: | ||||
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | ||||
; GFX9-NEXT: v_sub_co_u32_e32 v16, vcc, v0, v8 | ; GFX9-NEXT: v_sub_co_u32_e32 v16, vcc, v0, v8 | ||||
; GFX9-NEXT: v_subb_co_u32_e32 v17, vcc, v1, v9, vcc | ; GFX9-NEXT: v_subb_co_u32_e32 v17, vcc, v1, v9, vcc | ||||
; GFX9-NEXT: v_subb_co_u32_e32 v18, vcc, v2, v10, vcc | ; GFX9-NEXT: v_subb_co_u32_e32 v18, vcc, v2, v10, vcc | ||||
; GFX9-NEXT: v_subb_co_u32_e32 v19, vcc, v3, v11, vcc | ; GFX9-NEXT: v_subb_co_u32_e32 v19, vcc, v3, v11, vcc | ||||
; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, v[16:17], v[0:1] | ; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, v[16:17], v[0:1] | ||||
; GFX9-NEXT: v_bfrev_b32_e32 v20, 1 | |||||
; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc | ; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc | ||||
; GFX9-NEXT: v_cmp_lt_i64_e32 vcc, v[18:19], v[2:3] | ; GFX9-NEXT: v_cmp_lt_i64_e32 vcc, v[18:19], v[2:3] | ||||
; GFX9-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ; GFX9-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ||||
; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[18:19], v[2:3] | ; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[18:19], v[2:3] | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc | ||||
; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[8:9] | ; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[8:9] | ||||
; GFX9-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ; GFX9-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc | ||||
; GFX9-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[10:11] | ; GFX9-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[10:11] | ||||
; GFX9-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc | ; GFX9-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc | ||||
; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[10:11] | ; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[10:11] | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX9-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX9-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX9-NEXT: v_ashrrev_i32_e32 v2, 31, v19 | ; GFX9-NEXT: v_ashrrev_i32_e32 v2, 31, v19 | ||||
; GFX9-NEXT: v_add_co_u32_e32 v3, vcc, v2, v20 | ; GFX9-NEXT: v_bfrev_b32_e32 v1, 1 | ||||
; GFX9-NEXT: v_add_co_u32_e32 v3, vcc, v2, v1 | |||||
; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v0, v16, v2, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v0, v16, v2, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v17, v2, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v17, v2, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v2, v18, v2, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v2, v18, v2, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v3, v19, v3, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v3, v19, v3, vcc | ||||
; GFX9-NEXT: v_sub_co_u32_e32 v8, vcc, v4, v12 | ; GFX9-NEXT: v_sub_co_u32_e32 v8, vcc, v4, v12 | ||||
; GFX9-NEXT: v_subb_co_u32_e32 v9, vcc, v5, v13, vcc | ; GFX9-NEXT: v_subb_co_u32_e32 v9, vcc, v5, v13, vcc | ||||
; GFX9-NEXT: v_subb_co_u32_e32 v10, vcc, v6, v14, vcc | ; GFX9-NEXT: v_subb_co_u32_e32 v10, vcc, v6, v14, vcc | ||||
; GFX9-NEXT: v_subb_co_u32_e32 v11, vcc, v7, v15, vcc | ; GFX9-NEXT: v_subb_co_u32_e32 v11, vcc, v7, v15, vcc | ||||
; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, v[8:9], v[4:5] | ; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, v[8:9], v[4:5] | ||||
; GFX9-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc | ; GFX9-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc | ||||
; GFX9-NEXT: v_cmp_lt_i64_e32 vcc, v[10:11], v[6:7] | ; GFX9-NEXT: v_cmp_lt_i64_e32 vcc, v[10:11], v[6:7] | ||||
; GFX9-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ; GFX9-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ||||
; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[10:11], v[6:7] | ; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[10:11], v[6:7] | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc | ||||
; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[12:13] | ; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, 0, v[12:13] | ||||
; GFX9-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ; GFX9-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc | ||||
; GFX9-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[14:15] | ; GFX9-NEXT: v_cmp_lt_i64_e32 vcc, 0, v[14:15] | ||||
; GFX9-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc | ; GFX9-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc | ||||
; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[14:15] | ; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[14:15] | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v5, v6, v5, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v5, v6, v5, vcc | ||||
