aacebaeab5
For the majority of cases, this is a neutral or positive change. There are even testcases that greatly benefit from it, but some regressions are possible. There is #140040 for GlobalISel that'd need to be fixed but it's only a one instruction regression and I think it can be fixed later. Solves #64591
3503 lines
131 KiB
LLVM
3503 lines
131 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
|
|
; RUN: llc -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=SI %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=VI %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s
|
|
|
|
define <5 x float> @bitcast_v5i32_to_v5f32(<5 x i32> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v5i32_to_v5f32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: ; %bb.1: ; %cmp.true
|
|
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
|
|
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v3
|
|
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
|
|
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v1
|
|
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
|
|
; SI-NEXT: ; %bb.2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v5i32_to_v5f32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
|
|
; VI-NEXT: v_add_u32_e32 v3, vcc, 3, v3
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
|
|
; VI-NEXT: v_add_u32_e32 v1, vcc, 3, v1
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
|
|
; VI-NEXT: ; %bb.2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5i32_to_v5f32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_add_u32_e32 v4, 3, v4
|
|
; GFX9-NEXT: v_add_u32_e32 v3, 3, v3
|
|
; GFX9-NEXT: v_add_u32_e32 v2, 3, v2
|
|
; GFX9-NEXT: v_add_u32_e32 v1, 3, v1
|
|
; GFX9-NEXT: v_add_u32_e32 v0, 3, v0
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5i32_to_v5f32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v4, 3, v4
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v3, 3, v3
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v2, 3, v2
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v1, 3, v1
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v0, 3, v0
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <5 x i32> %a, splat (i32 3)
|
|
%a2 = bitcast <5 x i32> %a1 to <5 x float>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x i32> %a to <5 x float>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x float> %phi
|
|
}
|
|
|
|
define inreg <5 x float> @bitcast_v5i32_to_v5f32_scalar(<5 x i32> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v5i32_to_v5f32_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB1_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: s_cbranch_execnz .LBB1_3
|
|
; SI-NEXT: .LBB1_2: ; %cmp.true
|
|
; SI-NEXT: s_add_i32 s20, s20, 3
|
|
; SI-NEXT: s_add_i32 s19, s19, 3
|
|
; SI-NEXT: s_add_i32 s18, s18, 3
|
|
; SI-NEXT: s_add_i32 s17, s17, 3
|
|
; SI-NEXT: s_add_i32 s16, s16, 3
|
|
; SI-NEXT: .LBB1_3: ; %end
|
|
; SI-NEXT: v_mov_b32_e32 v0, s16
|
|
; SI-NEXT: v_mov_b32_e32 v1, s17
|
|
; SI-NEXT: v_mov_b32_e32 v2, s18
|
|
; SI-NEXT: v_mov_b32_e32 v3, s19
|
|
; SI-NEXT: v_mov_b32_e32 v4, s20
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB1_4:
|
|
; SI-NEXT: s_branch .LBB1_2
|
|
;
|
|
; VI-LABEL: bitcast_v5i32_to_v5f32_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB1_4
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB1_3
|
|
; VI-NEXT: .LBB1_2: ; %cmp.true
|
|
; VI-NEXT: s_add_i32 s20, s20, 3
|
|
; VI-NEXT: s_add_i32 s19, s19, 3
|
|
; VI-NEXT: s_add_i32 s18, s18, 3
|
|
; VI-NEXT: s_add_i32 s17, s17, 3
|
|
; VI-NEXT: s_add_i32 s16, s16, 3
|
|
; VI-NEXT: .LBB1_3: ; %end
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB1_4:
|
|
; VI-NEXT: s_branch .LBB1_2
|
|
;
|
|
; GFX9-LABEL: bitcast_v5i32_to_v5f32_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB1_4
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB1_3
|
|
; GFX9-NEXT: .LBB1_2: ; %cmp.true
|
|
; GFX9-NEXT: s_add_i32 s20, s20, 3
|
|
; GFX9-NEXT: s_add_i32 s19, s19, 3
|
|
; GFX9-NEXT: s_add_i32 s18, s18, 3
|
|
; GFX9-NEXT: s_add_i32 s17, s17, 3
|
|
; GFX9-NEXT: s_add_i32 s16, s16, 3
|
|
; GFX9-NEXT: .LBB1_3: ; %end
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB1_4:
|
|
; GFX9-NEXT: s_branch .LBB1_2
|
|
;
|
|
; GFX11-LABEL: bitcast_v5i32_to_v5f32_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s4, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB1_4
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s4
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB1_3
|
|
; GFX11-NEXT: .LBB1_2: ; %cmp.true
|
|
; GFX11-NEXT: s_add_i32 s16, s16, 3
|
|
; GFX11-NEXT: s_add_i32 s3, s3, 3
|
|
; GFX11-NEXT: s_add_i32 s2, s2, 3
|
|
; GFX11-NEXT: s_add_i32 s1, s1, 3
|
|
; GFX11-NEXT: s_add_i32 s0, s0, 3
|
|
; GFX11-NEXT: .LBB1_3: ; %end
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, s16
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB1_4:
|
|
; GFX11-NEXT: s_branch .LBB1_2
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <5 x i32> %a, splat (i32 3)
|
|
%a2 = bitcast <5 x i32> %a1 to <5 x float>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x i32> %a to <5 x float>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x float> %phi
|
|
}
|
|
|
|
define <5 x i32> @bitcast_v5f32_to_v5i32(<5 x float> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v5f32_to_v5i32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: ; %bb.1: ; %cmp.true
|
|
; SI-NEXT: v_add_f32_e32 v4, 1.0, v4
|
|
; SI-NEXT: v_add_f32_e32 v3, 1.0, v3
|
|
; SI-NEXT: v_add_f32_e32 v2, 1.0, v2
|
|
; SI-NEXT: v_add_f32_e32 v1, 1.0, v1
|
|
; SI-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; SI-NEXT: ; %bb.2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v5f32_to_v5i32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_add_f32_e32 v4, 1.0, v4
|
|
; VI-NEXT: v_add_f32_e32 v3, 1.0, v3
|
|
; VI-NEXT: v_add_f32_e32 v2, 1.0, v2
|
|
; VI-NEXT: v_add_f32_e32 v1, 1.0, v1
|
|
; VI-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; VI-NEXT: ; %bb.2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5f32_to_v5i32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_add_f32_e32 v4, 1.0, v4
|
|
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v3
|
|
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v2
|
|
; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
|
|
; GFX9-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5f32_to_v5i32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_dual_add_f32 v4, 1.0, v4 :: v_dual_add_f32 v3, 1.0, v3
|
|
; GFX11-NEXT: v_dual_add_f32 v2, 1.0, v2 :: v_dual_add_f32 v1, 1.0, v1
|
|
; GFX11-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <5 x float> %a, splat (float 1.000000e+00)
|
|
%a2 = bitcast <5 x float> %a1 to <5 x i32>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x float> %a to <5 x i32>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x i32> %phi
|
|
}
|
|
|
|
define inreg <5 x i32> @bitcast_v5f32_to_v5i32_scalar(<5 x float> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v5f32_to_v5i32_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB3_3
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: s_cbranch_execnz .LBB3_4
|
|
; SI-NEXT: .LBB3_2: ; %cmp.true
|
|
; SI-NEXT: v_add_f32_e64 v4, s20, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v3, s19, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v2, s18, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v1, s17, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v0, s16, 1.0
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB3_3:
|
|
; SI-NEXT: s_branch .LBB3_2
|
|
; SI-NEXT: .LBB3_4:
|
|
; SI-NEXT: v_mov_b32_e32 v0, s16
|
|
; SI-NEXT: v_mov_b32_e32 v1, s17
|
|
; SI-NEXT: v_mov_b32_e32 v2, s18
|
|
; SI-NEXT: v_mov_b32_e32 v3, s19
|
|
; SI-NEXT: v_mov_b32_e32 v4, s20
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v5f32_to_v5i32_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB3_3
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB3_4
|
|
; VI-NEXT: .LBB3_2: ; %cmp.true
|
|
; VI-NEXT: v_add_f32_e64 v4, s20, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v3, s19, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v2, s18, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v1, s17, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v0, s16, 1.0
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB3_3:
|
|
; VI-NEXT: s_branch .LBB3_2
|
|
; VI-NEXT: .LBB3_4:
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5f32_to_v5i32_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB3_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB3_4
|
|
; GFX9-NEXT: .LBB3_2: ; %cmp.true
|
|
; GFX9-NEXT: v_add_f32_e64 v4, s20, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v3, s19, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v2, s18, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v1, s17, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v0, s16, 1.0
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB3_3:
|
|
; GFX9-NEXT: s_branch .LBB3_2
|
|
; GFX9-NEXT: .LBB3_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5f32_to_v5i32_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s5, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB3_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s5
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB3_4
|
|
; GFX11-NEXT: .LBB3_2: ; %cmp.true
|
|
; GFX11-NEXT: v_add_f32_e64 v4, s4, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v3, s3, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v2, s2, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v1, s1, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v0, s0, 1.0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB3_3:
|
|
; GFX11-NEXT: s_branch .LBB3_2
|
|
; GFX11-NEXT: .LBB3_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, s4
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <5 x float> %a, splat (float 1.000000e+00)
|
|
%a2 = bitcast <5 x float> %a1 to <5 x i32>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x float> %a to <5 x i32>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x i32> %phi
|
|
}
|
|
|
|
define <10 x i16> @bitcast_v5i32_to_v10i16(<5 x i32> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v5i32_to_v10i16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_mov_b32_e32 v8, v4
|
|
; SI-NEXT: v_mov_b32_e32 v6, v3
|
|
; SI-NEXT: v_mov_b32_e32 v4, v2
|
|
; SI-NEXT: v_mov_b32_e32 v2, v1
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $vgpr3
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB4_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB4_4
|
|
; SI-NEXT: .LBB4_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB4_3: ; %cmp.false
|
|
; SI-NEXT: v_alignbit_b32 v9, s4, v8, 16
|
|
; SI-NEXT: v_alignbit_b32 v5, v6, v4, 16
|
|
; SI-NEXT: v_alignbit_b32 v1, v2, v0, 16
|
|
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v6
|
|
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v2
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB4_2
|
|
; SI-NEXT: .LBB4_4: ; %cmp.true
|
|
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
|
|
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
|
|
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
|
|
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
|
|
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
|
|
; SI-NEXT: v_alignbit_b32 v5, v6, v4, 16
|
|
; SI-NEXT: v_alignbit_b32 v1, v2, v0, 16
|
|
; SI-NEXT: v_alignbit_b32 v9, s4, v8, 16
|
|
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v6
|
|
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v2
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v5i32_to_v10i16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
|
|
; VI-NEXT: v_add_u32_e32 v3, vcc, 3, v3
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
|
|
; VI-NEXT: v_add_u32_e32 v1, vcc, 3, v1
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
|
|
; VI-NEXT: ; %bb.2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5i32_to_v10i16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_add_u32_e32 v4, 3, v4
|
|
; GFX9-NEXT: v_add_u32_e32 v3, 3, v3
|
|
; GFX9-NEXT: v_add_u32_e32 v2, 3, v2
|
|
; GFX9-NEXT: v_add_u32_e32 v1, 3, v1
|
|
; GFX9-NEXT: v_add_u32_e32 v0, 3, v0
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5i32_to_v10i16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v4, 3, v4
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v3, 3, v3
