xref: /llvm-project/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.intersect_ray.ll (revision 2d6d723a85c2d007b0359c206d66cd2e5a9f00e1) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; TODO: Run these for global isel as well.
3; RUN: llc -mtriple=amdgcn -mcpu=gfx1013 -verify-machineinstrs < %s | FileCheck -check-prefixes=PRE-GFX12,GFX10,GFX1013 %s
4; RUN: llc -mtriple=amdgcn -mcpu=gfx1030 -verify-machineinstrs < %s | FileCheck -check-prefixes=PRE-GFX12,GFX10,GFX1030 %s
5; RUN: not --crash llc -mtriple=amdgcn -mcpu=gfx1012 -verify-machineinstrs < %s 2>&1 | FileCheck -check-prefix=ERR %s
6; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=PRE-GFX12,GFX11 %s
7; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
8; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
9
10; uint4 llvm.amdgcn.image.bvh.intersect.ray.i32.v4f32(uint node_ptr, float ray_extent, float3 ray_origin, float3 ray_dir, float3 ray_inv_dir, uint4 texture_descr)
11; uint4 llvm.amdgcn.image.bvh.intersect.ray.i32.v4f16(uint node_ptr, float ray_extent, float3 ray_origin, half3 ray_dir, half3 ray_inv_dir, uint4 texture_descr)
12; uint4 llvm.amdgcn.image.bvh.intersect.ray.i64.v4f32(ulong node_ptr, float ray_extent, float3 ray_origin, float3 ray_dir, float3 ray_inv_dir, uint4 texture_descr)
13; uint4 llvm.amdgcn.image.bvh.intersect.ray.i64.v4f16(ulong node_ptr, float ray_extent, float3 ray_origin, half3 ray_dir, half3 ray_inv_dir, uint4 texture_descr)
14
15declare <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i32.v4f32(i32, float, <3 x float>, <3 x float>, <3 x float>, <4 x i32>)
16declare <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i32.v4f16(i32, float, <3 x float>, <3 x half>, <3 x half>, <4 x i32>)
17declare <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i64.v4f32(i64, float, <3 x float>, <3 x float>, <3 x float>, <4 x i32>)
18declare <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i64.v4f16(i64, float, <3 x float>, <3 x half>, <3 x half>, <4 x i32>)
19
20; ERR: in function image_bvh_intersect_ray{{.*}}intrinsic not supported on subtarget
21; Arguments are flattened to represent the actual VGPR_A layout, so we have no
22; extra moves in the generated kernel.
23define amdgpu_ps <4 x float> @image_bvh_intersect_ray(i32 %node_ptr, float %ray_extent, float %ray_origin_x, float %ray_origin_y, float %ray_origin_z, float %ray_dir_x, float %ray_dir_y, float %ray_dir_z, float %ray_inv_dir_x, float %ray_inv_dir_y, float %ray_inv_dir_z, <4 x i32> inreg %tdescr) {
24; PRE-GFX12-LABEL: image_bvh_intersect_ray:
25; PRE-GFX12:       ; %bb.0: ; %main_body
26; PRE-GFX12-NEXT:    image_bvh_intersect_ray v[0:3], v[0:10], s[0:3]
27; PRE-GFX12-NEXT:    s_waitcnt vmcnt(0)
28; PRE-GFX12-NEXT:    ; return to shader part epilog
29;
30; GFX12-LABEL: image_bvh_intersect_ray:
31; GFX12:       ; %bb.0: ; %main_body
32; GFX12-NEXT:    image_bvh_intersect_ray v[0:3], [v0, v1, v[2:4], v[5:7], v[8:10]], s[0:3]
33; GFX12-NEXT:    s_wait_bvhcnt 0x0
34; GFX12-NEXT:    ; return to shader part epilog
35main_body:
36  %ray_origin0 = insertelement <3 x float> undef, float %ray_origin_x, i32 0
37  %ray_origin1 = insertelement <3 x float> %ray_origin0, float %ray_origin_y, i32 1
38  %ray_origin = insertelement <3 x float> %ray_origin1, float %ray_origin_z, i32 2
39  %ray_dir0 = insertelement <3 x float> undef, float %ray_dir_x, i32 0
40  %ray_dir1 = insertelement <3 x float> %ray_dir0, float %ray_dir_y, i32 1
41  %ray_dir = insertelement <3 x float> %ray_dir1, float %ray_dir_z, i32 2
42  %ray_inv_dir0 = insertelement <3 x float> undef, float %ray_inv_dir_x, i32 0
43  %ray_inv_dir1 = insertelement <3 x float> %ray_inv_dir0, float %ray_inv_dir_y, i32 1
44  %ray_inv_dir = insertelement <3 x float> %ray_inv_dir1, float %ray_inv_dir_z, i32 2
45  %v = call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i32.v4f32(i32 %node_ptr, float %ray_extent, <3 x float> %ray_origin, <3 x float> %ray_dir, <3 x float> %ray_inv_dir, <4 x i32> %tdescr)
46 %r = bitcast <4 x i32> %v to <4 x float>
47 ret <4 x float> %r
48}
49
50define amdgpu_ps <4 x float> @image_bvh_intersect_ray_a16(i32 inreg %node_ptr, float inreg %ray_extent, <3 x float> inreg %ray_origin, <3 x half> inreg %ray_dir, <3 x half> inreg %ray_inv_dir, <4 x i32> inreg %tdescr) {
51; GFX10-LABEL: image_bvh_intersect_ray_a16:
52; GFX10:       ; %bb.0: ; %main_body
53; GFX10-NEXT:    s_mov_b32 s15, s12
54; GFX10-NEXT:    s_mov_b32 s12, s9
55; GFX10-NEXT:    s_lshr_b32 s9, s7, 16
56; GFX10-NEXT:    s_pack_ll_b32_b16 s6, s6, s7
57; GFX10-NEXT:    s_pack_ll_b32_b16 s7, s9, s8
58; GFX10-NEXT:    v_mov_b32_e32 v0, s0
59; GFX10-NEXT:    v_mov_b32_e32 v1, s1
60; GFX10-NEXT:    v_mov_b32_e32 v2, s2
61; GFX10-NEXT:    v_mov_b32_e32 v3, s3
62; GFX10-NEXT:    v_mov_b32_e32 v4, s4
63; GFX10-NEXT:    v_mov_b32_e32 v5, s5
64; GFX10-NEXT:    v_mov_b32_e32 v6, s6
65; GFX10-NEXT:    v_mov_b32_e32 v7, s7
66; GFX10-NEXT:    s_mov_b32 s14, s11
67; GFX10-NEXT:    s_mov_b32 s13, s10
68; GFX10-NEXT:    image_bvh_intersect_ray v[0:3], v[0:7], s[12:15] a16
69; GFX10-NEXT:    s_waitcnt vmcnt(0)
70; GFX10-NEXT:    ; return to shader part epilog
71;
72; GFX11-LABEL: image_bvh_intersect_ray_a16:
73; GFX11:       ; %bb.0: ; %main_body