; GFX9-NEXT: v_xor_b32_e32 v4, v5, v4 | ; GFX9-NEXT: v_xor_b32_e32 v4, v5, v4 | ||||
; GFX9-NEXT: v_ashrrev_i32_e32 v6, 31, v11 | ; GFX9-NEXT: v_ashrrev_i32_e32 v6, 31, v11 | ||||
; GFX9-NEXT: v_add_co_u32_e32 v7, vcc, v6, v20 | ; GFX9-NEXT: v_add_co_u32_e32 v7, vcc, 0x80000000, v6 | ||||
; GFX9-NEXT: v_and_b32_e32 v4, 1, v4 | ; GFX9-NEXT: v_and_b32_e32 v4, 1, v4 | ||||
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4 | ; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v4, v8, v6, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v4, v8, v6, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v5, v9, v6, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v5, v9, v6, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v6, v10, v6, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v6, v10, v6, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v7, v11, v7, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v7, v11, v7, vcc | ||||
; GFX9-NEXT: s_setpc_b64 s[30:31] | ; GFX9-NEXT: s_setpc_b64 s[30:31] | ||||
; | ; | ||||
▲ Show 20 Lines • Show All 134 Lines • ▼ Show 20 Lines | |||||
; GFX6-NEXT: v_cmp_eq_u64_e64 vcc, s[10:11], 0 | ; GFX6-NEXT: v_cmp_eq_u64_e64 vcc, s[10:11], 0 | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[0:1] | ; GFX6-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[0:1] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX6-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX6-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX6-NEXT: s_ashr_i32 s0, s19, 31 | ; GFX6-NEXT: s_ashr_i32 s0, s19, 31 | ||||
; GFX6-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX6-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX6-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX6-NEXT: s_add_u32 s1, s0, 0x80000000 | ||||
; GFX6-NEXT: v_mov_b32_e32 v1, s0 | ; GFX6-NEXT: v_mov_b32_e32 v1, s0 | ||||
; GFX6-NEXT: v_mov_b32_e32 v2, s0 | ; GFX6-NEXT: v_mov_b32_e32 v2, s16 | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s16 | ; GFX6-NEXT: v_mov_b32_e32 v3, s17 | ||||
; GFX6-NEXT: v_mov_b32_e32 v4, s17 | |||||
; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v4, v2, v1, vcc | |||||
; GFX6-NEXT: v_cndmask_b32_e32 v5, v3, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v5, v3, v1, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v4, v4, v2, vcc | ; GFX6-NEXT: v_mov_b32_e32 v0, s1 | ||||
; GFX6-NEXT: v_mov_b32_e32 v0, s0 | |||||
; GFX6-NEXT: v_mov_b32_e32 v1, s1 | |||||
; GFX6-NEXT: v_mov_b32_e32 v2, s18 | ; GFX6-NEXT: v_mov_b32_e32 v2, s18 | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s19 | ; GFX6-NEXT: v_mov_b32_e32 v3, s19 | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v6, v2, v0, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v6, v2, v1, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v7, v3, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v7, v3, v0, vcc | ||||
; GFX6-NEXT: s_sub_u32 s0, s4, s12 | ; GFX6-NEXT: s_sub_u32 s0, s4, s12 | ||||
; GFX6-NEXT: v_mov_b32_e32 v2, s4 | ; GFX6-NEXT: v_mov_b32_e32 v2, s4 | ||||
; GFX6-NEXT: s_subb_u32 s1, s5, s13 | ; GFX6-NEXT: s_subb_u32 s1, s5, s13 | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s5 | ; GFX6-NEXT: v_mov_b32_e32 v3, s5 | ||||
; GFX6-NEXT: s_subb_u32 s2, s6, s14 | ; GFX6-NEXT: s_subb_u32 s2, s6, s14 | ||||
; GFX6-NEXT: v_mov_b32_e32 v0, s6 | ; GFX6-NEXT: v_mov_b32_e32 v0, s6 | ||||
; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[2:3] | ; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[2:3] | ||||
; GFX6-NEXT: s_subb_u32 s3, s7, s15 | ; GFX6-NEXT: s_subb_u32 s3, s7, s15 | ||||
Show All 9 Lines | |||||
; GFX6-NEXT: v_cmp_eq_u64_e64 vcc, s[14:15], 0 | ; GFX6-NEXT: v_cmp_eq_u64_e64 vcc, s[14:15], 0 | ||||
; GFX6-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[4:5] | ; GFX6-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[4:5] | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX6-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX6-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX6-NEXT: s_ashr_i32 s4, s3, 31 | ; GFX6-NEXT: s_ashr_i32 s4, s3, 31 | ||||
; GFX6-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX6-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX6-NEXT: s_add_u32 s5, s4, 0x80000000 | ; GFX6-NEXT: s_add_u32 s5, s4, 0x80000000 | ||||
; GFX6-NEXT: v_mov_b32_e32 v1, s4 | ; GFX6-NEXT: v_mov_b32_e32 v1, s4 | ||||
; GFX6-NEXT: v_mov_b32_e32 v2, s4 | ; GFX6-NEXT: v_mov_b32_e32 v2, s0 | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s0 | ; GFX6-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX6-NEXT: v_mov_b32_e32 v8, s1 | |||||
; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX6-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v0, v3, v1, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v1, v8, v2, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v2, v3, v1, vcc | ||||