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v2, 3, v2
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v1, 3, v1
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v0, 3, v0
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <5 x i32> %a, splat (i32 3)
|
|
%a2 = bitcast <5 x i32> %a1 to <10 x i16>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x i32> %a to <10 x i16>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x i16> %phi
|
|
}
|
|
|
|
define inreg <10 x i16> @bitcast_v5i32_to_v10i16_scalar(<5 x i32> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v5i32_to_v10i16_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB5_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: v_mov_b32_e32 v0, s20
|
|
; SI-NEXT: v_alignbit_b32 v9, s4, v0, 16
|
|
; SI-NEXT: v_mov_b32_e32 v0, s18
|
|
; SI-NEXT: v_alignbit_b32 v5, s19, v0, 16
|
|
; SI-NEXT: v_mov_b32_e32 v0, s16
|
|
; SI-NEXT: v_alignbit_b32 v1, s17, v0, 16
|
|
; SI-NEXT: s_lshr_b32 s6, s19, 16
|
|
; SI-NEXT: s_lshr_b32 s7, s17, 16
|
|
; SI-NEXT: s_cbranch_execnz .LBB5_3
|
|
; SI-NEXT: .LBB5_2: ; %cmp.true
|
|
; SI-NEXT: s_add_i32 s18, s18, 3
|
|
; SI-NEXT: s_add_i32 s16, s16, 3
|
|
; SI-NEXT: s_add_i32 s19, s19, 3
|
|
; SI-NEXT: v_mov_b32_e32 v0, s18
|
|
; SI-NEXT: s_add_i32 s20, s20, 3
|
|
; SI-NEXT: s_add_i32 s17, s17, 3
|
|
; SI-NEXT: v_alignbit_b32 v5, s19, v0, 16
|
|
; SI-NEXT: v_mov_b32_e32 v0, s16
|
|
; SI-NEXT: v_alignbit_b32 v1, s17, v0, 16
|
|
; SI-NEXT: v_mov_b32_e32 v0, s20
|
|
; SI-NEXT: v_alignbit_b32 v9, s4, v0, 16
|
|
; SI-NEXT: s_lshr_b32 s6, s19, 16
|
|
; SI-NEXT: s_lshr_b32 s7, s17, 16
|
|
; SI-NEXT: .LBB5_3: ; %end
|
|
; SI-NEXT: v_mov_b32_e32 v0, s16
|
|
; SI-NEXT: v_mov_b32_e32 v2, s17
|
|
; SI-NEXT: v_mov_b32_e32 v3, s7
|
|
; SI-NEXT: v_mov_b32_e32 v4, s18
|
|
; SI-NEXT: v_mov_b32_e32 v6, s19
|
|
; SI-NEXT: v_mov_b32_e32 v7, s6
|
|
; SI-NEXT: v_mov_b32_e32 v8, s20
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB5_4:
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $sgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $sgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_branch .LBB5_2
|
|
;
|
|
; VI-LABEL: bitcast_v5i32_to_v10i16_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB5_4
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB5_3
|
|
; VI-NEXT: .LBB5_2: ; %cmp.true
|
|
; VI-NEXT: s_add_i32 s20, s20, 3
|
|
; VI-NEXT: s_add_i32 s19, s19, 3
|
|
; VI-NEXT: s_add_i32 s18, s18, 3
|
|
; VI-NEXT: s_add_i32 s17, s17, 3
|
|
; VI-NEXT: s_add_i32 s16, s16, 3
|
|
; VI-NEXT: .LBB5_3: ; %end
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB5_4:
|
|
; VI-NEXT: s_branch .LBB5_2
|
|
;
|
|
; GFX9-LABEL: bitcast_v5i32_to_v10i16_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB5_4
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB5_3
|
|
; GFX9-NEXT: .LBB5_2: ; %cmp.true
|
|
; GFX9-NEXT: s_add_i32 s20, s20, 3
|
|
; GFX9-NEXT: s_add_i32 s19, s19, 3
|
|
; GFX9-NEXT: s_add_i32 s18, s18, 3
|
|
; GFX9-NEXT: s_add_i32 s17, s17, 3
|
|
; GFX9-NEXT: s_add_i32 s16, s16, 3
|
|
; GFX9-NEXT: .LBB5_3: ; %end
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB5_4:
|
|
; GFX9-NEXT: s_branch .LBB5_2
|
|
;
|
|
; GFX11-LABEL: bitcast_v5i32_to_v10i16_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s4, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB5_4
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s4
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB5_3
|
|
; GFX11-NEXT: .LBB5_2: ; %cmp.true
|
|
; GFX11-NEXT: s_add_i32 s16, s16, 3
|
|
; GFX11-NEXT: s_add_i32 s3, s3, 3
|
|
; GFX11-NEXT: s_add_i32 s2, s2, 3
|
|
; GFX11-NEXT: s_add_i32 s1, s1, 3
|
|
; GFX11-NEXT: s_add_i32 s0, s0, 3
|
|
; GFX11-NEXT: .LBB5_3: ; %end
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, s16
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB5_4:
|
|
; GFX11-NEXT: s_branch .LBB5_2
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <5 x i32> %a, splat (i32 3)
|
|
%a2 = bitcast <5 x i32> %a1 to <10 x i16>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x i32> %a to <10 x i16>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x i16> %phi
|
|
}
|
|
|
|
define <5 x i32> @bitcast_v10i16_to_v5i32(<10 x i16> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v10i16_to_v5i32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_mov_b32_e32 v13, v4
|
|
; SI-NEXT: v_mov_b32_e32 v12, v2
|
|
; SI-NEXT: v_mov_b32_e32 v11, v0
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
|
|
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v1
|
|
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v3
|
|
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v5
|
|
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
|
|
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v9
|
|
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB6_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB6_4
|
|
; SI-NEXT: .LBB6_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB6_3: ; %cmp.false
|
|
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v11
|
|
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v12
|
|
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v13
|
|
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v6
|
|
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v8
|
|
; SI-NEXT: v_or_b32_e32 v0, v0, v15
|
|
; SI-NEXT: v_or_b32_e32 v1, v1, v14
|
|
; SI-NEXT: v_or_b32_e32 v2, v2, v10
|
|
; SI-NEXT: v_or_b32_e32 v3, v3, v7
|
|
; SI-NEXT: v_or_b32_e32 v4, v4, v5
|
|
; SI-NEXT: ; implicit-def: $vgpr11
|
|
; SI-NEXT: ; implicit-def: $vgpr12
|
|
; SI-NEXT: ; implicit-def: $vgpr13
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr8
|
|
; SI-NEXT: ; implicit-def: $vgpr15
|
|
; SI-NEXT: ; implicit-def: $vgpr14
|
|
; SI-NEXT: ; implicit-def: $vgpr10
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB6_2
|
|
; SI-NEXT: .LBB6_4: ; %cmp.true
|
|
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v11
|
|
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v12
|
|
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v13
|
|
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v6
|
|
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v8
|
|
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
|
|
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
|
|
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
|
|
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
|
|
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
|
|
; SI-NEXT: v_or_b32_e32 v0, v15, v0
|
|
; SI-NEXT: s_mov_b32 s6, 0x30000
|
|
; SI-NEXT: v_or_b32_e32 v1, v14, v1
|
|
; SI-NEXT: v_or_b32_e32 v2, v10, v2
|
|
; SI-NEXT: v_or_b32_e32 v3, v7, v3
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: v_add_i32_e32 v0, vcc, 0x30000, v0
|
|
; SI-NEXT: v_add_i32_e32 v1, vcc, s6, v1
|
|
; SI-NEXT: v_add_i32_e32 v2, vcc, 0x30000, v2
|
|
; SI-NEXT: v_add_i32_e32 v3, vcc, 0x30000, v3
|
|
; SI-NEXT: v_add_i32_e32 v4, vcc, 0x30000, v4
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v10i16_to_v5i32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: s_cbranch_execz .LBB6_2
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_mov_b32_e32 v6, 3
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v4
|
|
; VI-NEXT: v_add_u16_sdwa v4, v4, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v3
|
|
; VI-NEXT: v_add_u16_sdwa v3, v3, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v3, v5, v3
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v2
|
|
; VI-NEXT: v_add_u16_sdwa v2, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v2, v5, v2
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v1
|
|
; VI-NEXT: v_add_u16_sdwa v1, v1, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v1, v5, v1
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v0
|
|
; VI-NEXT: v_add_u16_sdwa v0, v0, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v0, v5, v0
|
|
; VI-NEXT: .LBB6_2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10i16_to_v5i32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10i16_to_v5i32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <10 x i16> %a, splat (i16 3)
|
|
%a2 = bitcast <10 x i16> %a1 to <5 x i32>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x i16> %a to <5 x i32>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x i32> %phi
|
|
}
|
|
|
|
define inreg <5 x i32> @bitcast_v10i16_to_v5i32_scalar(<10 x i16> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v10i16_to_v5i32_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s26, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB7_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: s_and_b32 s4, s16, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s5, s17, 16
|
|
; SI-NEXT: s_or_b32 s4, s4, s5
|
|
; SI-NEXT: s_and_b32 s5, s18, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s6, s19, 16
|
|
; SI-NEXT: s_or_b32 s5, s5, s6
|
|
; SI-NEXT: s_and_b32 s6, s20, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s7, s21, 16
|
|
; SI-NEXT: s_or_b32 s6, s6, s7
|
|
; SI-NEXT: s_and_b32 s7, s22, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s8, s23, 16
|
|
; SI-NEXT: s_or_b32 s7, s7, s8
|
|
; SI-NEXT: s_and_b32 s8, s24, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s9, s25, 16
|
|
; SI-NEXT: s_or_b32 s8, s8, s9
|
|
; SI-NEXT: s_cbranch_execnz .LBB7_3
|
|
; SI-NEXT: .LBB7_2: ; %cmp.true
|
|
; SI-NEXT: s_add_i32 s16, s16, 3
|
|
; SI-NEXT: s_and_b32 s4, s16, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s5, s17, 16
|
|
; SI-NEXT: s_add_i32 s18, s18, 3
|
|
; SI-NEXT: s_or_b32 s4, s5, s4
|
|
; SI-NEXT: s_and_b32 s5, s18, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s6, s19, 16
|
|
; SI-NEXT: s_add_i32 s20, s20, 3
|
|
; SI-NEXT: s_or_b32 s5, s6, s5
|
|
; SI-NEXT: s_and_b32 s6, s20, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s7, s21, 16
|
|
; SI-NEXT: s_add_i32 s22, s22, 3
|
|
; SI-NEXT: s_or_b32 s6, s7, s6
|
|
; SI-NEXT: s_and_b32 s7, s22, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s8, s23, 16
|
|
; SI-NEXT: s_add_i32 s24, s24, 3
|
|
; SI-NEXT: s_or_b32 s7, s8, s7
|
|
; SI-NEXT: s_and_b32 s8, s24, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s9, s25, 16
|
|
; SI-NEXT: s_or_b32 s8, s9, s8
|
|
; SI-NEXT: s_add_i32 s4, s4, 0x30000
|
|
; SI-NEXT: s_add_i32 s5, s5, 0x30000
|
|
; SI-NEXT: s_add_i32 s6, s6, 0x30000
|
|
; SI-NEXT: s_add_i32 s7, s7, 0x30000
|
|
; SI-NEXT: s_add_i32 s8, s8, 0x30000
|
|
; SI-NEXT: .LBB7_3: ; %end
|
|
; SI-NEXT: v_mov_b32_e32 v0, s4
|
|
; SI-NEXT: v_mov_b32_e32 v1, s5
|
|
; SI-NEXT: v_mov_b32_e32 v2, s6
|
|
; SI-NEXT: v_mov_b32_e32 v3, s7
|
|
; SI-NEXT: v_mov_b32_e32 v4, s8
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB7_4:
|
|
; SI-NEXT: ; implicit-def: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8
|
|
; SI-NEXT: s_branch .LBB7_2
|
|
;
|
|
; VI-LABEL: bitcast_v10i16_to_v5i32_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB7_4
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB7_3
|
|
; VI-NEXT: .LBB7_2: ; %cmp.true
|
|
; VI-NEXT: s_add_i32 s5, s20, 3
|
|
; VI-NEXT: s_and_b32 s4, s20, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s5, s19, 3
|
|
; VI-NEXT: s_add_i32 s20, s4, 0x30000
|
|
; VI-NEXT: s_and_b32 s4, s19, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s5, s18, 3
|
|
; VI-NEXT: s_add_i32 s19, s4, 0x30000
|
|
; VI-NEXT: s_and_b32 s4, s18, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s5, s17, 3
|
|
; VI-NEXT: s_add_i32 s18, s4, 0x30000
|
|
; VI-NEXT: s_and_b32 s4, s17, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s5, s16, 3
|
|
; VI-NEXT: s_add_i32 s17, s4, 0x30000
|
|
; VI-NEXT: s_and_b32 s4, s16, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s16, s4, 0x30000
|
|
; VI-NEXT: .LBB7_3: ; %end
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB7_4:
|
|
; VI-NEXT: s_branch .LBB7_2
|
|
;
|
|
; GFX9-LABEL: bitcast_v10i16_to_v5i32_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB7_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB7_4
|
|
; GFX9-NEXT: .LBB7_2: ; %cmp.true
|
|
; GFX9-NEXT: v_pk_add_u16 v4, s20, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v3, s19, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v2, s18, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v1, s17, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v0, s16, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB7_3:
|
|
; GFX9-NEXT: s_branch .LBB7_2
|
|