74; GFX11-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
75; GFX11-NEXT:    s_lshr_b32 s2, s7, 16
76; GFX11-NEXT:    s_lshr_b32 s3, s5, 16
77; GFX11-NEXT:    v_dual_mov_b32 v6, s0 :: v_dual_mov_b32 v7, s1
78; GFX11-NEXT:    s_pack_ll_b32_b16 s2, s3, s2
79; GFX11-NEXT:    s_pack_ll_b32_b16 s3, s5, s7
80; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
81; GFX11-NEXT:    v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s3
82; GFX11-NEXT:    s_pack_ll_b32_b16 s4, s6, s8
83; GFX11-NEXT:    v_dual_mov_b32 v4, s2 :: v_dual_mov_b32 v5, s4
84; GFX11-NEXT:    s_mov_b32 s15, s12
85; GFX11-NEXT:    s_mov_b32 s14, s11
86; GFX11-NEXT:    s_mov_b32 s13, s10
87; GFX11-NEXT:    s_mov_b32 s12, s9
88; GFX11-NEXT:    image_bvh_intersect_ray v[0:3], [v6, v7, v[0:2], v[3:5]], s[12:15] a16
89; GFX11-NEXT:    s_waitcnt vmcnt(0)
90; GFX11-NEXT:    ; return to shader part epilog
91;
92; GFX12-SDAG-LABEL: image_bvh_intersect_ray_a16:
93; GFX12-SDAG:       ; %bb.0: ; %main_body
94; GFX12-SDAG-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
95; GFX12-SDAG-NEXT:    s_lshr_b32 s2, s7, 16
96; GFX12-SDAG-NEXT:    s_lshr_b32 s3, s5, 16
97; GFX12-SDAG-NEXT:    v_dual_mov_b32 v6, s0 :: v_dual_mov_b32 v7, s1
98; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
99; GFX12-SDAG-NEXT:    s_pack_ll_b32_b16 s2, s3, s2
100; GFX12-SDAG-NEXT:    s_pack_ll_b32_b16 s3, s5, s7
101; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
102; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s3
103; GFX12-SDAG-NEXT:    s_pack_ll_b32_b16 s4, s6, s8
104; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
105; GFX12-SDAG-NEXT:    v_dual_mov_b32 v4, s2 :: v_dual_mov_b32 v5, s4
106; GFX12-SDAG-NEXT:    s_mov_b32 s15, s12
107; GFX12-SDAG-NEXT:    s_mov_b32 s14, s11
108; GFX12-SDAG-NEXT:    s_mov_b32 s13, s10
109; GFX12-SDAG-NEXT:    s_mov_b32 s12, s9
110; GFX12-SDAG-NEXT:    image_bvh_intersect_ray v[0:3], [v6, v7, v[0:2], v[3:5]], s[12:15] a16
111; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
112; GFX12-SDAG-NEXT:    ; return to shader part epilog
113;
114; GFX12-GISEL-LABEL: image_bvh_intersect_ray_a16:
115; GFX12-GISEL:       ; %bb.0: ; %main_body
116; GFX12-GISEL-NEXT:    s_mov_b32 s20, s2
117; GFX12-GISEL-NEXT:    s_mov_b32 s22, s4
118; GFX12-GISEL-NEXT:    s_pack_ll_b32_b16 s4, s7, s5
119; GFX12-GISEL-NEXT:    s_mov_b32 s21, s3
120; GFX12-GISEL-NEXT:    s_pack_hh_b32_b16 s5, s7, s5
121; GFX12-GISEL-NEXT:    s_pack_ll_b32_b16 s6, s8, s6
122; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s20 :: v_dual_mov_b32 v1, s21
123; GFX12-GISEL-NEXT:    v_dual_mov_b32 v2, s22 :: v_dual_mov_b32 v3, s4
124; GFX12-GISEL-NEXT:    v_dual_mov_b32 v6, s0 :: v_dual_mov_b32 v7, s1
125; GFX12-GISEL-NEXT:    v_dual_mov_b32 v4, s5 :: v_dual_mov_b32 v5, s6
126; GFX12-GISEL-NEXT:    s_mov_b32 s16, s9
127; GFX12-GISEL-NEXT:    s_mov_b32 s17, s10
128; GFX12-GISEL-NEXT:    s_mov_b32 s18, s11
129; GFX12-GISEL-NEXT:    s_mov_b32 s19, s12
130; GFX12-GISEL-NEXT:    image_bvh_intersect_ray v[0:3], [v6, v7, v[0:2], v[3:5]], s[16:19] a16
131; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
132; GFX12-GISEL-NEXT:    ; return to shader part epilog
133main_body:
134  %v = call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i32.v4f16(i32 %node_ptr, float %ray_extent, <3 x float> %ray_origin, <3 x half> %ray_dir, <3 x half> %ray_inv_dir, <4 x i32> %tdescr)
135  %r = bitcast <4 x i32> %v to <4 x float>
136  ret <4 x float> %r
137}
138
139; Arguments are flattened to represent the actual VGPR_A layout, so we have no
140; extra moves in the generated kernel.
141define amdgpu_ps <4 x float> @image_bvh64_intersect_ray(<2 x i32> %node_ptr_vec, float %ray_extent, float %ray_origin_x, float %ray_origin_y, float %ray_origin_z, float %ray_dir_x, float %ray_dir_y, float %ray_dir_z, float %ray_inv_dir_x, float %ray_inv_dir_y, float %ray_inv_dir_z, <4 x i32> inreg %tdescr) {
142; PRE-GFX12-LABEL: image_bvh64_intersect_ray:
143; PRE-GFX12:       ; %bb.0: ; %main_body
144; PRE-GFX12-NEXT:    image_bvh64_intersect_ray v[0:3], v[0:11], s[0:3]
145; PRE-GFX12-NEXT:    s_waitcnt vmcnt(0)
146; PRE-GFX12-NEXT:    ; return to shader part epilog
147;
148; GFX12-LABEL: image_bvh64_intersect_ray:
149; GFX12:       ; %bb.0: ; %main_body
150; GFX12-NEXT:    image_bvh64_intersect_ray v[0:3], [v[0:1], v2, v[3:5], v[6:8], v[9:11]], s[0:3]
151; GFX12-NEXT:    s_wait_bvhcnt 0x0
152; GFX12-NEXT:    ; return to shader part epilog
153main_body:
154  %node_ptr = bitcast <2 x i32> %node_ptr_vec to i64
155  %ray_origin0 = insertelement <3 x float> undef, float %ray_origin_x, i32 0
156  %ray_origin1 = insertelement <3 x float> %ray_origin0, float %ray_origin_y, i32 1
157  %ray_origin = insertelement <3 x float> %ray_origin1, float %ray_origin_z, i32 2
158  %ray_dir0 = insertelement <3 x float> undef, float %ray_dir_x, i32 0
159  %ray_dir1 = insertelement <3 x float> %ray_dir0, float %ray_dir_y, i32 1
160  %ray_dir = insertelement <3 x float> %ray_dir1, float %ray_dir_z, i32 2
161  %ray_inv_dir0 = insertelement <3 x float> undef, float %ray_inv_dir_x, i32 0
162  %ray_inv_dir1 = insertelement <3 x float> %ray_inv_dir0, float %ray_inv_dir_y, i32 1
163  %ray_inv_dir = insertelement <3 x float> %ray_inv_dir1, float %ray_inv_dir_z, i32 2