; GFX6-NEXT: v_mov_b32_e32 v3, s5 | ; GFX6-NEXT: v_mov_b32_e32 v3, s5 | ||||
; GFX6-NEXT: v_mov_b32_e32 v8, s2 | ; GFX6-NEXT: v_mov_b32_e32 v8, s2 | ||||
; GFX6-NEXT: v_mov_b32_e32 v9, s3 | ; GFX6-NEXT: v_mov_b32_e32 v9, s3 | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v2, v8, v2, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v1, v8, v1, vcc | ||||
; GFX6-NEXT: v_cndmask_b32_e32 v3, v9, v3, vcc | ; GFX6-NEXT: v_cndmask_b32_e32 v3, v9, v3, vcc | ||||
; GFX6-NEXT: v_readfirstlane_b32 s0, v5 | ; GFX6-NEXT: v_readfirstlane_b32 s0, v4 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s1, v4 | ; GFX6-NEXT: v_readfirstlane_b32 s1, v5 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s2, v6 | ; GFX6-NEXT: v_readfirstlane_b32 s2, v6 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s3, v7 | ; GFX6-NEXT: v_readfirstlane_b32 s3, v7 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s4, v0 | ; GFX6-NEXT: v_readfirstlane_b32 s4, v0 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s5, v1 | ; GFX6-NEXT: v_readfirstlane_b32 s5, v2 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s6, v2 | ; GFX6-NEXT: v_readfirstlane_b32 s6, v1 | ||||
; GFX6-NEXT: v_readfirstlane_b32 s7, v3 | ; GFX6-NEXT: v_readfirstlane_b32 s7, v3 | ||||
; GFX6-NEXT: ; return to shader part epilog | ; GFX6-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX8-LABEL: s_ssubsat_v2i128: | ; GFX8-LABEL: s_ssubsat_v2i128: | ||||
; GFX8: ; %bb.0: | ; GFX8: ; %bb.0: | ||||
; GFX8-NEXT: s_sub_u32 s16, s0, s8 | ; GFX8-NEXT: s_sub_u32 s16, s0, s8 | ||||
; GFX8-NEXT: s_subb_u32 s17, s1, s9 | ; GFX8-NEXT: s_subb_u32 s17, s1, s9 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ||||
Show All 20 Lines | |||||
; GFX8-NEXT: s_and_b32 s0, 1, s2 | ; GFX8-NEXT: s_and_b32 s0, 1, s2 | ||||
; GFX8-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 | ; GFX8-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX8-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX8-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX8-NEXT: s_ashr_i32 s0, s19, 31 | ; GFX8-NEXT: s_ashr_i32 s0, s19, 31 | ||||
; GFX8-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX8-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX8-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX8-NEXT: s_add_u32 s1, s0, 0x80000000 | ||||
; GFX8-NEXT: v_mov_b32_e32 v1, s0 | ; GFX8-NEXT: v_mov_b32_e32 v1, s0 | ||||
; GFX8-NEXT: v_mov_b32_e32 v2, s0 | ; GFX8-NEXT: v_mov_b32_e32 v2, s16 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s16 | ; GFX8-NEXT: v_mov_b32_e32 v3, s17 | ||||
; GFX8-NEXT: v_mov_b32_e32 v4, s17 | |||||
; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v4, v2, v1, vcc | |||||
; GFX8-NEXT: v_cndmask_b32_e32 v5, v3, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v5, v3, v1, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v4, v4, v2, vcc | ; GFX8-NEXT: v_mov_b32_e32 v0, s1 | ||||
; GFX8-NEXT: v_mov_b32_e32 v0, s0 | |||||
; GFX8-NEXT: v_mov_b32_e32 v1, s1 | |||||
; GFX8-NEXT: v_mov_b32_e32 v2, s18 | ; GFX8-NEXT: v_mov_b32_e32 v2, s18 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s19 | ; GFX8-NEXT: v_mov_b32_e32 v3, s19 | ||||
; GFX8-NEXT: s_sub_u32 s0, s4, s12 | ; GFX8-NEXT: s_sub_u32 s0, s4, s12 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v6, v2, v0, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v6, v2, v1, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v7, v3, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v7, v3, v0, vcc | ||||
; GFX8-NEXT: s_subb_u32 s1, s5, s13 | ; GFX8-NEXT: s_subb_u32 s1, s5, s13 | ||||
; GFX8-NEXT: v_mov_b32_e32 v2, s4 | ; GFX8-NEXT: v_mov_b32_e32 v2, s4 | ||||
; GFX8-NEXT: s_subb_u32 s2, s6, s14 | ; GFX8-NEXT: s_subb_u32 s2, s6, s14 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s5 | ; GFX8-NEXT: v_mov_b32_e32 v3, s5 | ||||
; GFX8-NEXT: s_subb_u32 s3, s7, s15 | ; GFX8-NEXT: s_subb_u32 s3, s7, s15 | ||||
; GFX8-NEXT: v_mov_b32_e32 v0, s6 | ; GFX8-NEXT: v_mov_b32_e32 v0, s6 | ||||
; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[2:3] | ; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[2:3] | ||||
; GFX8-NEXT: v_mov_b32_e32 v1, s7 | ; GFX8-NEXT: v_mov_b32_e32 v1, s7 | ||||
Show All 14 Lines | |||||
; GFX8-NEXT: s_and_b32 s4, 1, s6 | ; GFX8-NEXT: s_and_b32 s4, 1, s6 | ||||
; GFX8-NEXT: v_cmp_ne_u32_e64 vcc, 0, s4 | ; GFX8-NEXT: v_cmp_ne_u32_e64 vcc, 0, s4 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX8-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX8-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX8-NEXT: s_ashr_i32 s4, s3, 31 | ; GFX8-NEXT: s_ashr_i32 s4, s3, 31 | ||||
; GFX8-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX8-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX8-NEXT: s_add_u32 s5, s4, 0x80000000 | ; GFX8-NEXT: s_add_u32 s5, s4, 0x80000000 | ||||