; GFX9-NEXT: .LBB7_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10i16_to_v5i32_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s5, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB7_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s5
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB7_4
|
|
; GFX11-NEXT: .LBB7_2: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_u16 v4, s4, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v3, s3, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v2, s2, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v1, s1, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v0, s0, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB7_3:
|
|
; GFX11-NEXT: s_branch .LBB7_2
|
|
; GFX11-NEXT: .LBB7_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, s4
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <10 x i16> %a, splat (i16 3)
|
|
%a2 = bitcast <10 x i16> %a1 to <5 x i32>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x i16> %a to <5 x i32>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x i32> %phi
|
|
}
|
|
|
|
define <10 x half> @bitcast_v5i32_to_v10f16(<5 x i32> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v5i32_to_v10f16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_mov_b32_e32 v14, v4
|
|
; SI-NEXT: v_mov_b32_e32 v13, v3
|
|
; SI-NEXT: v_mov_b32_e32 v12, v2
|
|
; SI-NEXT: v_mov_b32_e32 v11, v1
|
|
; SI-NEXT: v_mov_b32_e32 v10, v0
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; SI-NEXT: ; implicit-def: $vgpr0
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $vgpr2
|
|
; SI-NEXT: ; implicit-def: $vgpr3
|
|
; SI-NEXT: ; implicit-def: $vgpr4
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr8
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB8_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB8_4
|
|
; SI-NEXT: .LBB8_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB8_3: ; %cmp.false
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v13
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v12
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v11
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v10
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v13
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v12
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v11
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v10
|
|
; SI-NEXT: ; implicit-def: $vgpr10
|
|
; SI-NEXT: ; implicit-def: $vgpr11
|
|
; SI-NEXT: ; implicit-def: $vgpr12
|
|
; SI-NEXT: ; implicit-def: $vgpr13
|
|
; SI-NEXT: ; implicit-def: $vgpr14
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB8_2
|
|
; SI-NEXT: .LBB8_4: ; %cmp.true
|
|
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v10
|
|
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v11
|
|
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v12
|
|
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v13
|
|
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v9
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v1
|
|
; SI-NEXT: v_lshrrev_b32_e32 v9, 16, v9
|
|
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v7
|
|
; SI-NEXT: v_lshrrev_b32_e32 v5, 16, v5
|
|
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v3
|
|
; SI-NEXT: v_lshrrev_b32_e32 v1, 16, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v5i32_to_v10f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
|
|
; VI-NEXT: v_add_u32_e32 v3, vcc, 3, v3
|
|
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
|
|
; VI-NEXT: v_add_u32_e32 v1, vcc, 3, v1
|
|
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
|
|
; VI-NEXT: ; %bb.2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5i32_to_v10f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_add_u32_e32 v4, 3, v4
|
|
; GFX9-NEXT: v_add_u32_e32 v3, 3, v3
|
|
; GFX9-NEXT: v_add_u32_e32 v2, 3, v2
|
|
; GFX9-NEXT: v_add_u32_e32 v1, 3, v1
|
|
; GFX9-NEXT: v_add_u32_e32 v0, 3, v0
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5i32_to_v10f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v4, 3, v4
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v3, 3, v3
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v2, 3, v2
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v1, 3, v1
|
|
; GFX11-NEXT: v_add_nc_u32_e32 v0, 3, v0
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <5 x i32> %a, splat (i32 3)
|
|
%a2 = bitcast <5 x i32> %a1 to <10 x half>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x i32> %a to <10 x half>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x half> %phi
|
|
}
|
|
|
|
define inreg <10 x half> @bitcast_v5i32_to_v10f16_scalar(<5 x i32> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v5i32_to_v10f16_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB9_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: s_lshr_b32 s4, s20, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s19, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s18, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s17, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s16, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, s4
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, s20
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, s19
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, s18
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, s17
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, s16
|
|
; SI-NEXT: s_cbranch_execnz .LBB9_3
|
|
; SI-NEXT: .LBB9_2: ; %cmp.true
|
|
; SI-NEXT: s_add_i32 s20, s20, 3
|
|
; SI-NEXT: s_add_i32 s19, s19, 3
|
|
; SI-NEXT: s_lshr_b32 s4, s20, 16
|
|
; SI-NEXT: s_add_i32 s18, s18, 3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s19, 16
|
|
; SI-NEXT: s_add_i32 s17, s17, 3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s18, 16
|
|
; SI-NEXT: s_add_i32 s16, s16, 3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s17, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s16, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, s20
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, s19
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, s18
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, s17
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, s16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, s4
|
|
; SI-NEXT: .LBB9_3: ; %end
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB9_4:
|
|
; SI-NEXT: ; implicit-def: $vgpr0
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $vgpr2
|
|
; SI-NEXT: ; implicit-def: $vgpr3
|
|
; SI-NEXT: ; implicit-def: $vgpr4
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr8
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_branch .LBB9_2
|
|
;
|
|
; VI-LABEL: bitcast_v5i32_to_v10f16_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB9_4
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB9_3
|
|
; VI-NEXT: .LBB9_2: ; %cmp.true
|
|
; VI-NEXT: s_add_i32 s20, s20, 3
|
|
; VI-NEXT: s_add_i32 s19, s19, 3
|
|
; VI-NEXT: s_add_i32 s18, s18, 3
|
|
; VI-NEXT: s_add_i32 s17, s17, 3
|
|
; VI-NEXT: s_add_i32 s16, s16, 3
|
|
; VI-NEXT: .LBB9_3: ; %end
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB9_4:
|
|
; VI-NEXT: s_branch .LBB9_2
|
|
;
|
|
; GFX9-LABEL: bitcast_v5i32_to_v10f16_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB9_4
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB9_3
|
|
; GFX9-NEXT: .LBB9_2: ; %cmp.true
|
|
; GFX9-NEXT: s_add_i32 s20, s20, 3
|
|
; GFX9-NEXT: s_add_i32 s19, s19, 3
|
|
; GFX9-NEXT: s_add_i32 s18, s18, 3
|
|
; GFX9-NEXT: s_add_i32 s17, s17, 3
|
|
; GFX9-NEXT: s_add_i32 s16, s16, 3
|
|
; GFX9-NEXT: .LBB9_3: ; %end
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB9_4:
|
|
; GFX9-NEXT: s_branch .LBB9_2
|
|
;
|
|
; GFX11-LABEL: bitcast_v5i32_to_v10f16_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s4, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB9_4
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s4
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB9_3
|
|
; GFX11-NEXT: .LBB9_2: ; %cmp.true
|
|
; GFX11-NEXT: s_add_i32 s16, s16, 3
|
|
; GFX11-NEXT: s_add_i32 s3, s3, 3
|
|
; GFX11-NEXT: s_add_i32 s2, s2, 3
|
|
; GFX11-NEXT: s_add_i32 s1, s1, 3
|
|
; GFX11-NEXT: s_add_i32 s0, s0, 3
|
|
; GFX11-NEXT: .LBB9_3: ; %end
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, s16
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB9_4:
|
|
; GFX11-NEXT: s_branch .LBB9_2
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <5 x i32> %a, splat (i32 3)
|
|
%a2 = bitcast <5 x i32> %a1 to <10 x half>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x i32> %a to <10 x half>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x half> %phi
|
|
}
|
|
|
|
define <5 x i32> @bitcast_v10f16_to_v5i32(<10 x half> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v10f16_to_v5i32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v17, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v16, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v15, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v14, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v13, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v12, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v11, v7
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v7, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v9
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v8
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
|
|
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB10_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB10_4
|
|
; SI-NEXT: .LBB10_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB10_3: ; %cmp.false
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v17
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v15
|
|
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v13
|
|
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v11
|
|
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v6
|
|
; SI-NEXT: v_or_b32_e32 v0, v16, v0
|
|
; SI-NEXT: v_or_b32_e32 v1, v14, v1
|
|
; SI-NEXT: v_or_b32_e32 v2, v12, v2
|
|
; SI-NEXT: v_or_b32_e32 v3, v7, v3
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: ; implicit-def: $vgpr17
|
|
; SI-NEXT: ; implicit-def: $vgpr16
|
|
; SI-NEXT: ; implicit-def: $vgpr15
|
|
; SI-NEXT: ; implicit-def: $vgpr14
|
|
; SI-NEXT: ; implicit-def: $vgpr13
|
|
; SI-NEXT: ; implicit-def: $vgpr12
|
|
; SI-NEXT: ; implicit-def: $vgpr11
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB10_2
|
|
; SI-NEXT: .LBB10_4: ; %cmp.true
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v17
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v15
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v14
|
|
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
|
|
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v12
|
|
; SI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v13
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_or_b32_e32 v1, v3, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v11
|
|
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
|
|
; SI-NEXT: v_or_b32_e32 v2, v4, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
|
|
; SI-NEXT: v_or_b32_e32 v3, v4, v3
|
|
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v6
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v10f16_to_v5i32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: s_cbranch_execz .LBB10_2
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_mov_b32_e32 v5, 0x200
|
|
; VI-NEXT: v_add_f16_sdwa v6, v4, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v4, 0x200, v4
|
|
; VI-NEXT: v_or_b32_e32 v4, v4, v6
|
|
; VI-NEXT: v_add_f16_sdwa v6, v3, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v3, 0x200, v3
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v6
|
|
; VI-NEXT: v_add_f16_sdwa v6, v2, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, 0x200, v2
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v6
|
|
; VI-NEXT: v_add_f16_sdwa v6, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v1, 0x200, v1
|
|
; VI-NEXT: v_add_f16_sdwa v5, v0, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v0, 0x200, v0
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v6
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v5
|
|
; VI-NEXT: .LBB10_2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10f16_to_v5i32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: s_movk_i32 s6, 0x200
|
|