164  %v = call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i64.v4f32(i64 %node_ptr, float %ray_extent, <3 x float> %ray_origin, <3 x float> %ray_dir, <3 x float> %ray_inv_dir, <4 x i32> %tdescr)
165 %r = bitcast <4 x i32> %v to <4 x float>
166 ret <4 x float> %r
167}
168
169define amdgpu_ps <4 x float> @image_bvh64_intersect_ray_a16(i64 inreg %node_ptr, float inreg %ray_extent, <3 x float> inreg %ray_origin, <3 x half> inreg %ray_dir, <3 x half> inreg %ray_inv_dir, <4 x i32> inreg %tdescr) {
170; GFX10-LABEL: image_bvh64_intersect_ray_a16:
171; GFX10:       ; %bb.0: ; %main_body
172; GFX10-NEXT:    s_mov_b32 s14, s12
173; GFX10-NEXT:    s_mov_b32 s12, s10
174; GFX10-NEXT:    s_lshr_b32 s10, s8, 16
175; GFX10-NEXT:    s_pack_ll_b32_b16 s7, s7, s8
176; GFX10-NEXT:    s_pack_ll_b32_b16 s8, s10, s9
177; GFX10-NEXT:    v_mov_b32_e32 v0, s0
178; GFX10-NEXT:    v_mov_b32_e32 v1, s1
179; GFX10-NEXT:    v_mov_b32_e32 v2, s2
180; GFX10-NEXT:    v_mov_b32_e32 v3, s3
181; GFX10-NEXT:    v_mov_b32_e32 v4, s4
182; GFX10-NEXT:    v_mov_b32_e32 v5, s5
183; GFX10-NEXT:    v_mov_b32_e32 v6, s6
184; GFX10-NEXT:    v_mov_b32_e32 v7, s7
185; GFX10-NEXT:    v_mov_b32_e32 v8, s8
186; GFX10-NEXT:    s_mov_b32 s15, s13
187; GFX10-NEXT:    s_mov_b32 s13, s11
188; GFX10-NEXT:    image_bvh64_intersect_ray v[0:3], v[0:8], s[12:15] a16
189; GFX10-NEXT:    s_waitcnt vmcnt(0)
190; GFX10-NEXT:    ; return to shader part epilog
191;
192; GFX11-LABEL: image_bvh64_intersect_ray_a16:
193; GFX11:       ; %bb.0: ; %main_body
194; GFX11-NEXT:    v_dual_mov_b32 v0, s3 :: v_dual_mov_b32 v1, s4
195; GFX11-NEXT:    v_dual_mov_b32 v2, s5 :: v_dual_mov_b32 v7, s1
196; GFX11-NEXT:    s_lshr_b32 s3, s6, 16
197; GFX11-NEXT:    s_pack_ll_b32_b16 s1, s6, s8
198; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
199; GFX11-NEXT:    v_dual_mov_b32 v6, s0 :: v_dual_mov_b32 v3, s1
200; GFX11-NEXT:    s_lshr_b32 s0, s8, 16
201; GFX11-NEXT:    v_mov_b32_e32 v8, s2
202; GFX11-NEXT:    s_pack_ll_b32_b16 s0, s3, s0
203; GFX11-NEXT:    s_pack_ll_b32_b16 s3, s7, s9
204; GFX11-NEXT:    v_dual_mov_b32 v4, s0 :: v_dual_mov_b32 v5, s3
205; GFX11-NEXT:    s_mov_b32 s15, s13
206; GFX11-NEXT:    s_mov_b32 s14, s12
207; GFX11-NEXT:    s_mov_b32 s13, s11
208; GFX11-NEXT:    s_mov_b32 s12, s10
209; GFX11-NEXT:    image_bvh64_intersect_ray v[0:3], [v[6:7], v8, v[0:2], v[3:5]], s[12:15] a16
210; GFX11-NEXT:    s_waitcnt vmcnt(0)
211; GFX11-NEXT:    ; return to shader part epilog
212;
213; GFX12-SDAG-LABEL: image_bvh64_intersect_ray_a16:
214; GFX12-SDAG:       ; %bb.0: ; %main_body
215; GFX12-SDAG-NEXT:    v_dual_mov_b32 v0, s3 :: v_dual_mov_b32 v1, s4
216; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, s5 :: v_dual_mov_b32 v7, s1
217; GFX12-SDAG-NEXT:    s_lshr_b32 s3, s6, 16
218; GFX12-SDAG-NEXT:    s_pack_ll_b32_b16 s1, s6, s8
219; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
220; GFX12-SDAG-NEXT:    v_dual_mov_b32 v6, s0 :: v_dual_mov_b32 v3, s1
221; GFX12-SDAG-NEXT:    s_lshr_b32 s0, s8, 16
222; GFX12-SDAG-NEXT:    v_mov_b32_e32 v8, s2
223; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
224; GFX12-SDAG-NEXT:    s_pack_ll_b32_b16 s0, s3, s0
225; GFX12-SDAG-NEXT:    s_pack_ll_b32_b16 s3, s7, s9
226; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
227; GFX12-SDAG-NEXT:    v_dual_mov_b32 v4, s0 :: v_dual_mov_b32 v5, s3
228; GFX12-SDAG-NEXT:    s_mov_b32 s15, s13
229; GFX12-SDAG-NEXT:    s_mov_b32 s14, s12
230; GFX12-SDAG-NEXT:    s_mov_b32 s13, s11
231; GFX12-SDAG-NEXT:    s_mov_b32 s12, s10
232; GFX12-SDAG-NEXT:    image_bvh64_intersect_ray v[0:3], [v[6:7], v8, v[0:2], v[3:5]], s[12:15] a16
233; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
234; GFX12-SDAG-NEXT:    ; return to shader part epilog
235;
236; GFX12-GISEL-LABEL: image_bvh64_intersect_ray_a16:
237; GFX12-GISEL:       ; %bb.0: ; %main_body
238; GFX12-GISEL-NEXT:    s_mov_b32 s20, s3
239; GFX12-GISEL-NEXT:    s_mov_b32 s21, s4
240; GFX12-GISEL-NEXT:    s_pack_ll_b32_b16 s4, s8, s6
241; GFX12-GISEL-NEXT:    s_mov_b32 s22, s5
242; GFX12-GISEL-NEXT:    s_pack_hh_b32_b16 s5, s8, s6
243; GFX12-GISEL-NEXT:    s_pack_ll_b32_b16 s6, s9, s7
244; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s20 :: v_dual_mov_b32 v3, s4
245; GFX12-GISEL-NEXT:    v_dual_mov_b32 v7, s1 :: v_dual_mov_b32 v6, s0
246; GFX12-GISEL-NEXT:    v_dual_mov_b32 v8, s2 :: v_dual_mov_b32 v1, s21
247; GFX12-GISEL-NEXT:    v_dual_mov_b32 v2, s22 :: v_dual_mov_b32 v5, s6
248; GFX12-GISEL-NEXT:    v_mov_b32_e32 v4, s5
249; GFX12-GISEL-NEXT:    s_mov_b32 s16, s10
250; GFX12-GISEL-NEXT:    s_mov_b32 s17, s11
251; GFX12-GISEL-NEXT:    s_mov_b32 s18, s12
252; GFX12-GISEL-NEXT:    s_mov_b32 s19, s13
253; GFX12-GISEL-NEXT:    image_bvh64_intersect_ray v[0:3], [v[6:7], v8, v[0:2], v[3:5]], s[16:19] a16
254; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
255; GFX12-GISEL-NEXT:    ; return to shader part epilog
256main_body:
257  %v = call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i64.v4f16(i64 %node_ptr, float %ray_extent, <3 x float> %ray_origin, <3 x half> %ray_dir, <3 x half> %ray_inv_dir, <4 x i32> %tdescr)
258  %r = bitcast <4 x i32> %v to <4 x float>
259  ret <4 x float> %r
260}
261
262; TODO: NSA reassign is very limited and cannot work with VGPR tuples and subregs.