; GFX8-NEXT: v_mov_b32_e32 v1, s4 | ; GFX8-NEXT: v_mov_b32_e32 v1, s4 | ||||
; GFX8-NEXT: v_mov_b32_e32 v2, s4 | ; GFX8-NEXT: v_mov_b32_e32 v2, s0 | ||||
; GFX8-NEXT: v_mov_b32_e32 v3, s0 | ; GFX8-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX8-NEXT: v_mov_b32_e32 v8, s1 | |||||
; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX8-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v0, v3, v1, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v1, v8, v2, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v2, v3, v1, vcc | ||||
; GFX8-NEXT: v_mov_b32_e32 v2, s4 | |||||
; GFX8-NEXT: v_mov_b32_e32 v3, s5 | ; GFX8-NEXT: v_mov_b32_e32 v3, s5 | ||||
; GFX8-NEXT: v_mov_b32_e32 v8, s2 | ; GFX8-NEXT: v_mov_b32_e32 v8, s2 | ||||
; GFX8-NEXT: v_mov_b32_e32 v9, s3 | ; GFX8-NEXT: v_mov_b32_e32 v9, s3 | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v2, v8, v2, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v1, v8, v1, vcc | ||||
; GFX8-NEXT: v_cndmask_b32_e32 v3, v9, v3, vcc | ; GFX8-NEXT: v_cndmask_b32_e32 v3, v9, v3, vcc | ||||
; GFX8-NEXT: v_readfirstlane_b32 s0, v5 | ; GFX8-NEXT: v_readfirstlane_b32 s0, v4 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s1, v4 | ; GFX8-NEXT: v_readfirstlane_b32 s1, v5 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s2, v6 | ; GFX8-NEXT: v_readfirstlane_b32 s2, v6 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s3, v7 | ; GFX8-NEXT: v_readfirstlane_b32 s3, v7 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s4, v0 | ; GFX8-NEXT: v_readfirstlane_b32 s4, v0 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s5, v1 | ; GFX8-NEXT: v_readfirstlane_b32 s5, v2 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s6, v2 | ; GFX8-NEXT: v_readfirstlane_b32 s6, v1 | ||||
; GFX8-NEXT: v_readfirstlane_b32 s7, v3 | ; GFX8-NEXT: v_readfirstlane_b32 s7, v3 | ||||
; GFX8-NEXT: ; return to shader part epilog | ; GFX8-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX9-LABEL: s_ssubsat_v2i128: | ; GFX9-LABEL: s_ssubsat_v2i128: | ||||
; GFX9: ; %bb.0: | ; GFX9: ; %bb.0: | ||||
; GFX9-NEXT: s_sub_u32 s16, s0, s8 | ; GFX9-NEXT: s_sub_u32 s16, s0, s8 | ||||
; GFX9-NEXT: s_subb_u32 s17, s1, s9 | ; GFX9-NEXT: s_subb_u32 s17, s1, s9 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ||||
Show All 20 Lines | |||||
; GFX9-NEXT: s_and_b32 s0, 1, s2 | ; GFX9-NEXT: s_and_b32 s0, 1, s2 | ||||
; GFX9-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 | ; GFX9-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX9-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX9-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX9-NEXT: s_ashr_i32 s0, s19, 31 | ; GFX9-NEXT: s_ashr_i32 s0, s19, 31 | ||||
; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX9-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX9-NEXT: s_add_u32 s1, s0, 0x80000000 | ||||
; GFX9-NEXT: v_mov_b32_e32 v1, s0 | ; GFX9-NEXT: v_mov_b32_e32 v1, s0 | ||||
; GFX9-NEXT: v_mov_b32_e32 v2, s0 | ; GFX9-NEXT: v_mov_b32_e32 v2, s16 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s16 | ; GFX9-NEXT: v_mov_b32_e32 v3, s17 | ||||
; GFX9-NEXT: v_mov_b32_e32 v4, s17 | |||||
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v4, v2, v1, vcc | |||||
; GFX9-NEXT: v_cndmask_b32_e32 v5, v3, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v5, v3, v1, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v2, vcc | ; GFX9-NEXT: v_mov_b32_e32 v0, s1 | ||||
; GFX9-NEXT: v_mov_b32_e32 v0, s0 | |||||
; GFX9-NEXT: v_mov_b32_e32 v1, s1 | |||||
; GFX9-NEXT: v_mov_b32_e32 v2, s18 | ; GFX9-NEXT: v_mov_b32_e32 v2, s18 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s19 | ; GFX9-NEXT: v_mov_b32_e32 v3, s19 | ||||
; GFX9-NEXT: s_sub_u32 s0, s4, s12 | ; GFX9-NEXT: s_sub_u32 s0, s4, s12 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v6, v2, v0, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v6, v2, v1, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v7, v3, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v7, v3, v0, vcc | ||||
; GFX9-NEXT: s_subb_u32 s1, s5, s13 | ; GFX9-NEXT: s_subb_u32 s1, s5, s13 | ||||
; GFX9-NEXT: v_mov_b32_e32 v2, s4 | ; GFX9-NEXT: v_mov_b32_e32 v2, s4 | ||||
; GFX9-NEXT: s_subb_u32 s2, s6, s14 | ; GFX9-NEXT: s_subb_u32 s2, s6, s14 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s5 | ; GFX9-NEXT: v_mov_b32_e32 v3, s5 | ||||
; GFX9-NEXT: s_subb_u32 s3, s7, s15 | ; GFX9-NEXT: s_subb_u32 s3, s7, s15 | ||||
; GFX9-NEXT: v_mov_b32_e32 v0, s6 | ; GFX9-NEXT: v_mov_b32_e32 v0, s6 | ||||
; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[2:3] | ; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[2:3] | ||||
; GFX9-NEXT: v_mov_b32_e32 v1, s7 | ; GFX9-NEXT: v_mov_b32_e32 v1, s7 | ||||
Show All 14 Lines | |||||
; GFX9-NEXT: s_and_b32 s4, 1, s6 | ; GFX9-NEXT: s_and_b32 s4, 1, s6 | ||||