; GFX9-NEXT: v_pk_add_f16 v4, v4, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v3, v3, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v2, v2, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v1, v1, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v0, v0, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10f16_to_v5i32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_f16 v4, 0x200, v4 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v3, 0x200, v3 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v2, 0x200, v2 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v1, 0x200, v1 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v0, 0x200, v0 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <10 x half> %a, splat (half 0xH0200)
|
|
%a2 = bitcast <10 x half> %a1 to <5 x i32>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x half> %a to <5 x i32>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x i32> %phi
|
|
}
|
|
|
|
define inreg <5 x i32> @bitcast_v10f16_to_v5i32_scalar(<10 x half> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v10f16_to_v5i32_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v14, s17
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v13, s16
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v12, s19
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v11, s18
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v10, s21
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v9, s20
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v8, s23
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v7, s22
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, s25
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, s24
|
|
; SI-NEXT: s_cmp_lg_u32 s26, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB11_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v14
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v12
|
|
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v10
|
|
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v8
|
|
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v6
|
|
; SI-NEXT: v_or_b32_e32 v0, v13, v0
|
|
; SI-NEXT: v_or_b32_e32 v1, v11, v1
|
|
; SI-NEXT: v_or_b32_e32 v2, v9, v2
|
|
; SI-NEXT: v_or_b32_e32 v3, v7, v3
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: s_cbranch_execnz .LBB11_3
|
|
; SI-NEXT: .LBB11_2: ; %cmp.true
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v13
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v12
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v11
|
|
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
|
|
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v9
|
|
; SI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v10
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_or_b32_e32 v1, v3, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v8
|
|
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
|
|
; SI-NEXT: v_or_b32_e32 v2, v4, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
|
|
; SI-NEXT: v_or_b32_e32 v3, v4, v3
|
|
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v6
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: .LBB11_3: ; %end
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB11_4:
|
|
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4
|
|
; SI-NEXT: s_branch .LBB11_2
|
|
;
|
|
; VI-LABEL: bitcast_v10f16_to_v5i32_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB11_3
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB11_4
|
|
; VI-NEXT: .LBB11_2: ; %cmp.true
|
|
; VI-NEXT: s_lshr_b32 s4, s20, 16
|
|
; VI-NEXT: v_mov_b32_e32 v0, 0x200
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, s20, v0
|
|
; VI-NEXT: s_lshr_b32 s4, s19, 16
|
|
; VI-NEXT: v_or_b32_e32 v4, v2, v1
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, s19, v0
|
|
; VI-NEXT: s_lshr_b32 s4, s18, 16
|
|
; VI-NEXT: v_or_b32_e32 v3, v2, v1
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, s18, v0
|
|
; VI-NEXT: s_lshr_b32 s4, s17, 16
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v1
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v5, s17, v0
|
|
; VI-NEXT: s_lshr_b32 s4, s16, 16
|
|
; VI-NEXT: v_or_b32_e32 v1, v5, v1
|
|
; VI-NEXT: v_mov_b32_e32 v5, s4
|
|
; VI-NEXT: v_add_f16_sdwa v5, v5, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v0, s16, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v5
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB11_3:
|
|
; VI-NEXT: s_branch .LBB11_2
|
|
; VI-NEXT: .LBB11_4:
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10f16_to_v5i32_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB11_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB11_4
|
|
; GFX9-NEXT: .LBB11_2: ; %cmp.true
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0x200
|
|
; GFX9-NEXT: v_pk_add_f16 v4, s20, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v3, s19, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v2, s18, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v1, s17, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v0, s16, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB11_3:
|
|
; GFX9-NEXT: s_branch .LBB11_2
|
|
; GFX9-NEXT: .LBB11_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10f16_to_v5i32_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s5, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB11_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s5
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB11_4
|
|
; GFX11-NEXT: .LBB11_2: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_f16 v4, 0x200, s4 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v3, 0x200, s3 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v2, 0x200, s2 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v1, 0x200, s1 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v0, 0x200, s0 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB11_3:
|
|
; GFX11-NEXT: s_branch .LBB11_2
|
|
; GFX11-NEXT: .LBB11_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, s4
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <10 x half> %a, splat (half 0xH0200)
|
|
%a2 = bitcast <10 x half> %a1 to <5 x i32>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x half> %a to <5 x i32>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x i32> %phi
|
|
}
|
|
|
|
define <10 x i16> @bitcast_v5f32_to_v10i16(<5 x float> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v5f32_to_v10i16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_mov_b32_e32 v8, v4
|
|
; SI-NEXT: v_mov_b32_e32 v6, v3
|
|
; SI-NEXT: v_mov_b32_e32 v4, v2
|
|
; SI-NEXT: v_mov_b32_e32 v2, v1
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $vgpr3
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB12_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB12_4
|
|
; SI-NEXT: .LBB12_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB12_3: ; %cmp.false
|
|
; SI-NEXT: v_alignbit_b32 v9, s4, v8, 16
|
|
; SI-NEXT: v_alignbit_b32 v5, v6, v4, 16
|
|
; SI-NEXT: v_alignbit_b32 v1, v2, v0, 16
|
|
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v6
|
|
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v2
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB12_2
|
|
; SI-NEXT: .LBB12_4: ; %cmp.true
|
|
; SI-NEXT: v_add_f32_e32 v8, 1.0, v8
|
|
; SI-NEXT: v_add_f32_e32 v2, 1.0, v2
|
|
; SI-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; SI-NEXT: v_add_f32_e32 v6, 1.0, v6
|
|
; SI-NEXT: v_add_f32_e32 v4, 1.0, v4
|
|
; SI-NEXT: v_alignbit_b32 v5, v6, v4, 16
|
|
; SI-NEXT: v_alignbit_b32 v1, v2, v0, 16
|
|
; SI-NEXT: v_alignbit_b32 v9, s4, v8, 16
|
|
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v6
|
|
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v2
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v5f32_to_v10i16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_add_f32_e32 v4, 1.0, v4
|
|
; VI-NEXT: v_add_f32_e32 v3, 1.0, v3
|
|
; VI-NEXT: v_add_f32_e32 v2, 1.0, v2
|
|
; VI-NEXT: v_add_f32_e32 v1, 1.0, v1
|
|
; VI-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; VI-NEXT: ; %bb.2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5f32_to_v10i16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_add_f32_e32 v4, 1.0, v4
|
|
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v3
|
|
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v2
|
|
; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
|
|
; GFX9-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5f32_to_v10i16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_dual_add_f32 v4, 1.0, v4 :: v_dual_add_f32 v3, 1.0, v3
|
|
; GFX11-NEXT: v_dual_add_f32 v2, 1.0, v2 :: v_dual_add_f32 v1, 1.0, v1
|
|
; GFX11-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <5 x float> %a, splat (float 1.000000e+00)
|
|
%a2 = bitcast <5 x float> %a1 to <10 x i16>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x float> %a to <10 x i16>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x i16> %phi
|
|
}
|
|
|
|
define inreg <10 x i16> @bitcast_v5f32_to_v10i16_scalar(<5 x float> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v5f32_to_v10i16_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB13_3
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: v_mov_b32_e32 v0, s20
|
|
; SI-NEXT: v_alignbit_b32 v9, s4, v0, 16
|
|
; SI-NEXT: v_mov_b32_e32 v0, s18
|
|
; SI-NEXT: v_alignbit_b32 v5, s19, v0, 16
|
|
; SI-NEXT: v_mov_b32_e32 v0, s16
|
|
; SI-NEXT: v_alignbit_b32 v1, s17, v0, 16
|
|
; SI-NEXT: s_lshr_b32 s7, s19, 16
|
|
; SI-NEXT: s_lshr_b32 s6, s17, 16
|
|
; SI-NEXT: s_cbranch_execnz .LBB13_4
|
|
; SI-NEXT: .LBB13_2: ; %cmp.true
|
|
; SI-NEXT: v_add_f32_e64 v8, s20, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v2, s17, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v0, s16, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v6, s19, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v4, s18, 1.0
|
|
; SI-NEXT: v_alignbit_b32 v5, v6, v4, 16
|
|
; SI-NEXT: v_alignbit_b32 v1, v2, v0, 16
|
|
; SI-NEXT: v_alignbit_b32 v9, s4, v8, 16
|
|
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v6
|
|
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v2
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB13_3:
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $sgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $sgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_branch .LBB13_2
|
|
; SI-NEXT: .LBB13_4:
|
|
; SI-NEXT: v_mov_b32_e32 v0, s16
|
|
; SI-NEXT: v_mov_b32_e32 v2, s17
|
|
; SI-NEXT: v_mov_b32_e32 v4, s18
|
|
; SI-NEXT: v_mov_b32_e32 v6, s19
|
|
; SI-NEXT: v_mov_b32_e32 v8, s20
|
|
; SI-NEXT: v_mov_b32_e32 v3, s6
|
|
; SI-NEXT: v_mov_b32_e32 v7, s7
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v5f32_to_v10i16_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB13_3
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB13_4
|
|
; VI-NEXT: .LBB13_2: ; %cmp.true
|
|
; VI-NEXT: v_add_f32_e64 v4, s20, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v3, s19, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v2, s18, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v1, s17, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v0, s16, 1.0
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB13_3:
|
|
; VI-NEXT: s_branch .LBB13_2
|
|
; VI-NEXT: .LBB13_4:
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: v_mov_b32_e32 v5, s21
|
|
; VI-NEXT: v_mov_b32_e32 v6, s22
|
|
; VI-NEXT: v_mov_b32_e32 v7, s23
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5f32_to_v10i16_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB13_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB13_4
|
|
; GFX9-NEXT: .LBB13_2: ; %cmp.true
|
|
; GFX9-NEXT: v_add_f32_e64 v4, s20, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v3, s19, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v2, s18, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v1, s17, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v0, s16, 1.0
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB13_3:
|
|
; GFX9-NEXT: s_branch .LBB13_2
|
|
; GFX9-NEXT: .LBB13_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s21
|
|
; GFX9-NEXT: v_mov_b32_e32 v6, s22
|
|
; GFX9-NEXT: v_mov_b32_e32 v7, s23
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5f32_to_v10i16_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s8, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB13_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s8
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB13_4
|
|
; GFX11-NEXT: .LBB13_2: ; %cmp.true