263
264define amdgpu_kernel void @image_bvh_intersect_ray_nsa_reassign(ptr %p_node_ptr, ptr %p_ray, <4 x i32> inreg %tdescr) {
265; GFX1013-LABEL: image_bvh_intersect_ray_nsa_reassign:
266; GFX1013:       ; %bb.0: ; %main_body
267; GFX1013-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x24
268; GFX1013-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
269; GFX1013-NEXT:    v_mov_b32_e32 v6, 4.0
270; GFX1013-NEXT:    v_mov_b32_e32 v7, 0x40a00000
271; GFX1013-NEXT:    v_mov_b32_e32 v8, 0x40c00000
272; GFX1013-NEXT:    v_mov_b32_e32 v9, 0x40e00000
273; GFX1013-NEXT:    v_mov_b32_e32 v10, 0x41000000
274; GFX1013-NEXT:    s_waitcnt lgkmcnt(0)
275; GFX1013-NEXT:    v_add_co_u32 v2, s0, s8, v0
276; GFX1013-NEXT:    v_add_co_ci_u32_e64 v3, s0, s9, 0, s0
277; GFX1013-NEXT:    v_add_co_u32 v4, s0, s10, v0
278; GFX1013-NEXT:    v_add_co_ci_u32_e64 v5, s0, s11, 0, s0
279; GFX1013-NEXT:    flat_load_dword v0, v[2:3]
280; GFX1013-NEXT:    flat_load_dword v1, v[4:5]
281; GFX1013-NEXT:    v_mov_b32_e32 v2, 0
282; GFX1013-NEXT:    v_mov_b32_e32 v3, 1.0
283; GFX1013-NEXT:    v_mov_b32_e32 v4, 2.0
284; GFX1013-NEXT:    v_mov_b32_e32 v5, 0x40400000
285; GFX1013-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
286; GFX1013-NEXT:    image_bvh_intersect_ray v[0:3], v[0:10], s[12:15]
287; GFX1013-NEXT:    s_waitcnt vmcnt(0)
288; GFX1013-NEXT:    flat_store_dwordx4 v[0:1], v[0:3]
289; GFX1013-NEXT:    s_endpgm
290;
291; GFX1030-LABEL: image_bvh_intersect_ray_nsa_reassign:
292; GFX1030:       ; %bb.0: ; %main_body
293; GFX1030-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x24
294; GFX1030-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
295; GFX1030-NEXT:    v_mov_b32_e32 v10, 0x41000000
296; GFX1030-NEXT:    v_mov_b32_e32 v9, 0x40e00000
297; GFX1030-NEXT:    v_mov_b32_e32 v8, 0x40c00000
298; GFX1030-NEXT:    v_mov_b32_e32 v7, 0x40a00000
299; GFX1030-NEXT:    v_mov_b32_e32 v6, 4.0
300; GFX1030-NEXT:    v_mov_b32_e32 v5, 0x40400000
301; GFX1030-NEXT:    v_mov_b32_e32 v4, 2.0
302; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
303; GFX1030-NEXT:    v_add_co_u32 v0, s0, s0, v2
304; GFX1030-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
305; GFX1030-NEXT:    v_add_co_u32 v2, s0, s2, v2
306; GFX1030-NEXT:    v_add_co_ci_u32_e64 v3, null, s3, 0, s0
307; GFX1030-NEXT:    flat_load_dword v0, v[0:1]
308; GFX1030-NEXT:    flat_load_dword v1, v[2:3]
309; GFX1030-NEXT:    v_mov_b32_e32 v2, 0
310; GFX1030-NEXT:    v_mov_b32_e32 v3, 1.0
311; GFX1030-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
312; GFX1030-NEXT:    image_bvh_intersect_ray v[0:3], v[0:10], s[4:7]
313; GFX1030-NEXT:    s_waitcnt vmcnt(0)
314; GFX1030-NEXT:    flat_store_dwordx4 v[0:1], v[0:3]
315; GFX1030-NEXT:    s_endpgm
316;
317; GFX11-LABEL: image_bvh_intersect_ray_nsa_reassign:
318; GFX11:       ; %bb.0: ; %main_body
319; GFX11-NEXT:    s_load_b256 s[0:7], s[4:5], 0x24
320; GFX11-NEXT:    v_dual_mov_b32 v7, 1.0 :: v_dual_and_b32 v0, 0x3ff, v0
321; GFX11-NEXT:    v_dual_mov_b32 v5, 0x40a00000 :: v_dual_mov_b32 v6, 0
322; GFX11-NEXT:    v_mov_b32_e32 v8, 2.0
323; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
324; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
325; GFX11-NEXT:    v_mov_b32_e32 v4, 4.0
326; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
327; GFX11-NEXT:    v_add_co_u32 v0, s0, s0, v2
328; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
329; GFX11-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
330; GFX11-NEXT:    v_add_co_u32 v2, s0, s2, v2
331; GFX11-NEXT:    v_add_co_ci_u32_e64 v3, null, s3, 0, s0
332; GFX11-NEXT:    flat_load_b32 v9, v[0:1]
333; GFX11-NEXT:    flat_load_b32 v10, v[2:3]
334; GFX11-NEXT:    v_mov_b32_e32 v1, 0x40e00000
335; GFX11-NEXT:    v_mov_b32_e32 v0, 0x40c00000
336; GFX11-NEXT:    v_mov_b32_e32 v2, 0x41000000
337; GFX11-NEXT:    v_mov_b32_e32 v3, 0x40400000
338; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
339; GFX11-NEXT:    image_bvh_intersect_ray v[0:3], [v9, v10, v[6:8], v[3:5], v[0:2]], s[4:7]
340; GFX11-NEXT:    s_waitcnt vmcnt(0)
341; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
342; GFX11-NEXT:    s_endpgm
343;
344; GFX12-SDAG-LABEL: image_bvh_intersect_ray_nsa_reassign:
345; GFX12-SDAG:       ; %bb.0: ; %main_body
346; GFX12-SDAG-NEXT:    s_load_b256 s[0:7], s[4:5], 0x24
347; GFX12-SDAG-NEXT:    v_dual_mov_b32 v7, 1.0 :: v_dual_and_b32 v0, 0x3ff, v0
348; GFX12-SDAG-NEXT:    v_dual_mov_b32 v5, 0x40a00000 :: v_dual_mov_b32 v6, 0
349; GFX12-SDAG-NEXT:    v_mov_b32_e32 v8, 2.0
350; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
351; GFX12-SDAG-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
352; GFX12-SDAG-NEXT:    v_mov_b32_e32 v4, 4.0
353; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
354; GFX12-SDAG-NEXT:    v_add_co_u32 v0, s0, s0, v2
355; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
356; GFX12-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
357; GFX12-SDAG-NEXT:    v_add_co_u32 v2, s0, s2, v2
358; GFX12-SDAG-NEXT:    v_add_co_ci_u32_e64 v3, null, s3, 0, s0
359; GFX12-SDAG-NEXT:    flat_load_b32 v9, v[0:1]
360; GFX12-SDAG-NEXT:    flat_load_b32 v10, v[2:3]
361; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, 0x40e00000
362; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0x40c00000
363; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0x41000000
364; GFX12-SDAG-NEXT:    v_mov_b32_e32 v3, 0x40400000
365; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
366; GFX12-SDAG-NEXT:    image_bvh_intersect_ray v[0:3], [v9, v10, v[6:8], v[3:5], v[0:2]], s[4:7]
367; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
368; GFX12-SDAG-NEXT:    flat_store_b128 v[0:1], v[0:3]
369; GFX12-SDAG-NEXT:    s_endpgm
370;
371; GFX12-GISEL-LABEL: image_bvh_intersect_ray_nsa_reassign:
372; GFX12-GISEL:       ; %bb.0: ; %main_body
373; GFX12-GISEL-NEXT:    s_load_b256 s[0:7], s[4:5], 0x24
374; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
375; GFX12-GISEL-NEXT:    s_mov_b32 s10, 0x40a00000
376; GFX12-GISEL-NEXT:    s_mov_b32 s9, 4.0
377; GFX12-GISEL-NEXT:    s_mov_b32 s8, 0x40400000
378; GFX12-GISEL-NEXT:    s_mov_b32 s12, 0x40c00000
379; GFX12-GISEL-NEXT:    v_lshlrev_b32_e32 v4, 2, v0
380; GFX12-GISEL-NEXT:    s_mov_b32 s14, 0x41000000
381; GFX12-GISEL-NEXT:    s_mov_b32 s13, 0x40e00000