; GFX9-NEXT: v_cmp_ne_u32_e64 vcc, 0, s4 | ; GFX9-NEXT: v_cmp_ne_u32_e64 vcc, 0, s4 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc | ||||
; GFX9-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX9-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX9-NEXT: s_ashr_i32 s4, s3, 31 | ; GFX9-NEXT: s_ashr_i32 s4, s3, 31 | ||||
; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX9-NEXT: s_add_u32 s5, s4, 0x80000000 | ; GFX9-NEXT: s_add_u32 s5, s4, 0x80000000 | ||||
; GFX9-NEXT: v_mov_b32_e32 v1, s4 | ; GFX9-NEXT: v_mov_b32_e32 v1, s4 | ||||
; GFX9-NEXT: v_mov_b32_e32 v2, s4 | ; GFX9-NEXT: v_mov_b32_e32 v2, s0 | ||||
; GFX9-NEXT: v_mov_b32_e32 v3, s0 | ; GFX9-NEXT: v_mov_b32_e32 v3, s1 | ||||
; GFX9-NEXT: v_mov_b32_e32 v8, s1 | |||||
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v1, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v1, v8, v2, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v2, v3, v1, vcc | ||||
; GFX9-NEXT: v_mov_b32_e32 v2, s4 | |||||
; GFX9-NEXT: v_mov_b32_e32 v3, s5 | ; GFX9-NEXT: v_mov_b32_e32 v3, s5 | ||||
; GFX9-NEXT: v_mov_b32_e32 v8, s2 | ; GFX9-NEXT: v_mov_b32_e32 v8, s2 | ||||
; GFX9-NEXT: v_mov_b32_e32 v9, s3 | ; GFX9-NEXT: v_mov_b32_e32 v9, s3 | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v2, v8, v2, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v1, v8, v1, vcc | ||||
; GFX9-NEXT: v_cndmask_b32_e32 v3, v9, v3, vcc | ; GFX9-NEXT: v_cndmask_b32_e32 v3, v9, v3, vcc | ||||
; GFX9-NEXT: v_readfirstlane_b32 s0, v5 | ; GFX9-NEXT: v_readfirstlane_b32 s0, v4 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s1, v4 | ; GFX9-NEXT: v_readfirstlane_b32 s1, v5 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s2, v6 | ; GFX9-NEXT: v_readfirstlane_b32 s2, v6 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s3, v7 | ; GFX9-NEXT: v_readfirstlane_b32 s3, v7 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s4, v0 | ; GFX9-NEXT: v_readfirstlane_b32 s4, v0 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s5, v1 | ; GFX9-NEXT: v_readfirstlane_b32 s5, v2 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s6, v2 | ; GFX9-NEXT: v_readfirstlane_b32 s6, v1 | ||||
; GFX9-NEXT: v_readfirstlane_b32 s7, v3 | ; GFX9-NEXT: v_readfirstlane_b32 s7, v3 | ||||
; GFX9-NEXT: ; return to shader part epilog | ; GFX9-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX10-LABEL: s_ssubsat_v2i128: | ; GFX10-LABEL: s_ssubsat_v2i128: | ||||
; GFX10: ; %bb.0: | ; GFX10: ; %bb.0: | ||||
; GFX10-NEXT: s_sub_u32 s16, s0, s8 | ; GFX10-NEXT: s_sub_u32 s16, s0, s8 | ||||
; GFX10-NEXT: s_subb_u32 s17, s1, s9 | ; GFX10-NEXT: s_subb_u32 s17, s1, s9 | ||||
; GFX10-NEXT: s_subb_u32 s18, s2, s10 | ; GFX10-NEXT: s_subb_u32 s18, s2, s10 | ||||
; GFX10-NEXT: v_cmp_lt_u64_e64 s0, s[16:17], s[0:1] | ; GFX10-NEXT: v_cmp_lt_u64_e64 s0, s[16:17], s[0:1] | ||||
; GFX10-NEXT: s_subb_u32 s19, s3, s11 | ; GFX10-NEXT: s_subb_u32 s19, s3, s11 | ||||
; GFX10-NEXT: s_cmp_eq_u64 s[18:19], s[2:3] | ; GFX10-NEXT: s_cmp_eq_u64 s[18:19], s[2:3] | ||||
; GFX10-NEXT: s_cselect_b32 s20, 1, 0 | ; GFX10-NEXT: s_cselect_b32 s20, 1, 0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0 | ; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0 | ||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s0, s[18:19], s[2:3] | ; GFX10-NEXT: v_cmp_lt_i64_e64 s0, s[18:19], s[2:3] | ||||
; GFX10-NEXT: v_cmp_gt_u64_e64 s2, s[8:9], 0 | ; GFX10-NEXT: v_cmp_gt_u64_e64 s2, s[8:9], 0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0 | ; GFX10-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0 | ||||
; GFX10-NEXT: s_and_b32 s0, 1, s20 | ; GFX10-NEXT: s_and_b32 s0, 1, s20 | ||||
; GFX10-NEXT: s_cmp_eq_u64 s[10:11], 0 | ; GFX10-NEXT: s_cmp_eq_u64 s[10:11], 0 | ||||
; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s0 | |||||
; GFX10-NEXT: s_cselect_b32 s1, 1, 0 | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, s2 | ; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, s2 | ||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s2, s[10:11], 0 | ; GFX10-NEXT: v_cmp_gt_i64_e64 s2, s[10:11], 0 | ||||
; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s0 | |||||
; GFX10-NEXT: s_cselect_b32 s1, 1, 0 | |||||
; GFX10-NEXT: s_ashr_i32 s8, s19, 31 | |||||
; GFX10-NEXT: s_and_b32 s1, 1, s1 | ; GFX10-NEXT: s_and_b32 s1, 1, s1 | ||||
; GFX10-NEXT: s_ashr_i32 s0, s19, 31 | ; GFX10-NEXT: s_add_u32 s9, s8, 0x80000000 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v3, 0, 1, s2 | |||||
; GFX10-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo | ||||
; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s1 | ; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s1 | ||||
; GFX10-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX10-NEXT: s_sub_u32 s0, s4, s12 | ||||
; GFX10-NEXT: s_sub_u32 s8, s4, s12 | ; GFX10-NEXT: s_subb_u32 s1, s5, s13 | ||||