|
|
; GFX11-NEXT: v_add_f32_e64 v4, s4, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v3, s3, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v2, s2, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v1, s1, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v0, s0, 1.0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB13_3:
|
|
; GFX11-NEXT: s_branch .LBB13_2
|
|
; GFX11-NEXT: .LBB13_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s5
|
|
; GFX11-NEXT: v_dual_mov_b32 v6, s6 :: v_dual_mov_b32 v7, s7
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <5 x float> %a, splat (float 1.000000e+00)
|
|
%a2 = bitcast <5 x float> %a1 to <10 x i16>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x float> %a to <10 x i16>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x i16> %phi
|
|
}
|
|
|
|
define <5 x float> @bitcast_v10i16_to_v5f32(<10 x i16> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v10i16_to_v5f32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_mov_b32_e32 v13, v4
|
|
; SI-NEXT: v_mov_b32_e32 v12, v2
|
|
; SI-NEXT: v_mov_b32_e32 v11, v0
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
|
|
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v1
|
|
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v3
|
|
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v5
|
|
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
|
|
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v9
|
|
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB14_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB14_4
|
|
; SI-NEXT: .LBB14_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB14_3: ; %cmp.false
|
|
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v11
|
|
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v12
|
|
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v13
|
|
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v6
|
|
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v8
|
|
; SI-NEXT: v_or_b32_e32 v0, v0, v15
|
|
; SI-NEXT: v_or_b32_e32 v1, v1, v14
|
|
; SI-NEXT: v_or_b32_e32 v2, v2, v10
|
|
; SI-NEXT: v_or_b32_e32 v3, v3, v7
|
|
; SI-NEXT: v_or_b32_e32 v4, v4, v5
|
|
; SI-NEXT: ; implicit-def: $vgpr11
|
|
; SI-NEXT: ; implicit-def: $vgpr12
|
|
; SI-NEXT: ; implicit-def: $vgpr13
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr8
|
|
; SI-NEXT: ; implicit-def: $vgpr15
|
|
; SI-NEXT: ; implicit-def: $vgpr14
|
|
; SI-NEXT: ; implicit-def: $vgpr10
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB14_2
|
|
; SI-NEXT: .LBB14_4: ; %cmp.true
|
|
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v11
|
|
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v12
|
|
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v13
|
|
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v6
|
|
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v8
|
|
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
|
|
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
|
|
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
|
|
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
|
|
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
|
|
; SI-NEXT: v_or_b32_e32 v0, v15, v0
|
|
; SI-NEXT: s_mov_b32 s6, 0x30000
|
|
; SI-NEXT: v_or_b32_e32 v1, v14, v1
|
|
; SI-NEXT: v_or_b32_e32 v2, v10, v2
|
|
; SI-NEXT: v_or_b32_e32 v3, v7, v3
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: v_add_i32_e32 v0, vcc, 0x30000, v0
|
|
; SI-NEXT: v_add_i32_e32 v1, vcc, s6, v1
|
|
; SI-NEXT: v_add_i32_e32 v2, vcc, 0x30000, v2
|
|
; SI-NEXT: v_add_i32_e32 v3, vcc, 0x30000, v3
|
|
; SI-NEXT: v_add_i32_e32 v4, vcc, 0x30000, v4
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v10i16_to_v5f32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: s_cbranch_execz .LBB14_2
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_mov_b32_e32 v6, 3
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v4
|
|
; VI-NEXT: v_add_u16_sdwa v4, v4, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v3
|
|
; VI-NEXT: v_add_u16_sdwa v3, v3, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v3, v5, v3
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v2
|
|
; VI-NEXT: v_add_u16_sdwa v2, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v2, v5, v2
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v1
|
|
; VI-NEXT: v_add_u16_sdwa v1, v1, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v1, v5, v1
|
|
; VI-NEXT: v_add_u16_e32 v5, 3, v0
|
|
; VI-NEXT: v_add_u16_sdwa v0, v0, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v0, v5, v0
|
|
; VI-NEXT: .LBB14_2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10i16_to_v5f32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10i16_to_v5f32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <10 x i16> %a, splat (i16 3)
|
|
%a2 = bitcast <10 x i16> %a1 to <5 x float>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x i16> %a to <5 x float>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x float> %phi
|
|
}
|
|
|
|
define inreg <5 x float> @bitcast_v10i16_to_v5f32_scalar(<10 x i16> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v10i16_to_v5f32_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s26, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB15_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: s_and_b32 s4, s16, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s5, s17, 16
|
|
; SI-NEXT: s_or_b32 s4, s4, s5
|
|
; SI-NEXT: s_and_b32 s5, s18, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s6, s19, 16
|
|
; SI-NEXT: s_or_b32 s5, s5, s6
|
|
; SI-NEXT: s_and_b32 s6, s20, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s7, s21, 16
|
|
; SI-NEXT: s_or_b32 s6, s6, s7
|
|
; SI-NEXT: s_and_b32 s7, s22, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s8, s23, 16
|
|
; SI-NEXT: s_or_b32 s7, s7, s8
|
|
; SI-NEXT: s_and_b32 s8, s24, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s9, s25, 16
|
|
; SI-NEXT: s_or_b32 s8, s8, s9
|
|
; SI-NEXT: s_cbranch_execnz .LBB15_3
|
|
; SI-NEXT: .LBB15_2: ; %cmp.true
|
|
; SI-NEXT: s_add_i32 s16, s16, 3
|
|
; SI-NEXT: s_and_b32 s4, s16, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s5, s17, 16
|
|
; SI-NEXT: s_add_i32 s18, s18, 3
|
|
; SI-NEXT: s_or_b32 s4, s5, s4
|
|
; SI-NEXT: s_and_b32 s5, s18, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s6, s19, 16
|
|
; SI-NEXT: s_add_i32 s20, s20, 3
|
|
; SI-NEXT: s_or_b32 s5, s6, s5
|
|
; SI-NEXT: s_and_b32 s6, s20, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s7, s21, 16
|
|
; SI-NEXT: s_add_i32 s22, s22, 3
|
|
; SI-NEXT: s_or_b32 s6, s7, s6
|
|
; SI-NEXT: s_and_b32 s7, s22, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s8, s23, 16
|
|
; SI-NEXT: s_add_i32 s24, s24, 3
|
|
; SI-NEXT: s_or_b32 s7, s8, s7
|
|
; SI-NEXT: s_and_b32 s8, s24, 0xffff
|
|
; SI-NEXT: s_lshl_b32 s9, s25, 16
|
|
; SI-NEXT: s_or_b32 s8, s9, s8
|
|
; SI-NEXT: s_add_i32 s4, s4, 0x30000
|
|
; SI-NEXT: s_add_i32 s5, s5, 0x30000
|
|
; SI-NEXT: s_add_i32 s6, s6, 0x30000
|
|
; SI-NEXT: s_add_i32 s7, s7, 0x30000
|
|
; SI-NEXT: s_add_i32 s8, s8, 0x30000
|
|
; SI-NEXT: .LBB15_3: ; %end
|
|
; SI-NEXT: v_mov_b32_e32 v0, s4
|
|
; SI-NEXT: v_mov_b32_e32 v1, s5
|
|
; SI-NEXT: v_mov_b32_e32 v2, s6
|
|
; SI-NEXT: v_mov_b32_e32 v3, s7
|
|
; SI-NEXT: v_mov_b32_e32 v4, s8
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB15_4:
|
|
; SI-NEXT: ; implicit-def: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8
|
|
; SI-NEXT: s_branch .LBB15_2
|
|
;
|
|
; VI-LABEL: bitcast_v10i16_to_v5f32_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB15_4
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB15_3
|
|
; VI-NEXT: .LBB15_2: ; %cmp.true
|
|
; VI-NEXT: s_add_i32 s5, s20, 3
|
|
; VI-NEXT: s_and_b32 s4, s20, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s5, s19, 3
|
|
; VI-NEXT: s_add_i32 s20, s4, 0x30000
|
|
; VI-NEXT: s_and_b32 s4, s19, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s5, s18, 3
|
|
; VI-NEXT: s_add_i32 s19, s4, 0x30000
|
|
; VI-NEXT: s_and_b32 s4, s18, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s5, s17, 3
|
|
; VI-NEXT: s_add_i32 s18, s4, 0x30000
|
|
; VI-NEXT: s_and_b32 s4, s17, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s5, s16, 3
|
|
; VI-NEXT: s_add_i32 s17, s4, 0x30000
|
|
; VI-NEXT: s_and_b32 s4, s16, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s16, s4, 0x30000
|
|
; VI-NEXT: .LBB15_3: ; %end
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB15_4:
|
|
; VI-NEXT: s_branch .LBB15_2
|
|
;
|
|
; GFX9-LABEL: bitcast_v10i16_to_v5f32_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB15_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB15_4
|
|
; GFX9-NEXT: .LBB15_2: ; %cmp.true
|
|
; GFX9-NEXT: v_pk_add_u16 v4, s20, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v3, s19, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v2, s18, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v1, s17, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v0, s16, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB15_3:
|
|
; GFX9-NEXT: s_branch .LBB15_2
|
|
; GFX9-NEXT: .LBB15_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10i16_to_v5f32_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s5, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB15_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s5
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB15_4
|
|
; GFX11-NEXT: .LBB15_2: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_u16 v4, s4, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v3, s3, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v2, s2, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v1, s1, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v0, s0, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB15_3:
|
|
; GFX11-NEXT: s_branch .LBB15_2
|
|
; GFX11-NEXT: .LBB15_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, s4
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <10 x i16> %a, splat (i16 3)
|
|
%a2 = bitcast <10 x i16> %a1 to <5 x float>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x i16> %a to <5 x float>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x float> %phi
|
|
}
|
|
|
|
define <10 x half> @bitcast_v5f32_to_v10f16(<5 x float> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v5f32_to_v10f16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_mov_b32_e32 v14, v4
|
|
; SI-NEXT: v_mov_b32_e32 v13, v3
|
|
; SI-NEXT: v_mov_b32_e32 v12, v2
|
|
; SI-NEXT: v_mov_b32_e32 v11, v1
|
|
; SI-NEXT: v_mov_b32_e32 v10, v0
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; SI-NEXT: ; implicit-def: $vgpr0
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $vgpr2
|
|
; SI-NEXT: ; implicit-def: $vgpr3
|
|
; SI-NEXT: ; implicit-def: $vgpr4
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr8
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB16_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB16_4
|
|
; SI-NEXT: .LBB16_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB16_3: ; %cmp.false
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v13
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v12
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v11
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v0
|
|
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v10
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v13
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v12
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v11
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v10
|
|
; SI-NEXT: ; implicit-def: $vgpr10
|
|
; SI-NEXT: ; implicit-def: $vgpr11
|
|
; SI-NEXT: ; implicit-def: $vgpr12
|
|
; SI-NEXT: ; implicit-def: $vgpr13
|
|
; SI-NEXT: ; implicit-def: $vgpr14
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB16_2
|
|
; SI-NEXT: .LBB16_4: ; %cmp.true
|
|
; SI-NEXT: v_add_f32_e32 v1, 1.0, v10
|
|
; SI-NEXT: v_add_f32_e32 v3, 1.0, v11
|
|
; SI-NEXT: v_add_f32_e32 v5, 1.0, v12
|
|
; SI-NEXT: v_add_f32_e32 v7, 1.0, v13
|
|
; SI-NEXT: v_add_f32_e32 v9, 1.0, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v9
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v1
|
|
; SI-NEXT: v_lshrrev_b32_e32 v9, 16, v9
|
|
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v7
|
|
; SI-NEXT: v_lshrrev_b32_e32 v5, 16, v5
|
|
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v3
|
|
; SI-NEXT: v_lshrrev_b32_e32 v1, 16, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v5f32_to_v10f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_add_f32_e32 v4, 1.0, v4