382; GFX12-GISEL-NEXT:    v_mov_b32_e32 v6, s12
383; GFX12-GISEL-NEXT:    v_dual_mov_b32 v8, s14 :: v_dual_mov_b32 v7, s13
384; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
385; GFX12-GISEL-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
386; GFX12-GISEL-NEXT:    s_mov_b32 s2, 2.0
387; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
388; GFX12-GISEL-NEXT:    s_mov_b32 s0, 0
389; GFX12-GISEL-NEXT:    s_mov_b32 s1, 1.0
390; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
391; GFX12-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, v0, v4
392; GFX12-GISEL-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
393; GFX12-GISEL-NEXT:    v_add_co_u32 v2, vcc_lo, v2, v4
394; GFX12-GISEL-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
395; GFX12-GISEL-NEXT:    flat_load_b32 v9, v[0:1]
396; GFX12-GISEL-NEXT:    flat_load_b32 v10, v[2:3]
397; GFX12-GISEL-NEXT:    s_wait_alu 0xfffe
398; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v3, s8
399; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, s1 :: v_dual_mov_b32 v2, s2
400; GFX12-GISEL-NEXT:    v_dual_mov_b32 v4, s9 :: v_dual_mov_b32 v5, s10
401; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
402; GFX12-GISEL-NEXT:    image_bvh_intersect_ray v[0:3], [v9, v10, v[0:2], v[3:5], v[6:8]], s[4:7]
403; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
404; GFX12-GISEL-NEXT:    flat_store_b128 v[0:1], v[0:3]
405; GFX12-GISEL-NEXT:    s_endpgm
406main_body:
407  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
408  %gep_node_ptr = getelementptr inbounds i32, ptr %p_node_ptr, i32 %lid
409  %node_ptr = load i32, ptr %gep_node_ptr, align 4
410  %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
411  %ray_extent = load float, ptr %gep_ray, align 4
412  %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
413  %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1
414  %ray_origin = insertelement <3 x float> %ray_origin1, float 2.0, i32 2
415  %ray_dir0 = insertelement <3 x float> undef, float 3.0, i32 0
416  %ray_dir1 = insertelement <3 x float> %ray_dir0, float 4.0, i32 1
417  %ray_dir = insertelement <3 x float> %ray_dir1, float 5.0, i32 2
418  %ray_inv_dir0 = insertelement <3 x float> undef, float 6.0, i32 0
419  %ray_inv_dir1 = insertelement <3 x float> %ray_inv_dir0, float 7.0, i32 1
420  %ray_inv_dir = insertelement <3 x float> %ray_inv_dir1, float 8.0, i32 2
421  %v = call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i32.v4f32(i32 %node_ptr, float %ray_extent, <3 x float> %ray_origin, <3 x float> %ray_dir, <3 x float> %ray_inv_dir, <4 x i32> %tdescr)
422  store <4 x i32> %v, ptr undef
423  ret void
424}
425
426define amdgpu_kernel void @image_bvh_intersect_ray_a16_nsa_reassign(ptr %p_node_ptr, ptr %p_ray, <4 x i32> inreg %tdescr) {
427; GFX1013-LABEL: image_bvh_intersect_ray_a16_nsa_reassign:
428; GFX1013:       ; %bb.0: ; %main_body
429; GFX1013-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x24
430; GFX1013-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
431; GFX1013-NEXT:    v_mov_b32_e32 v6, 0x46004500
432; GFX1013-NEXT:    v_mov_b32_e32 v7, 0x48004700
433; GFX1013-NEXT:    s_waitcnt lgkmcnt(0)
434; GFX1013-NEXT:    v_add_co_u32 v2, s0, s8, v0
435; GFX1013-NEXT:    v_add_co_ci_u32_e64 v3, s0, s9, 0, s0
436; GFX1013-NEXT:    v_add_co_u32 v4, s0, s10, v0
437; GFX1013-NEXT:    v_add_co_ci_u32_e64 v5, s0, s11, 0, s0
438; GFX1013-NEXT:    flat_load_dword v0, v[2:3]
439; GFX1013-NEXT:    flat_load_dword v1, v[4:5]
440; GFX1013-NEXT:    v_mov_b32_e32 v2, 0
441; GFX1013-NEXT:    v_mov_b32_e32 v3, 1.0
442; GFX1013-NEXT:    v_mov_b32_e32 v4, 2.0
443; GFX1013-NEXT:    v_mov_b32_e32 v5, 0x44004200
444; GFX1013-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
445; GFX1013-NEXT:    image_bvh_intersect_ray v[0:3], v[0:7], s[12:15] a16
446; GFX1013-NEXT:    s_waitcnt vmcnt(0)
447; GFX1013-NEXT:    flat_store_dwordx4 v[0:1], v[0:3]
448; GFX1013-NEXT:    s_endpgm
449;
450; GFX1030-LABEL: image_bvh_intersect_ray_a16_nsa_reassign:
451; GFX1030:       ; %bb.0: ; %main_body
452; GFX1030-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x24
453; GFX1030-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
454; GFX1030-NEXT:    v_mov_b32_e32 v4, 2.0
455; GFX1030-NEXT:    v_mov_b32_e32 v5, 0x44004200
456; GFX1030-NEXT:    v_mov_b32_e32 v6, 0x46004500
457; GFX1030-NEXT:    v_mov_b32_e32 v7, 0x48004700
458; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
459; GFX1030-NEXT:    v_add_co_u32 v0, s0, s0, v2
460; GFX1030-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
461; GFX1030-NEXT:    v_add_co_u32 v2, s0, s2, v2
462; GFX1030-NEXT:    v_add_co_ci_u32_e64 v3, null, s3, 0, s0
463; GFX1030-NEXT:    flat_load_dword v0, v[0:1]
464; GFX1030-NEXT:    flat_load_dword v1, v[2:3]
465; GFX1030-NEXT:    v_mov_b32_e32 v2, 0
466; GFX1030-NEXT:    v_mov_b32_e32 v3, 1.0
467; GFX1030-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
468; GFX1030-NEXT:    image_bvh_intersect_ray v[0:3], v[0:7], s[4:7] a16
469; GFX1030-NEXT:    s_waitcnt vmcnt(0)
470; GFX1030-NEXT:    flat_store_dwordx4 v[0:1], v[0:3]
471; GFX1030-NEXT:    s_endpgm
472;
473; GFX11-LABEL: image_bvh_intersect_ray_a16_nsa_reassign:
474; GFX11:       ; %bb.0: ; %main_body
475; GFX11-NEXT:    s_load_b256 s[0:7], s[4:5], 0x24
476; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
477; GFX11-NEXT:    v_dual_mov_b32 v4, 1.0 :: v_dual_mov_b32 v5, 2.0
478; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
479; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
480; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
481; GFX11-NEXT:    v_add_co_u32 v0, s0, s0, v2
482; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
483; GFX11-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
484; GFX11-NEXT:    v_add_co_u32 v2, s0, s2, v2
485; GFX11-NEXT:    v_add_co_ci_u32_e64 v3, null, s3, 0, s0
486; GFX11-NEXT:    flat_load_b32 v6, v[0:1]
487; GFX11-NEXT:    flat_load_b32 v7, v[2:3]
488; GFX11-NEXT:    v_mov_b32_e32 v1, 0x47004400
489; GFX11-NEXT:    v_mov_b32_e32 v0, 0x46004200
490; GFX11-NEXT:    v_dual_mov_b32 v2, 0x48004500 :: v_dual_mov_b32 v3, 0
491; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
492; GFX11-NEXT:    image_bvh_intersect_ray v[0:3], [v6, v7, v[3:5], v[0:2]], s[4:7] a16
493; GFX11-NEXT:    s_waitcnt vmcnt(0)
494; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
495; GFX11-NEXT:    s_endpgm
496;
497; GFX12-SDAG-LABEL: image_bvh_intersect_ray_a16_nsa_reassign:
498; GFX12-SDAG:       ; %bb.0: ; %main_body
499; GFX12-SDAG-NEXT:    s_load_b256 s[0:7], s[4:5], 0x24
500; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
501; GFX12-SDAG-NEXT:    v_dual_mov_b32 v4, 1.0 :: v_dual_mov_b32 v5, 2.0
502; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
503; GFX12-SDAG-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
504; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
505; GFX12-SDAG-NEXT:    v_add_co_u32 v0, s0, s0, v2
506; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
507; GFX12-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
508; GFX12-SDAG-NEXT:    v_add_co_u32 v2, s0, s2, v2
509; GFX12-SDAG-NEXT:    v_add_co_ci_u32_e64 v3, null, s3, 0, s0
510; GFX12-SDAG-NEXT:    flat_load_b32 v6, v[0:1]
511; GFX12-SDAG-NEXT:    flat_load_b32 v7, v[2:3]
512; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, 0x47004400
513; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0x46004200
514; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0x48004500 :: v_dual_mov_b32 v3, 0
515; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
516; GFX12-SDAG-NEXT:    image_bvh_intersect_ray v[0:3], [v6, v7, v[3:5], v[0:2]], s[4:7] a16
517; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
518; GFX12-SDAG-NEXT:    flat_store_b128 v[0:1], v[0:3]
519; GFX12-SDAG-NEXT:    s_endpgm
520;
521; GFX12-GISEL-LABEL: image_bvh_intersect_ray_a16_nsa_reassign:
522; GFX12-GISEL:       ; %bb.0: ; %main_body
523; GFX12-GISEL-NEXT:    s_load_b256 s[0:7], s[4:5], 0x24
524; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
525; GFX12-GISEL-NEXT:    s_mov_b32 s8, 0x42004600
526; GFX12-GISEL-NEXT:    s_mov_b32 s9, 0x44004700
527; GFX12-GISEL-NEXT:    s_mov_b32 s10, 0x45004800
528; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
529; GFX12-GISEL-NEXT:    v_lshlrev_b32_e32 v4, 2, v0
530; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
531; GFX12-GISEL-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
532; GFX12-GISEL-NEXT:    s_mov_b32 s2, 2.0
533; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
534; GFX12-GISEL-NEXT:    s_mov_b32 s0, 0
535; GFX12-GISEL-NEXT:    s_mov_b32 s1, 1.0
536; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
537; GFX12-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, v0, v4
538; GFX12-GISEL-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
539; GFX12-GISEL-NEXT:    v_add_co_u32 v2, vcc_lo, v2, v4
540; GFX12-GISEL-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
541; GFX12-GISEL-NEXT:    flat_load_b32 v6, v[0:1]
542; GFX12-GISEL-NEXT:    flat_load_b32 v7, v[2:3]
543; GFX12-GISEL-NEXT:    s_wait_alu 0xfffe
544; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v3, s8
545; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, s1 :: v_dual_mov_b32 v2, s2
546; GFX12-GISEL-NEXT:    v_dual_mov_b32 v4, s9 :: v_dual_mov_b32 v5, s10
547; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
548; GFX12-GISEL-NEXT:    image_bvh_intersect_ray v[0:3], [v6, v7, v[0:2], v[3:5]], s[4:7] a16
549; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
550; GFX12-GISEL-NEXT:    flat_store_b128 v[0:1], v[0:3]
551; GFX12-GISEL-NEXT:    s_endpgm
552main_body:
553  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
554  %gep_node_ptr = getelementptr inbounds i32, ptr %p_node_ptr, i32 %lid
555  %node_ptr = load i32, ptr %gep_node_ptr, align 4
556  %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
557  %ray_extent = load float, ptr %gep_ray, align 4
558  %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
559  %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1
560  %ray_origin = insertelement <3 x float> %ray_origin1, float 2.0, i32 2
561  %ray_dir0 = insertelement <3 x half> undef, half 3.0, i32 0
562  %ray_dir1 = insertelement <3 x half> %ray_dir0, half 4.0, i32 1
563  %ray_dir = insertelement <3 x half> %ray_dir1, half 5.0, i32 2
564  %ray_inv_dir0 = insertelement <3 x half> undef, half 6.0, i32 0
565  %ray_inv_dir1 = insertelement <3 x half> %ray_inv_dir0, half 7.0, i32 1
566  %ray_inv_dir = insertelement <3 x half> %ray_inv_dir1, half 8.0, i32 2
567  %v = call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i32.v4f16(i32 %node_ptr, float %ray_extent, <3 x float> %ray_origin, <3 x half> %ray_dir, <3 x half> %ray_inv_dir, <4 x i32> %tdescr)
568  store <4 x i32> %v, ptr undef
569  ret void
570}
571
572define amdgpu_kernel void @image_bvh64_intersect_ray_nsa_reassign(ptr %p_ray, <4 x i32> inreg %tdescr) {
573; GFX1013-LABEL: image_bvh64_intersect_ray_nsa_reassign:
574; GFX1013:       ; %bb.0: ; %main_body
575; GFX1013-NEXT:    s_clause 0x1
576; GFX1013-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
577; GFX1013-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x34
578; GFX1013-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
579; GFX1013-NEXT:    v_mov_b32_e32 v3, 0
580; GFX1013-NEXT:    v_mov_b32_e32 v4, 1.0
581; GFX1013-NEXT:    v_mov_b32_e32 v5, 2.0
582; GFX1013-NEXT:    v_mov_b32_e32 v6, 0x40400000
583; GFX1013-NEXT:    v_mov_b32_e32 v7, 4.0
584; GFX1013-NEXT:    v_mov_b32_e32 v8, 0x40a00000
585; GFX1013-NEXT:    v_mov_b32_e32 v9, 0x40c00000
586; GFX1013-NEXT:    v_mov_b32_e32 v10, 0x40e00000
587; GFX1013-NEXT:    v_mov_b32_e32 v11, 0x41000000
588; GFX1013-NEXT:    s_waitcnt lgkmcnt(0)
589; GFX1013-NEXT:    v_add_co_u32 v0, s4, s6, v0
590; GFX1013-NEXT:    v_add_co_ci_u32_e64 v1, s4, s7, 0, s4
591; GFX1013-NEXT:    flat_load_dword v2, v[0:1]
592; GFX1013-NEXT:    v_mov_b32_e32 v0, 0xb36211c7
593; GFX1013-NEXT:    v_bfrev_b32_e32 v1, 4.0
594; GFX1013-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
595; GFX1013-NEXT:    image_bvh64_intersect_ray v[0:3], v[0:11], s[0:3]
596; GFX1013-NEXT:    s_waitcnt vmcnt(0)
597; GFX1013-NEXT:    flat_store_dwordx4 v[0:1], v[0:3]
598; GFX1013-NEXT:    s_endpgm
599;
600; GFX1030-LABEL: image_bvh64_intersect_ray_nsa_reassign:
601; GFX1030:       ; %bb.0: ; %main_body
602; GFX1030-NEXT:    s_clause 0x1
603; GFX1030-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
604; GFX1030-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x34
605; GFX1030-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