; GFX10-NEXT: s_subb_u32 s9, s5, s13 | ; GFX10-NEXT: s_subb_u32 s2, s6, s14 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v3, 0, 1, s2 | ; GFX10-NEXT: v_cmp_lt_u64_e64 s4, s[0:1], s[4:5] | ||||
; GFX10-NEXT: v_cmp_lt_u64_e64 s4, s[8:9], s[4:5] | |||||
; GFX10-NEXT: s_subb_u32 s10, s6, s14 | |||||
; GFX10-NEXT: s_subb_u32 s11, s7, s15 | |||||
; GFX10-NEXT: s_mov_b32 s3, s0 | |||||
; GFX10-NEXT: v_cndmask_b32_e32 v1, v3, v2, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e32 v1, v3, v2, vcc_lo | ||||
; GFX10-NEXT: s_cmp_eq_u64 s[10:11], s[6:7] | ; GFX10-NEXT: s_subb_u32 s3, s7, s15 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 | ; GFX10-NEXT: v_mov_b32_e32 v5, s0 | ||||
; GFX10-NEXT: v_cmp_lt_i64_e64 s4, s[10:11], s[6:7] | ; GFX10-NEXT: s_cmp_eq_u64 s[2:3], s[6:7] | ||||
; GFX10-NEXT: v_cmp_gt_u64_e64 s6, s[12:13], 0 | ; GFX10-NEXT: v_mov_b32_e32 v6, s1 | ||||
; GFX10-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX10-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX10-NEXT: v_mov_b32_e32 v1, s16 | ; GFX10-NEXT: v_cndmask_b32_e64 v1, 0, 1, s4 | ||||
; GFX10-NEXT: s_cselect_b32 s16, 1, 0 | ; GFX10-NEXT: v_cmp_lt_i64_e64 s4, s[2:3], s[6:7] | ||||
; GFX10-NEXT: v_mov_b32_e32 v2, s17 | ; GFX10-NEXT: v_cmp_gt_u64_e64 s6, s[12:13], 0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v4, 0, 1, s4 | ; GFX10-NEXT: s_cselect_b32 s10, 1, 0 | ||||
; GFX10-NEXT: s_and_b32 s4, 1, s16 | ; GFX10-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX10-NEXT: v_mov_b32_e32 v7, s3 | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, s4 | |||||
; GFX10-NEXT: s_and_b32 s4, 1, s10 | |||||
; GFX10-NEXT: s_cmp_eq_u64 s[14:15], 0 | ; GFX10-NEXT: s_cmp_eq_u64 s[14:15], 0 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v5, 0, 1, s6 | ; GFX10-NEXT: v_cndmask_b32_e64 v3, 0, 1, s6 | ||||
; GFX10-NEXT: v_cmp_gt_i64_e64 s6, s[14:15], 0 | ; GFX10-NEXT: v_cmp_gt_i64_e64 s6, s[14:15], 0 | ||||
; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s4 | ; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s4 | ||||
; GFX10-NEXT: s_cselect_b32 s5, 1, 0 | ; GFX10-NEXT: s_cselect_b32 s5, 1, 0 | ||||
; GFX10-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX10-NEXT: s_ashr_i32 s4, s3, 31 | ||||
; GFX10-NEXT: s_and_b32 s5, 1, s5 | ; GFX10-NEXT: s_and_b32 s5, 1, s5 | ||||
; GFX10-NEXT: s_mov_b32 s2, s0 | ; GFX10-NEXT: s_add_u32 s0, s4, 0x80000000 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v6, 0, 1, s6 | ; GFX10-NEXT: v_cndmask_b32_e64 v4, 0, 1, s6 | ||||
; GFX10-NEXT: v_cndmask_b32_e32 v3, v4, v3, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc_lo | ||||
; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s5 | ; GFX10-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s5 | ||||
; GFX10-NEXT: v_mov_b32_e32 v7, s11 | ; GFX10-NEXT: v_cndmask_b32_e32 v2, v4, v3, vcc_lo | ||||
; GFX10-NEXT: v_cndmask_b32_e32 v4, v6, v5, vcc_lo | ; GFX10-NEXT: v_mov_b32_e32 v3, s16 | ||||
; GFX10-NEXT: v_mov_b32_e32 v4, s17 | |||||
; GFX10-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 | ; GFX10-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 | ||||
; GFX10-NEXT: v_mov_b32_e32 v0, s18 | ; GFX10-NEXT: v_mov_b32_e32 v0, s18 | ||||
; GFX10-NEXT: v_mov_b32_e32 v5, s19 | ; GFX10-NEXT: v_xor_b32_e32 v1, v2, v1 | ||||
; GFX10-NEXT: v_mov_b32_e32 v6, s9 | ; GFX10-NEXT: v_mov_b32_e32 v2, s19 | ||||
; GFX10-NEXT: v_xor_b32_e32 v3, v4, v3 | ; GFX10-NEXT: v_cndmask_b32_e64 v3, v3, s8, vcc_lo | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s0, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e64 v4, v4, s8, vcc_lo | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v2, v2, s3, vcc_lo | ; GFX10-NEXT: v_and_b32_e32 v1, 1, v1 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s2, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, s8, vcc_lo | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v4, v5, s1, vcc_lo | ; GFX10-NEXT: v_cndmask_b32_e64 v2, v2, s9, vcc_lo | ||||
; GFX10-NEXT: v_and_b32_e32 v3, 1, v3 | ; GFX10-NEXT: v_readfirstlane_b32 s1, v4 | ||||
; GFX10-NEXT: v_mov_b32_e32 v5, s8 | ; GFX10-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 | ||||
; GFX10-NEXT: s_ashr_i32 s0, s11, 31 | ; GFX10-NEXT: v_mov_b32_e32 v1, s2 | ||||
; GFX10-NEXT: s_add_u32 s1, s0, 0x80000000 | |||||
; GFX10-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v3 | |||||
; GFX10-NEXT: v_mov_b32_e32 v3, s10 | |||||
; GFX10-NEXT: s_mov_b32 s3, s0 | |||||
; GFX10-NEXT: s_mov_b32 s2, s0 | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v5, v5, s0, vcc_lo | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v6, v6, s3, vcc_lo | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v3, v3, s2, vcc_lo | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v7, v7, s1, vcc_lo | |||||
; GFX10-NEXT: v_readfirstlane_b32 s0, v1 | |||||
; GFX10-NEXT: v_readfirstlane_b32 s1, v2 | |||||