|
|
; VI-NEXT: v_add_f32_e32 v3, 1.0, v3
|
|
; VI-NEXT: v_add_f32_e32 v2, 1.0, v2
|
|
; VI-NEXT: v_add_f32_e32 v1, 1.0, v1
|
|
; VI-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; VI-NEXT: ; %bb.2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5f32_to_v10f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_add_f32_e32 v4, 1.0, v4
|
|
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v3
|
|
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v2
|
|
; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
|
|
; GFX9-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5f32_to_v10f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_dual_add_f32 v4, 1.0, v4 :: v_dual_add_f32 v3, 1.0, v3
|
|
; GFX11-NEXT: v_dual_add_f32 v2, 1.0, v2 :: v_dual_add_f32 v1, 1.0, v1
|
|
; GFX11-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <5 x float> %a, splat (float 1.000000e+00)
|
|
%a2 = bitcast <5 x float> %a1 to <10 x half>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x float> %a to <10 x half>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x half> %phi
|
|
}
|
|
|
|
define inreg <10 x half> @bitcast_v5f32_to_v10f16_scalar(<5 x float> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v5f32_to_v10f16_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB17_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: s_lshr_b32 s4, s20, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s19, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s18, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s17, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, s4
|
|
; SI-NEXT: s_lshr_b32 s4, s16, 16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, s4
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, s20
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, s19
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, s18
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, s17
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, s16
|
|
; SI-NEXT: s_cbranch_execnz .LBB17_3
|
|
; SI-NEXT: .LBB17_2: ; %cmp.true
|
|
; SI-NEXT: v_add_f32_e64 v1, s16, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v3, s17, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v5, s18, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v7, s19, 1.0
|
|
; SI-NEXT: v_add_f32_e64 v9, s20, 1.0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v9
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v1
|
|
; SI-NEXT: v_lshrrev_b32_e32 v9, 16, v9
|
|
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v7
|
|
; SI-NEXT: v_lshrrev_b32_e32 v5, 16, v5
|
|
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v3
|
|
; SI-NEXT: v_lshrrev_b32_e32 v1, 16, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: .LBB17_3: ; %end
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB17_4:
|
|
; SI-NEXT: ; implicit-def: $vgpr0
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $vgpr2
|
|
; SI-NEXT: ; implicit-def: $vgpr3
|
|
; SI-NEXT: ; implicit-def: $vgpr4
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr8
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_branch .LBB17_2
|
|
;
|
|
; VI-LABEL: bitcast_v5f32_to_v10f16_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB17_3
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB17_4
|
|
; VI-NEXT: .LBB17_2: ; %cmp.true
|
|
; VI-NEXT: v_add_f32_e64 v4, s20, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v3, s19, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v2, s18, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v1, s17, 1.0
|
|
; VI-NEXT: v_add_f32_e64 v0, s16, 1.0
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB17_3:
|
|
; VI-NEXT: s_branch .LBB17_2
|
|
; VI-NEXT: .LBB17_4:
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: v_mov_b32_e32 v5, s21
|
|
; VI-NEXT: v_mov_b32_e32 v6, s22
|
|
; VI-NEXT: v_mov_b32_e32 v7, s23
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v5f32_to_v10f16_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB17_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB17_4
|
|
; GFX9-NEXT: .LBB17_2: ; %cmp.true
|
|
; GFX9-NEXT: v_add_f32_e64 v4, s20, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v3, s19, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v2, s18, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v1, s17, 1.0
|
|
; GFX9-NEXT: v_add_f32_e64 v0, s16, 1.0
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB17_3:
|
|
; GFX9-NEXT: s_branch .LBB17_2
|
|
; GFX9-NEXT: .LBB17_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s21
|
|
; GFX9-NEXT: v_mov_b32_e32 v6, s22
|
|
; GFX9-NEXT: v_mov_b32_e32 v7, s23
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v5f32_to_v10f16_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s8, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB17_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s8
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB17_4
|
|
; GFX11-NEXT: .LBB17_2: ; %cmp.true
|
|
; GFX11-NEXT: v_add_f32_e64 v4, s4, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v3, s3, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v2, s2, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v1, s1, 1.0
|
|
; GFX11-NEXT: v_add_f32_e64 v0, s0, 1.0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB17_3:
|
|
; GFX11-NEXT: s_branch .LBB17_2
|
|
; GFX11-NEXT: .LBB17_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s5
|
|
; GFX11-NEXT: v_dual_mov_b32 v6, s6 :: v_dual_mov_b32 v7, s7
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <5 x float> %a, splat (float 1.000000e+00)
|
|
%a2 = bitcast <5 x float> %a1 to <10 x half>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <5 x float> %a to <10 x half>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x half> %phi
|
|
}
|
|
|
|
define <5 x float> @bitcast_v10f16_to_v5f32(<10 x half> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v10f16_to_v5f32:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v17, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v16, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v15, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v14, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v13, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v12, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v11, v7
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v7, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v9
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v8
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
|
|
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB18_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB18_4
|
|
; SI-NEXT: .LBB18_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB18_3: ; %cmp.false
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v17
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v15
|
|
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v13
|
|
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v11
|
|
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v6
|
|
; SI-NEXT: v_or_b32_e32 v0, v16, v0
|
|
; SI-NEXT: v_or_b32_e32 v1, v14, v1
|
|
; SI-NEXT: v_or_b32_e32 v2, v12, v2
|
|
; SI-NEXT: v_or_b32_e32 v3, v7, v3
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: ; implicit-def: $vgpr17
|
|
; SI-NEXT: ; implicit-def: $vgpr16
|
|
; SI-NEXT: ; implicit-def: $vgpr15
|
|
; SI-NEXT: ; implicit-def: $vgpr14
|
|
; SI-NEXT: ; implicit-def: $vgpr13
|
|
; SI-NEXT: ; implicit-def: $vgpr12
|
|
; SI-NEXT: ; implicit-def: $vgpr11
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB18_2
|
|
; SI-NEXT: .LBB18_4: ; %cmp.true
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v17
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v15
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v14
|
|
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
|
|
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v12
|
|
; SI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v13
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_or_b32_e32 v1, v3, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v11
|
|
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
|
|
; SI-NEXT: v_or_b32_e32 v2, v4, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
|
|
; SI-NEXT: v_or_b32_e32 v3, v4, v3
|
|
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v6
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v10f16_to_v5f32:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: s_cbranch_execz .LBB18_2
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_mov_b32_e32 v5, 0x200
|
|
; VI-NEXT: v_add_f16_sdwa v6, v4, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v4, 0x200, v4
|
|
; VI-NEXT: v_or_b32_e32 v4, v4, v6
|
|
; VI-NEXT: v_add_f16_sdwa v6, v3, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v3, 0x200, v3
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v6
|
|
; VI-NEXT: v_add_f16_sdwa v6, v2, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, 0x200, v2
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v6
|
|
; VI-NEXT: v_add_f16_sdwa v6, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v1, 0x200, v1
|
|
; VI-NEXT: v_add_f16_sdwa v5, v0, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v0, 0x200, v0
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v6
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v5
|
|
; VI-NEXT: .LBB18_2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10f16_to_v5f32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: s_movk_i32 s6, 0x200
|
|
; GFX9-NEXT: v_pk_add_f16 v4, v4, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v3, v3, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v2, v2, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v1, v1, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v0, v0, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10f16_to_v5f32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_f16 v4, 0x200, v4 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v3, 0x200, v3 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v2, 0x200, v2 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v1, 0x200, v1 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v0, 0x200, v0 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <10 x half> %a, splat (half 0xH0200)
|
|
%a2 = bitcast <10 x half> %a1 to <5 x float>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x half> %a to <5 x float>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x float> %phi
|
|
}
|
|
|
|
define inreg <5 x float> @bitcast_v10f16_to_v5f32_scalar(<10 x half> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v10f16_to_v5f32_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v14, s17
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v13, s16
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v12, s19
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v11, s18
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v10, s21
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v9, s20
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v8, s23
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v7, s22
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, s25
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, s24
|
|
; SI-NEXT: s_cmp_lg_u32 s26, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB19_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v14
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v12
|
|
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v10
|
|
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v8
|
|
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v6
|
|
; SI-NEXT: v_or_b32_e32 v0, v13, v0
|
|
; SI-NEXT: v_or_b32_e32 v1, v11, v1
|
|
; SI-NEXT: v_or_b32_e32 v2, v9, v2
|
|
; SI-NEXT: v_or_b32_e32 v3, v7, v3
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: s_cbranch_execnz .LBB19_3
|
|
; SI-NEXT: .LBB19_2: ; %cmp.true
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v13
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v12
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v11
|
|
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
|
|
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v9
|
|
; SI-NEXT: v_or_b32_e32 v0, v1, v0
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v10
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_or_b32_e32 v1, v3, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v8