606; GFX1030-NEXT:    v_mov_b32_e32 v3, 0
607; GFX1030-NEXT:    v_mov_b32_e32 v11, 0x41000000
608; GFX1030-NEXT:    v_mov_b32_e32 v10, 0x40e00000
609; GFX1030-NEXT:    v_mov_b32_e32 v9, 0x40c00000
610; GFX1030-NEXT:    v_mov_b32_e32 v8, 0x40a00000
611; GFX1030-NEXT:    v_mov_b32_e32 v7, 4.0
612; GFX1030-NEXT:    v_mov_b32_e32 v6, 0x40400000
613; GFX1030-NEXT:    v_mov_b32_e32 v5, 2.0
614; GFX1030-NEXT:    v_mov_b32_e32 v4, 1.0
615; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
616; GFX1030-NEXT:    v_add_co_u32 v0, s4, s6, v0
617; GFX1030-NEXT:    v_add_co_ci_u32_e64 v1, null, s7, 0, s4
618; GFX1030-NEXT:    flat_load_dword v2, v[0:1]
619; GFX1030-NEXT:    v_bfrev_b32_e32 v1, 4.0
620; GFX1030-NEXT:    v_mov_b32_e32 v0, 0xb36211c7
621; GFX1030-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
622; GFX1030-NEXT:    image_bvh64_intersect_ray v[0:3], v[0:11], s[0:3]
623; GFX1030-NEXT:    s_waitcnt vmcnt(0)
624; GFX1030-NEXT:    flat_store_dwordx4 v[0:1], v[0:3]
625; GFX1030-NEXT:    s_endpgm
626;
627; GFX11-LABEL: image_bvh64_intersect_ray_nsa_reassign:
628; GFX11:       ; %bb.0: ; %main_body
629; GFX11-NEXT:    s_clause 0x1
630; GFX11-NEXT:    s_load_b64 s[6:7], s[4:5], 0x24
631; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
632; GFX11-NEXT:    v_dual_mov_b32 v7, 1.0 :: v_dual_and_b32 v0, 0x3ff, v0
633; GFX11-NEXT:    v_mov_b32_e32 v2, 0x41000000
634; GFX11-NEXT:    v_dual_mov_b32 v3, 0x40400000 :: v_dual_mov_b32 v4, 4.0
635; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
636; GFX11-NEXT:    v_dual_mov_b32 v5, 0x40a00000 :: v_dual_lshlrev_b32 v0, 2, v0
637; GFX11-NEXT:    v_mov_b32_e32 v6, 0
638; GFX11-NEXT:    v_dual_mov_b32 v8, 2.0 :: v_dual_mov_b32 v9, 0xb36211c7
639; GFX11-NEXT:    v_bfrev_b32_e32 v10, 4.0
640; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
641; GFX11-NEXT:    v_add_co_u32 v0, s4, s6, v0
642; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
643; GFX11-NEXT:    v_add_co_ci_u32_e64 v1, null, s7, 0, s4
644; GFX11-NEXT:    flat_load_b32 v11, v[0:1]
645; GFX11-NEXT:    v_mov_b32_e32 v0, 0x40c00000
646; GFX11-NEXT:    v_mov_b32_e32 v1, 0x40e00000
647; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
648; GFX11-NEXT:    image_bvh64_intersect_ray v[0:3], [v[9:10], v11, v[6:8], v[3:5], v[0:2]], s[0:3]
649; GFX11-NEXT:    s_waitcnt vmcnt(0)
650; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
651; GFX11-NEXT:    s_endpgm
652;
653; GFX12-SDAG-LABEL: image_bvh64_intersect_ray_nsa_reassign:
654; GFX12-SDAG:       ; %bb.0: ; %main_body
655; GFX12-SDAG-NEXT:    s_clause 0x1
656; GFX12-SDAG-NEXT:    s_load_b64 s[6:7], s[4:5], 0x24
657; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
658; GFX12-SDAG-NEXT:    v_dual_mov_b32 v7, 1.0 :: v_dual_and_b32 v0, 0x3ff, v0
659; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0x41000000
660; GFX12-SDAG-NEXT:    v_dual_mov_b32 v3, 0x40400000 :: v_dual_mov_b32 v4, 4.0
661; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3)
662; GFX12-SDAG-NEXT:    v_dual_mov_b32 v5, 0x40a00000 :: v_dual_lshlrev_b32 v0, 2, v0
663; GFX12-SDAG-NEXT:    v_mov_b32_e32 v6, 0
664; GFX12-SDAG-NEXT:    v_dual_mov_b32 v8, 2.0 :: v_dual_mov_b32 v9, 0xb36211c7
665; GFX12-SDAG-NEXT:    v_bfrev_b32_e32 v10, 4.0
666; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
667; GFX12-SDAG-NEXT:    v_add_co_u32 v0, s4, s6, v0
668; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
669; GFX12-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, s7, 0, s4
670; GFX12-SDAG-NEXT:    flat_load_b32 v11, v[0:1]
671; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0x40c00000
672; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, 0x40e00000
673; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
674; GFX12-SDAG-NEXT:    image_bvh64_intersect_ray v[0:3], [v[9:10], v11, v[6:8], v[3:5], v[0:2]], s[0:3]
675; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
676; GFX12-SDAG-NEXT:    flat_store_b128 v[0:1], v[0:3]
677; GFX12-SDAG-NEXT:    s_endpgm
678;
679; GFX12-GISEL-LABEL: image_bvh64_intersect_ray_nsa_reassign:
680; GFX12-GISEL:       ; %bb.0: ; %main_body
681; GFX12-GISEL-NEXT:    s_clause 0x1
682; GFX12-GISEL-NEXT:    s_load_b64 s[6:7], s[4:5], 0x24
683; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
684; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
685; GFX12-GISEL-NEXT:    s_mov_b32 s5, 1.0
686; GFX12-GISEL-NEXT:    s_mov_b32 s4, 0
687; GFX12-GISEL-NEXT:    v_mov_b32_e32 v9, 0xb36211c7
688; GFX12-GISEL-NEXT:    s_mov_b32 s8, 0x40400000
689; GFX12-GISEL-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
690; GFX12-GISEL-NEXT:    s_mov_b32 s12, 0x40c00000
691; GFX12-GISEL-NEXT:    s_mov_b32 s10, 0x40a00000
692; GFX12-GISEL-NEXT:    s_mov_b32 s9, 4.0
693; GFX12-GISEL-NEXT:    s_mov_b32 s14, 0x41000000
694; GFX12-GISEL-NEXT:    s_mov_b32 s13, 0x40e00000
695; GFX12-GISEL-NEXT:    v_mov_b32_e32 v6, s12
696; GFX12-GISEL-NEXT:    v_bfrev_b32_e32 v10, 4.0
697; GFX12-GISEL-NEXT:    v_dual_mov_b32 v8, s14 :: v_dual_mov_b32 v3, s8
698; GFX12-GISEL-NEXT:    v_dual_mov_b32 v4, s9 :: v_dual_mov_b32 v7, s13
699; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
700; GFX12-GISEL-NEXT:    v_dual_mov_b32 v5, s10 :: v_dual_mov_b32 v0, s6
701; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, s7
702; GFX12-GISEL-NEXT:    s_mov_b32 s6, 2.0
703; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
704; GFX12-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, v0, v2
705; GFX12-GISEL-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
706; GFX12-GISEL-NEXT:    flat_load_b32 v11, v[0:1]
707; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5
708; GFX12-GISEL-NEXT:    s_wait_alu 0xfffe
709; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, s6
710; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
711; GFX12-GISEL-NEXT:    image_bvh64_intersect_ray v[0:3], [v[9:10], v11, v[0:2], v[3:5], v[6:8]], s[0:3]
712; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
713; GFX12-GISEL-NEXT:    flat_store_b128 v[0:1], v[0:3]
714; GFX12-GISEL-NEXT:    s_endpgm
715main_body:
716  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
717  %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
718  %ray_extent = load float, ptr %gep_ray, align 4
719  %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