; GFX10-NEXT: v_readfirstlane_b32 s2, v0 | ; GFX10-NEXT: v_readfirstlane_b32 s2, v0 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s3, v4 | ; GFX10-NEXT: v_readfirstlane_b32 s3, v2 | ||||
; GFX10-NEXT: v_cndmask_b32_e64 v5, v5, s4, vcc_lo | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v6, v6, s4, vcc_lo | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, s4, vcc_lo | |||||
; GFX10-NEXT: v_cndmask_b32_e64 v7, v7, s0, vcc_lo | |||||
; GFX10-NEXT: v_readfirstlane_b32 s0, v3 | |||||
; GFX10-NEXT: v_readfirstlane_b32 s4, v5 | ; GFX10-NEXT: v_readfirstlane_b32 s4, v5 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s5, v6 | ; GFX10-NEXT: v_readfirstlane_b32 s5, v6 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s6, v3 | ; GFX10-NEXT: v_readfirstlane_b32 s6, v1 | ||||
; GFX10-NEXT: v_readfirstlane_b32 s7, v7 | ; GFX10-NEXT: v_readfirstlane_b32 s7, v7 | ||||
; GFX10-NEXT: ; return to shader part epilog | ; GFX10-NEXT: ; return to shader part epilog | ||||
; | ; | ||||
; GFX11-LABEL: s_ssubsat_v2i128: | ; GFX11-LABEL: s_ssubsat_v2i128: | ||||
; GFX11: ; %bb.0: | ; GFX11: ; %bb.0: | ||||
; GFX11-NEXT: s_sub_u32 s16, s0, s8 | ; GFX11-NEXT: s_sub_u32 s16, s0, s8 | ||||
; GFX11-NEXT: s_subb_u32 s17, s1, s9 | ; GFX11-NEXT: s_subb_u32 s17, s1, s9 | ||||
; GFX11-NEXT: s_subb_u32 s18, s2, s10 | ; GFX11-NEXT: s_subb_u32 s18, s2, s10 | ||||
; GFX11-NEXT: v_cmp_lt_u64_e64 s0, s[16:17], s[0:1] | ; GFX11-NEXT: v_cmp_lt_u64_e64 s0, s[16:17], s[0:1] | ||||
; GFX11-NEXT: s_subb_u32 s19, s3, s11 | ; GFX11-NEXT: s_subb_u32 s19, s3, s11 | ||||
; GFX11-NEXT: s_cmp_eq_u64 s[18:19], s[2:3] | ; GFX11-NEXT: s_cmp_eq_u64 s[18:19], s[2:3] | ||||
; GFX11-NEXT: s_cselect_b32 s20, 1, 0 | ; GFX11-NEXT: s_cselect_b32 s20, 1, 0 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0 | ; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0 | ||||
; GFX11-NEXT: v_cmp_lt_i64_e64 s0, s[18:19], s[2:3] | ; GFX11-NEXT: v_cmp_lt_i64_e64 s0, s[18:19], s[2:3] | ||||
; GFX11-NEXT: v_cmp_gt_u64_e64 s2, s[8:9], 0 | ; GFX11-NEXT: v_cmp_gt_u64_e64 s2, s[8:9], 0 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0 | ; GFX11-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0 | ||||
; GFX11-NEXT: s_and_b32 s0, 1, s20 | ; GFX11-NEXT: s_and_b32 s0, 1, s20 | ||||
; GFX11-NEXT: s_cmp_eq_u64 s[10:11], 0 | ; GFX11-NEXT: s_cmp_eq_u64 s[10:11], 0 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, s2 | ; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, s2 | ||||
; GFX11-NEXT: v_cmp_gt_i64_e64 s2, s[10:11], 0 | ; GFX11-NEXT: v_cmp_gt_i64_e64 s2, s[10:11], 0 | ||||
; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s0 | ; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s0 | ||||
; GFX11-NEXT: s_cselect_b32 s1, 1, 0 | ; GFX11-NEXT: s_cselect_b32 s1, 1, 0 | ||||
; GFX11-NEXT: s_ashr_i32 s0, s19, 31 | ; GFX11-NEXT: s_ashr_i32 s8, s19, 31 | ||||
; GFX11-NEXT: s_and_b32 s1, 1, s1 | ; GFX11-NEXT: s_and_b32 s1, 1, s1 | ||||
; GFX11-NEXT: s_mov_b32 s3, s0 | ; GFX11-NEXT: s_add_u32 s9, s8, 0x80000000 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s2 | ; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s2 | ||||
; GFX11-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo | ; GFX11-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo | ||||
; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s1 | ; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s1 | ||||
; GFX11-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX11-NEXT: s_sub_u32 s0, s4, s12 | ||||
; GFX11-NEXT: s_sub_u32 s8, s4, s12 | ; GFX11-NEXT: s_subb_u32 s1, s5, s13 | ||||
; GFX11-NEXT: s_subb_u32 s9, s5, s13 | ; GFX11-NEXT: s_subb_u32 s2, s6, s14 | ||||
; GFX11-NEXT: s_subb_u32 s10, s6, s14 | ; GFX11-NEXT: v_cmp_lt_u64_e64 s4, s[0:1], s[4:5] | ||||
; GFX11-NEXT: v_cndmask_b32_e32 v1, v3, v2, vcc_lo | ; GFX11-NEXT: v_cndmask_b32_e32 v1, v3, v2, vcc_lo | ||||
; GFX11-NEXT: v_cmp_lt_u64_e64 s4, s[8:9], s[4:5] | ; GFX11-NEXT: s_subb_u32 s3, s7, s15 | ||||
; GFX11-NEXT: s_subb_u32 s11, s7, s15 | ; GFX11-NEXT: v_dual_mov_b32 v6, s1 :: v_dual_mov_b32 v7, s3 | ||||
; GFX11-NEXT: s_mov_b32 s2, s0 | ; GFX11-NEXT: s_cmp_eq_u64 s[2:3], s[6:7] | ||||
; GFX11-NEXT: s_cmp_eq_u64 s[10:11], s[6:7] | |||||
; GFX11-NEXT: v_xor_b32_e32 v0, v1, v0 | ; GFX11-NEXT: v_xor_b32_e32 v0, v1, v0 | ||||
; GFX11-NEXT: v_mov_b32_e32 v1, s16 | ; GFX11-NEXT: v_cndmask_b32_e64 v1, 0, 1, s4 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 | ; GFX11-NEXT: v_cmp_lt_i64_e64 s4, s[2:3], s[6:7] | ||||
; GFX11-NEXT: v_cmp_lt_i64_e64 s4, s[10:11], s[6:7] | |||||
; GFX11-NEXT: v_cmp_gt_u64_e64 s6, s[12:13], 0 | ; GFX11-NEXT: v_cmp_gt_u64_e64 s6, s[12:13], 0 | ||||
; GFX11-NEXT: v_and_b32_e32 v0, 1, v0 | ; GFX11-NEXT: s_cselect_b32 s10, 1, 0 | ||||
; GFX11-NEXT: s_cselect_b32 s16, 1, 0 | ; GFX11-NEXT: v_mov_b32_e32 v5, s0 | ||||