|
|
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
|
|
; SI-NEXT: v_or_b32_e32 v2, v4, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
|
|
; SI-NEXT: v_or_b32_e32 v3, v4, v3
|
|
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v6
|
|
; SI-NEXT: v_or_b32_e32 v4, v5, v4
|
|
; SI-NEXT: .LBB19_3: ; %end
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB19_4:
|
|
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4
|
|
; SI-NEXT: s_branch .LBB19_2
|
|
;
|
|
; VI-LABEL: bitcast_v10f16_to_v5f32_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB19_3
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB19_4
|
|
; VI-NEXT: .LBB19_2: ; %cmp.true
|
|
; VI-NEXT: s_lshr_b32 s4, s20, 16
|
|
; VI-NEXT: v_mov_b32_e32 v0, 0x200
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, s20, v0
|
|
; VI-NEXT: s_lshr_b32 s4, s19, 16
|
|
; VI-NEXT: v_or_b32_e32 v4, v2, v1
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, s19, v0
|
|
; VI-NEXT: s_lshr_b32 s4, s18, 16
|
|
; VI-NEXT: v_or_b32_e32 v3, v2, v1
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, s18, v0
|
|
; VI-NEXT: s_lshr_b32 s4, s17, 16
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v1
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v5, s17, v0
|
|
; VI-NEXT: s_lshr_b32 s4, s16, 16
|
|
; VI-NEXT: v_or_b32_e32 v1, v5, v1
|
|
; VI-NEXT: v_mov_b32_e32 v5, s4
|
|
; VI-NEXT: v_add_f16_sdwa v5, v5, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v0, s16, v0
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v5
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB19_3:
|
|
; VI-NEXT: s_branch .LBB19_2
|
|
; VI-NEXT: .LBB19_4:
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10f16_to_v5f32_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB19_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB19_4
|
|
; GFX9-NEXT: .LBB19_2: ; %cmp.true
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0x200
|
|
; GFX9-NEXT: v_pk_add_f16 v4, s20, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v3, s19, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v2, s18, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v1, s17, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v0, s16, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB19_3:
|
|
; GFX9-NEXT: s_branch .LBB19_2
|
|
; GFX9-NEXT: .LBB19_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10f16_to_v5f32_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s5, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB19_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s5
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB19_4
|
|
; GFX11-NEXT: .LBB19_2: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_f16 v4, 0x200, s4 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v3, 0x200, s3 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v2, 0x200, s2 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v1, 0x200, s1 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v0, 0x200, s0 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB19_3:
|
|
; GFX11-NEXT: s_branch .LBB19_2
|
|
; GFX11-NEXT: .LBB19_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_mov_b32_e32 v4, s4
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <10 x half> %a, splat (half 0xH0200)
|
|
%a2 = bitcast <10 x half> %a1 to <5 x float>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x half> %a to <5 x float>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <5 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <5 x float> %phi
|
|
}
|
|
|
|
define <10 x half> @bitcast_v10i16_to_v10f16(<10 x i16> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v10i16_to_v10f16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_mov_b32_e32 v19, v9
|
|
; SI-NEXT: v_mov_b32_e32 v18, v8
|
|
; SI-NEXT: v_mov_b32_e32 v17, v7
|
|
; SI-NEXT: v_mov_b32_e32 v16, v6
|
|
; SI-NEXT: v_mov_b32_e32 v15, v5
|
|
; SI-NEXT: v_mov_b32_e32 v14, v4
|
|
; SI-NEXT: v_mov_b32_e32 v13, v3
|
|
; SI-NEXT: v_mov_b32_e32 v12, v2
|
|
; SI-NEXT: v_mov_b32_e32 v11, v1
|
|
; SI-NEXT: v_mov_b32_e32 v20, v0
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
|
|
; SI-NEXT: ; implicit-def: $vgpr0
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $vgpr2
|
|
; SI-NEXT: ; implicit-def: $vgpr3
|
|
; SI-NEXT: ; implicit-def: $vgpr4
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr8
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB20_3
|
|
; SI-NEXT: ; %bb.1: ; %Flow
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execnz .LBB20_4
|
|
; SI-NEXT: .LBB20_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB20_3: ; %cmp.false
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v20
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v11
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v12
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v13
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v14
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v15
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v17
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v18
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v19
|
|
; SI-NEXT: ; implicit-def: $vgpr20
|
|
; SI-NEXT: ; implicit-def: $vgpr11
|
|
; SI-NEXT: ; implicit-def: $vgpr12
|
|
; SI-NEXT: ; implicit-def: $vgpr13
|
|
; SI-NEXT: ; implicit-def: $vgpr14
|
|
; SI-NEXT: ; implicit-def: $vgpr15
|
|
; SI-NEXT: ; implicit-def: $vgpr16
|
|
; SI-NEXT: ; implicit-def: $vgpr17
|
|
; SI-NEXT: ; implicit-def: $vgpr18
|
|
; SI-NEXT: ; implicit-def: $vgpr19
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB20_2
|
|
; SI-NEXT: .LBB20_4: ; %cmp.true
|
|
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v19
|
|
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v18
|
|
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v17
|
|
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v16
|
|
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v15
|
|
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v14
|
|
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v13
|
|
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v12
|
|
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v11
|
|
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v20
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v8
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v10i16_to_v10f16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: s_cbranch_execz .LBB20_2
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_mov_b32_e32 v5, 3
|
|
; VI-NEXT: v_add_u16_sdwa v6, v0, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_u16_sdwa v7, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_u16_sdwa v8, v2, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_u16_sdwa v9, v3, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_u16_sdwa v5, v4, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_u16_e32 v4, 3, v4
|
|
; VI-NEXT: v_add_u16_e32 v3, 3, v3
|
|
; VI-NEXT: v_add_u16_e32 v2, 3, v2
|
|
; VI-NEXT: v_add_u16_e32 v1, 3, v1
|
|
; VI-NEXT: v_add_u16_e32 v0, 3, v0
|
|
; VI-NEXT: v_or_b32_e32 v4, v4, v5
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v9
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v8
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v7
|
|
; VI-NEXT: v_or_b32_e32 v0, v0, v6
|
|
; VI-NEXT: .LBB20_2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10i16_to_v10f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10i16_to_v10f16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <10 x i16> %a, splat (i16 3)
|
|
%a2 = bitcast <10 x i16> %a1 to <10 x half>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x i16> %a to <10 x half>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x half> %phi
|
|
}
|
|
|
|
define inreg <10 x half> @bitcast_v10i16_to_v10f16_scalar(<10 x i16> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v10i16_to_v10f16_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: s_cmp_lg_u32 s26, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB21_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, s16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, s17
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, s18
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, s19
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, s20
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, s21
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, s22
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, s23
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, s24
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, s25
|
|
; SI-NEXT: s_cbranch_execnz .LBB21_3
|
|
; SI-NEXT: .LBB21_2: ; %cmp.true
|
|
; SI-NEXT: s_add_i32 s25, s25, 3
|
|
; SI-NEXT: s_add_i32 s24, s24, 3
|
|
; SI-NEXT: s_add_i32 s23, s23, 3
|
|
; SI-NEXT: s_add_i32 s22, s22, 3
|
|
; SI-NEXT: s_add_i32 s21, s21, 3
|
|
; SI-NEXT: s_add_i32 s20, s20, 3
|
|
; SI-NEXT: s_add_i32 s19, s19, 3
|
|
; SI-NEXT: s_add_i32 s18, s18, 3
|
|
; SI-NEXT: s_add_i32 s17, s17, 3
|
|
; SI-NEXT: s_add_i32 s16, s16, 3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, s16
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, s17
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, s18
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, s19
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, s20
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, s21
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, s22
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, s23
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, s24
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, s25
|
|
; SI-NEXT: .LBB21_3: ; %end
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB21_4:
|
|
; SI-NEXT: ; implicit-def: $vgpr0
|
|
; SI-NEXT: ; implicit-def: $vgpr1
|
|
; SI-NEXT: ; implicit-def: $vgpr2
|
|
; SI-NEXT: ; implicit-def: $vgpr3
|
|
; SI-NEXT: ; implicit-def: $vgpr4
|
|
; SI-NEXT: ; implicit-def: $vgpr5
|
|
; SI-NEXT: ; implicit-def: $vgpr6
|
|
; SI-NEXT: ; implicit-def: $vgpr7
|
|
; SI-NEXT: ; implicit-def: $vgpr8
|
|
; SI-NEXT: ; implicit-def: $vgpr9
|
|
; SI-NEXT: s_branch .LBB21_2
|
|
;
|
|
; VI-LABEL: bitcast_v10i16_to_v10f16_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB21_4
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB21_3
|
|
; VI-NEXT: .LBB21_2: ; %cmp.true
|
|
; VI-NEXT: s_add_i32 s5, s16, 3
|
|
; VI-NEXT: s_add_i32 s7, s17, 3
|
|
; VI-NEXT: s_add_i32 s9, s18, 3
|
|
; VI-NEXT: s_add_i32 s11, s19, 3
|
|
; VI-NEXT: s_add_i32 s13, s20, 3
|
|
; VI-NEXT: s_and_b32 s4, s16, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s6, s17, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s8, s18, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s10, s19, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s12, s20, 0xffff0000
|
|
; VI-NEXT: s_and_b32 s13, s13, 0xffff
|
|
; VI-NEXT: s_and_b32 s11, s11, 0xffff
|
|
; VI-NEXT: s_and_b32 s9, s9, 0xffff
|
|
; VI-NEXT: s_and_b32 s7, s7, 0xffff
|
|
; VI-NEXT: s_and_b32 s5, s5, 0xffff
|
|
; VI-NEXT: s_or_b32 s12, s12, s13
|
|
; VI-NEXT: s_or_b32 s10, s10, s11
|
|
; VI-NEXT: s_or_b32 s8, s8, s9
|
|
; VI-NEXT: s_or_b32 s6, s6, s7
|
|
; VI-NEXT: s_or_b32 s4, s4, s5
|
|
; VI-NEXT: s_add_i32 s20, s12, 0x30000
|
|
; VI-NEXT: s_add_i32 s19, s10, 0x30000
|
|
; VI-NEXT: s_add_i32 s18, s8, 0x30000
|
|
; VI-NEXT: s_add_i32 s17, s6, 0x30000
|
|
; VI-NEXT: s_add_i32 s16, s4, 0x30000
|
|
; VI-NEXT: .LBB21_3: ; %end
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB21_4:
|
|
; VI-NEXT: s_branch .LBB21_2
|
|
;
|
|
; GFX9-LABEL: bitcast_v10i16_to_v10f16_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB21_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB21_4
|
|
; GFX9-NEXT: .LBB21_2: ; %cmp.true
|
|
; GFX9-NEXT: v_pk_add_u16 v4, s20, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v3, s19, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v2, s18, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v1, s17, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_u16 v0, s16, 3 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB21_3:
|
|
; GFX9-NEXT: s_branch .LBB21_2
|
|
; GFX9-NEXT: .LBB21_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s21
|
|
; GFX9-NEXT: v_mov_b32_e32 v6, s22
|
|
; GFX9-NEXT: v_mov_b32_e32 v7, s23
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10i16_to_v10f16_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s8, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB21_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s8
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB21_4
|
|
; GFX11-NEXT: .LBB21_2: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_u16 v4, s4, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v3, s3, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v2, s2, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v1, s1, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: v_pk_add_u16 v0, s0, 3 op_sel_hi:[1,0]
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB21_3:
|
|
; GFX11-NEXT: s_branch .LBB21_2
|
|
; GFX11-NEXT: .LBB21_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s5
|
|
; GFX11-NEXT: v_dual_mov_b32 v6, s6 :: v_dual_mov_b32 v7, s7
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = add <10 x i16> %a, splat (i16 3)
|
|
%a2 = bitcast <10 x i16> %a1 to <10 x half>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x i16> %a to <10 x half>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x half> %phi
|
|
}
|
|
|
|
define <10 x i16> @bitcast_v10f16_to_v10i16(<10 x half> %a, i32 %b) {
|
|
; SI-LABEL: bitcast_v10f16_to_v10i16:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
|
|
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
|
|
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; SI-NEXT: s_cbranch_execz .LBB22_2
|
|
; SI-NEXT: ; %bb.1: ; %cmp.true
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v8
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
|
|
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
|
|
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
|
|
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v9
|
|
; SI-NEXT: v_or_b32_e32 v8, v8, v10
|
|
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v7
|
|
; SI-NEXT: v_or_b32_e32 v6, v6, v10
|
|
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v3
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
|
|
; SI-NEXT: v_or_b32_e32 v2, v2, v10
|
|
; SI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; SI-NEXT: v_or_b32_e32 v4, v4, v5
|
|
; SI-NEXT: v_alignbit_b32 v1, v2, v1, 16
|
|
; SI-NEXT: v_alignbit_b32 v5, v6, v5, 16
|
|
; SI-NEXT: .LBB22_2: ; %end
|
|
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; VI-LABEL: bitcast_v10f16_to_v10i16:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; VI-NEXT: s_cbranch_execz .LBB22_2
|
|
; VI-NEXT: ; %bb.1: ; %cmp.true
|
|
; VI-NEXT: v_mov_b32_e32 v6, 0x200
|
|
; VI-NEXT: v_add_f16_e32 v5, 0x200, v0
|
|
; VI-NEXT: v_add_f16_sdwa v0, v0, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v7, 0x200, v1
|
|
; VI-NEXT: v_add_f16_sdwa v1, v1, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v8, 0x200, v2
|
|
; VI-NEXT: v_add_f16_sdwa v2, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v9, 0x200, v3
|
|
; VI-NEXT: v_add_f16_sdwa v3, v3, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v10, 0x200, v4
|
|
; VI-NEXT: v_add_f16_sdwa v4, v4, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v4, v10, v4
|
|
; VI-NEXT: v_or_b32_e32 v3, v9, v3
|
|
; VI-NEXT: v_or_b32_e32 v2, v8, v2
|
|
; VI-NEXT: v_or_b32_e32 v1, v7, v1
|
|
; VI-NEXT: v_or_b32_e32 v0, v5, v0
|
|
; VI-NEXT: .LBB22_2: ; %end
|
|
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10f16_to_v10i16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
|
|
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
|
|
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
|
|
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX9-NEXT: s_movk_i32 s6, 0x200
|
|
; GFX9-NEXT: v_pk_add_f16 v4, v4, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v3, v3, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v2, v2, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v1, v1, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v0, v0, s6 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: ; %bb.2: ; %end
|
|
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10f16_to_v10i16:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s0, exec_lo
|
|
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v5
|
|
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
|
|
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
|
|
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
|
|
; GFX11-NEXT: ; %bb.1: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_f16 v4, 0x200, v4 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v3, 0x200, v3 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v2, 0x200, v2 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v1, 0x200, v1 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v0, 0x200, v0 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: ; %bb.2: ; %end
|
|
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <10 x half> %a, splat (half 0xH0200)
|
|
%a2 = bitcast <10 x half> %a1 to <10 x i16>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x half> %a to <10 x i16>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x i16> %phi
|
|
}
|
|
|
|
define inreg <10 x i16> @bitcast_v10f16_to_v10i16_scalar(<10 x half> inreg %a, i32 inreg %b) {
|
|
; SI-LABEL: bitcast_v10f16_to_v10i16_scalar:
|
|
; SI: ; %bb.0:
|
|
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, s16
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, s17
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, s18
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, s19
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, s20
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, s21
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, s22
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v7, s23
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v8, s24
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v9, s25
|
|
; SI-NEXT: s_cmp_lg_u32 s26, 0
|
|
; SI-NEXT: s_cbranch_scc0 .LBB23_4
|
|
; SI-NEXT: ; %bb.1: ; %cmp.false
|
|
; SI-NEXT: s_cbranch_execnz .LBB23_3
|
|
; SI-NEXT: .LBB23_2: ; %cmp.true
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v9, v9
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v8, v8
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v7, v7
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v3, v3
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; SI-NEXT: v_cvt_f32_f16_e32 v4, v4
|
|
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
|
|
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
|
|
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
|
|
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
|
|
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
|
|
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
|
|
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
|
|
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
|
|
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
|
|
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
|
|
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
|
|
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v9
|
|
; SI-NEXT: v_or_b32_e32 v8, v8, v10
|
|
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v7
|
|
; SI-NEXT: v_or_b32_e32 v6, v6, v10
|
|
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v3
|
|
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
|
|
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
|
|
; SI-NEXT: v_or_b32_e32 v2, v2, v10
|
|
; SI-NEXT: v_or_b32_e32 v0, v0, v1
|
|
; SI-NEXT: v_or_b32_e32 v4, v4, v5
|
|
; SI-NEXT: v_alignbit_b32 v1, v2, v1, 16
|
|
; SI-NEXT: v_alignbit_b32 v5, v6, v5, 16
|
|
; SI-NEXT: .LBB23_3: ; %end
|
|
; SI-NEXT: s_setpc_b64 s[30:31]
|
|
; SI-NEXT: .LBB23_4:
|
|
; SI-NEXT: s_branch .LBB23_2
|
|
;
|
|
; VI-LABEL: bitcast_v10f16_to_v10i16_scalar:
|
|
; VI: ; %bb.0:
|
|
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; VI-NEXT: s_cmp_lg_u32 s21, 0
|
|
; VI-NEXT: s_cbranch_scc0 .LBB23_3
|
|
; VI-NEXT: ; %bb.1: ; %cmp.false
|
|
; VI-NEXT: s_cbranch_execnz .LBB23_4
|
|
; VI-NEXT: .LBB23_2: ; %cmp.true
|
|
; VI-NEXT: s_lshr_b32 s4, s16, 16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s4
|
|
; VI-NEXT: s_lshr_b32 s4, s17, 16
|
|
; VI-NEXT: v_mov_b32_e32 v2, s4
|
|
; VI-NEXT: s_lshr_b32 s4, s18, 16
|
|
; VI-NEXT: v_mov_b32_e32 v3, s4
|
|
; VI-NEXT: s_lshr_b32 s4, s19, 16
|
|
; VI-NEXT: v_mov_b32_e32 v4, s4
|
|
; VI-NEXT: s_lshr_b32 s4, s20, 16
|
|
; VI-NEXT: v_mov_b32_e32 v0, 0x200
|
|
; VI-NEXT: v_mov_b32_e32 v10, s4
|
|
; VI-NEXT: v_add_f16_e32 v5, s16, v0
|
|
; VI-NEXT: v_add_f16_sdwa v6, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v1, s17, v0
|
|
; VI-NEXT: v_add_f16_sdwa v7, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v2, s18, v0
|
|
; VI-NEXT: v_add_f16_sdwa v8, v3, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v3, s19, v0
|
|
; VI-NEXT: v_add_f16_sdwa v9, v4, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_add_f16_e32 v4, s20, v0
|
|
; VI-NEXT: v_add_f16_sdwa v0, v10, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
|
|
; VI-NEXT: v_or_b32_e32 v4, v4, v0
|
|
; VI-NEXT: v_or_b32_e32 v3, v3, v9
|
|
; VI-NEXT: v_or_b32_e32 v2, v2, v8
|
|
; VI-NEXT: v_or_b32_e32 v1, v1, v7
|
|
; VI-NEXT: v_or_b32_e32 v0, v5, v6
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
; VI-NEXT: .LBB23_3:
|
|
; VI-NEXT: s_branch .LBB23_2
|
|
; VI-NEXT: .LBB23_4:
|
|
; VI-NEXT: v_mov_b32_e32 v0, s16
|
|
; VI-NEXT: v_mov_b32_e32 v1, s17
|
|
; VI-NEXT: v_mov_b32_e32 v2, s18
|
|
; VI-NEXT: v_mov_b32_e32 v3, s19
|
|
; VI-NEXT: v_mov_b32_e32 v4, s20
|
|
; VI-NEXT: v_mov_b32_e32 v5, s21
|
|
; VI-NEXT: v_mov_b32_e32 v6, s22
|
|
; VI-NEXT: v_mov_b32_e32 v7, s23
|
|
; VI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: bitcast_v10f16_to_v10i16_scalar:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: s_cmp_lg_u32 s21, 0
|
|
; GFX9-NEXT: s_cbranch_scc0 .LBB23_3
|
|
; GFX9-NEXT: ; %bb.1: ; %cmp.false
|
|
; GFX9-NEXT: s_cbranch_execnz .LBB23_4
|
|
; GFX9-NEXT: .LBB23_2: ; %cmp.true
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, 0x200
|
|
; GFX9-NEXT: v_pk_add_f16 v4, s20, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v3, s19, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v2, s18, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v1, s17, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: v_pk_add_f16 v0, s16, v0 op_sel_hi:[1,0]
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX9-NEXT: .LBB23_3:
|
|
; GFX9-NEXT: s_branch .LBB23_2
|
|
; GFX9-NEXT: .LBB23_4:
|
|
; GFX9-NEXT: v_mov_b32_e32 v0, s16
|
|
; GFX9-NEXT: v_mov_b32_e32 v1, s17
|
|
; GFX9-NEXT: v_mov_b32_e32 v2, s18
|
|
; GFX9-NEXT: v_mov_b32_e32 v3, s19
|
|
; GFX9-NEXT: v_mov_b32_e32 v4, s20
|
|
; GFX9-NEXT: v_mov_b32_e32 v5, s21
|
|
; GFX9-NEXT: v_mov_b32_e32 v6, s22
|
|
; GFX9-NEXT: v_mov_b32_e32 v7, s23
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: bitcast_v10f16_to_v10i16_scalar:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: s_mov_b32 s4, s16
|
|
; GFX11-NEXT: s_cmp_lg_u32 s17, 0
|
|
; GFX11-NEXT: s_mov_b32 s8, 0
|
|
; GFX11-NEXT: s_cbranch_scc0 .LBB23_3
|
|
; GFX11-NEXT: ; %bb.1: ; %Flow
|
|
; GFX11-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s8
|
|
; GFX11-NEXT: s_cbranch_vccnz .LBB23_4
|
|
; GFX11-NEXT: .LBB23_2: ; %cmp.true
|
|
; GFX11-NEXT: v_pk_add_f16 v4, 0x200, s4 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v3, 0x200, s3 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v2, 0x200, s2 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v1, 0x200, s1 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: v_pk_add_f16 v0, 0x200, s0 op_sel_hi:[0,1]
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
; GFX11-NEXT: .LBB23_3:
|
|
; GFX11-NEXT: s_branch .LBB23_2
|
|
; GFX11-NEXT: .LBB23_4:
|
|
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
|
|
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
|
|
; GFX11-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s5
|
|
; GFX11-NEXT: v_dual_mov_b32 v6, s6 :: v_dual_mov_b32 v7, s7
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%cmp = icmp eq i32 %b, 0
|
|
br i1 %cmp, label %cmp.true, label %cmp.false
|
|
|
|
cmp.true:
|
|
%a1 = fadd <10 x half> %a, splat (half 0xH0200)
|
|
%a2 = bitcast <10 x half> %a1 to <10 x i16>
|
|
br label %end
|
|
|
|
cmp.false:
|
|
%a3 = bitcast <10 x half> %a to <10 x i16>
|
|
br label %end
|
|
|
|
end:
|
|
%phi = phi <10 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
|
|
ret <10 x i16> %phi
|
|
}
|
|
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
|
|
; GFX11-FAKE16: {{.*}}
|
|
; GFX11-TRUE16: {{.*}}
|