; GFX11-NEXT: v_mov_b32_e32 v7, s11 | ; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, s4 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v4, 0, 1, s4 | ; GFX11-NEXT: s_and_b32 s4, 1, s10 | ||||
; GFX11-NEXT: s_and_b32 s4, 1, s16 | |||||
; GFX11-NEXT: s_cmp_eq_u64 s[14:15], 0 | ; GFX11-NEXT: s_cmp_eq_u64 s[14:15], 0 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v5, 0, 1, s6 | ; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s6 | ||||
; GFX11-NEXT: v_cmp_gt_i64_e64 s6, s[14:15], 0 | ; GFX11-NEXT: v_cmp_gt_i64_e64 s6, s[14:15], 0 | ||||
; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s4 | ; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s4 | ||||
; GFX11-NEXT: s_cselect_b32 s5, 1, 0 | ; GFX11-NEXT: s_cselect_b32 s5, 1, 0 | ||||
; GFX11-NEXT: v_mov_b32_e32 v2, s17 | ; GFX11-NEXT: v_and_b32_e32 v0, 1, v0 | ||||
; GFX11-NEXT: s_and_b32 s5, 1, s5 | ; GFX11-NEXT: s_and_b32 s5, 1, s5 | ||||
; GFX11-NEXT: v_cndmask_b32_e32 v3, v4, v3, vcc_lo | ; GFX11-NEXT: s_ashr_i32 s4, s3, 31 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v6, 0, 1, s6 | ; GFX11-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc_lo | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v4, 0, 1, s6 | |||||
; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s5 | ; GFX11-NEXT: v_cmp_ne_u32_e64 vcc_lo, 0, s5 | ||||
; GFX11-NEXT: v_dual_cndmask_b32 v4, v6, v5 :: v_dual_mov_b32 v5, s19 | ; GFX11-NEXT: s_add_u32 s0, s4, 0x80000000 | ||||
; GFX11-NEXT: v_dual_cndmask_b32 v2, v4, v3 :: v_dual_mov_b32 v3, s16 | |||||
; GFX11-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 | ; GFX11-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 | ||||
; GFX11-NEXT: v_mov_b32_e32 v6, s9 | |||||
; GFX11-NEXT: v_xor_b32_e32 v3, v4, v3 | |||||
; GFX11-NEXT: v_mov_b32_e32 v0, s18 | ; GFX11-NEXT: v_mov_b32_e32 v0, s18 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s0, vcc_lo | ; GFX11-NEXT: v_xor_b32_e32 v1, v2, v1 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v2, v2, s3, vcc_lo | ; GFX11-NEXT: v_mov_b32_e32 v4, s17 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v4, v5, s1, vcc_lo | ; GFX11-NEXT: v_mov_b32_e32 v2, s19 | ||||
; GFX11-NEXT: v_and_b32_e32 v3, 1, v3 | ; GFX11-NEXT: v_cndmask_b32_e64 v3, v3, s8, vcc_lo | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s2, vcc_lo | ; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, s8, vcc_lo | ||||
; GFX11-NEXT: v_mov_b32_e32 v5, s8 | ; GFX11-NEXT: v_and_b32_e32 v1, 1, v1 | ||||
; GFX11-NEXT: s_ashr_i32 s0, s11, 31 | ; GFX11-NEXT: v_cndmask_b32_e64 v4, v4, s8, vcc_lo | ||||
; GFX11-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v3 | ; GFX11-NEXT: v_cndmask_b32_e64 v2, v2, s9, vcc_lo | ||||
; GFX11-NEXT: v_mov_b32_e32 v3, s10 | ; GFX11-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 | ||||
; GFX11-NEXT: s_add_u32 s1, s0, 0x80000000 | ; GFX11-NEXT: v_mov_b32_e32 v1, s2 | ||||
; GFX11-NEXT: s_mov_b32 s3, s0 | ; GFX11-NEXT: v_readfirstlane_b32 s1, v4 | ||||
; GFX11-NEXT: s_mov_b32 s2, s0 | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v5, v5, s0, vcc_lo | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v6, v6, s3, vcc_lo | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v3, v3, s2, vcc_lo | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v7, v7, s1, vcc_lo | |||||
; GFX11-NEXT: v_readfirstlane_b32 s0, v1 | |||||
; GFX11-NEXT: v_readfirstlane_b32 s1, v2 | |||||
; GFX11-NEXT: v_readfirstlane_b32 s2, v0 | ; GFX11-NEXT: v_readfirstlane_b32 s2, v0 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s3, v4 | ; GFX11-NEXT: v_readfirstlane_b32 s3, v2 | ||||
; GFX11-NEXT: v_cndmask_b32_e64 v5, v5, s4, vcc_lo | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v6, v6, s4, vcc_lo | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, s4, vcc_lo | |||||
; GFX11-NEXT: v_cndmask_b32_e64 v7, v7, s0, vcc_lo | |||||
; GFX11-NEXT: v_readfirstlane_b32 s0, v3 | |||||
; GFX11-NEXT: v_readfirstlane_b32 s4, v5 | ; GFX11-NEXT: v_readfirstlane_b32 s4, v5 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s5, v6 | ; GFX11-NEXT: v_readfirstlane_b32 s5, v6 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s6, v3 | ; GFX11-NEXT: v_readfirstlane_b32 s6, v1 | ||||
; GFX11-NEXT: v_readfirstlane_b32 s7, v7 | ; GFX11-NEXT: v_readfirstlane_b32 s7, v7 | ||||
; GFX11-NEXT: ; return to shader part epilog | ; GFX11-NEXT: ; return to shader part epilog | ||||
%result = call <2 x i128> @llvm.ssub.sat.v2i128(<2 x i128> %lhs, <2 x i128> %rhs) | %result = call <2 x i128> @llvm.ssub.sat.v2i128(<2 x i128> %lhs, <2 x i128> %rhs) | ||||
ret <2 x i128> %result | ret <2 x i128> %result | ||||
} | } | ||||
declare i7 @llvm.ssub.sat.i7(i7, i7) #0 | declare i7 @llvm.ssub.sat.i7(i7, i7) #0 | ||||
declare i8 @llvm.ssub.sat.i8(i8, i8) #0 | declare i8 @llvm.ssub.sat.i8(i8, i8) #0 | ||||
Show All 29 Lines |