720  %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1
721  %ray_origin = insertelement <3 x float> %ray_origin1, float 2.0, i32 2
722  %ray_dir0 = insertelement <3 x float> undef, float 3.0, i32 0
723  %ray_dir1 = insertelement <3 x float> %ray_dir0, float 4.0, i32 1
724  %ray_dir = insertelement <3 x float> %ray_dir1, float 5.0, i32 2
725  %ray_inv_dir0 = insertelement <3 x float> undef, float 6.0, i32 0
726  %ray_inv_dir1 = insertelement <3 x float> %ray_inv_dir0, float 7.0, i32 1
727  %ray_inv_dir = insertelement <3 x float> %ray_inv_dir1, float 8.0, i32 2
728  %v = call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i64.v4f32(i64 1111111111111, float %ray_extent, <3 x float> %ray_origin, <3 x float> %ray_dir, <3 x float> %ray_inv_dir, <4 x i32> %tdescr)
729  store <4 x i32> %v, ptr undef
730  ret void
731}
732
733define amdgpu_kernel void @image_bvh64_intersect_ray_a16_nsa_reassign(ptr %p_ray, <4 x i32> inreg %tdescr) {
734; GFX1013-LABEL: image_bvh64_intersect_ray_a16_nsa_reassign:
735; GFX1013:       ; %bb.0: ; %main_body
736; GFX1013-NEXT:    s_clause 0x1
737; GFX1013-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
738; GFX1013-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x34
739; GFX1013-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
740; GFX1013-NEXT:    v_mov_b32_e32 v3, 0
741; GFX1013-NEXT:    v_mov_b32_e32 v4, 1.0
742; GFX1013-NEXT:    v_mov_b32_e32 v5, 2.0
743; GFX1013-NEXT:    v_mov_b32_e32 v6, 0x44004200
744; GFX1013-NEXT:    v_mov_b32_e32 v7, 0x46004500
745; GFX1013-NEXT:    v_mov_b32_e32 v8, 0x48004700
746; GFX1013-NEXT:    s_waitcnt lgkmcnt(0)
747; GFX1013-NEXT:    v_add_co_u32 v0, s4, s6, v0
748; GFX1013-NEXT:    v_add_co_ci_u32_e64 v1, s4, s7, 0, s4
749; GFX1013-NEXT:    flat_load_dword v2, v[0:1]
750; GFX1013-NEXT:    v_mov_b32_e32 v0, 0xb36211c6
751; GFX1013-NEXT:    v_bfrev_b32_e32 v1, 4.0
752; GFX1013-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
753; GFX1013-NEXT:    image_bvh64_intersect_ray v[0:3], v[0:8], s[0:3] a16
754; GFX1013-NEXT:    s_waitcnt vmcnt(0)
755; GFX1013-NEXT:    flat_store_dwordx4 v[0:1], v[0:3]
756; GFX1013-NEXT:    s_endpgm
757;
758; GFX1030-LABEL: image_bvh64_intersect_ray_a16_nsa_reassign:
759; GFX1030:       ; %bb.0: ; %main_body
760; GFX1030-NEXT:    s_clause 0x1
761; GFX1030-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x24
762; GFX1030-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x34
763; GFX1030-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
764; GFX1030-NEXT:    v_mov_b32_e32 v3, 0
765; GFX1030-NEXT:    v_mov_b32_e32 v5, 2.0
766; GFX1030-NEXT:    v_mov_b32_e32 v4, 1.0
767; GFX1030-NEXT:    v_mov_b32_e32 v6, 0x44004200
768; GFX1030-NEXT:    v_mov_b32_e32 v7, 0x46004500
769; GFX1030-NEXT:    v_mov_b32_e32 v8, 0x48004700
770; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
771; GFX1030-NEXT:    v_add_co_u32 v0, s4, s6, v0
772; GFX1030-NEXT:    v_add_co_ci_u32_e64 v1, null, s7, 0, s4
773; GFX1030-NEXT:    flat_load_dword v2, v[0:1]
774; GFX1030-NEXT:    v_bfrev_b32_e32 v1, 4.0
775; GFX1030-NEXT:    v_mov_b32_e32 v0, 0xb36211c6
776; GFX1030-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
777; GFX1030-NEXT:    image_bvh64_intersect_ray v[0:3], v[0:8], s[0:3] a16
778; GFX1030-NEXT:    s_waitcnt vmcnt(0)
779; GFX1030-NEXT:    flat_store_dwordx4 v[0:1], v[0:3]
780; GFX1030-NEXT:    s_endpgm
781;
782; GFX11-LABEL: image_bvh64_intersect_ray_a16_nsa_reassign:
783; GFX11:       ; %bb.0: ; %main_body
784; GFX11-NEXT:    s_clause 0x1
785; GFX11-NEXT:    s_load_b64 s[6:7], s[4:5], 0x24
786; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
787; GFX11-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
788; GFX11-NEXT:    v_mov_b32_e32 v2, 0x48004500
789; GFX11-NEXT:    v_mov_b32_e32 v4, 1.0
790; GFX11-NEXT:    v_mov_b32_e32 v6, 0xb36211c6
791; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_3) | instid1(VALU_DEP_3)
792; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
793; GFX11-NEXT:    v_bfrev_b32_e32 v7, 4.0
794; GFX11-NEXT:    v_mov_b32_e32 v5, 2.0
795; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
796; GFX11-NEXT:    v_add_co_u32 v0, s4, s6, v0
797; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
798; GFX11-NEXT:    v_add_co_ci_u32_e64 v1, null, s7, 0, s4
799; GFX11-NEXT:    flat_load_b32 v8, v[0:1]
800; GFX11-NEXT:    v_mov_b32_e32 v1, 0x47004400
801; GFX11-NEXT:    v_mov_b32_e32 v0, 0x46004200
802; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
803; GFX11-NEXT:    image_bvh64_intersect_ray v[0:3], [v[6:7], v8, v[3:5], v[0:2]], s[0:3] a16
804; GFX11-NEXT:    s_waitcnt vmcnt(0)
805; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
806; GFX11-NEXT:    s_endpgm
807;
808; GFX12-SDAG-LABEL: image_bvh64_intersect_ray_a16_nsa_reassign:
809; GFX12-SDAG:       ; %bb.0: ; %main_body
810; GFX12-SDAG-NEXT:    s_clause 0x1
811; GFX12-SDAG-NEXT:    s_load_b64 s[6:7], s[4:5], 0x24
812; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
813; GFX12-SDAG-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
814; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0x48004500
815; GFX12-SDAG-NEXT:    v_mov_b32_e32 v4, 1.0
816; GFX12-SDAG-NEXT:    v_mov_b32_e32 v6, 0xb36211c6
817; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_3) | instid1(VALU_DEP_3)
818; GFX12-SDAG-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
819; GFX12-SDAG-NEXT:    v_bfrev_b32_e32 v7, 4.0
820; GFX12-SDAG-NEXT:    v_mov_b32_e32 v5, 2.0
821; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
822; GFX12-SDAG-NEXT:    v_add_co_u32 v0, s4, s6, v0
823; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
824; GFX12-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, s7, 0, s4
825; GFX12-SDAG-NEXT:    flat_load_b32 v8, v[0:1]
826; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, 0x47004400
827; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0x46004200
828; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
829; GFX12-SDAG-NEXT:    image_bvh64_intersect_ray v[0:3], [v[6:7], v8, v[3:5], v[0:2]], s[0:3] a16
830; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
831; GFX12-SDAG-NEXT:    flat_store_b128 v[0:1], v[0:3]
832; GFX12-SDAG-NEXT:    s_endpgm
833;
834; GFX12-GISEL-LABEL: image_bvh64_intersect_ray_a16_nsa_reassign:
835; GFX12-GISEL:       ; %bb.0: ; %main_body
836; GFX12-GISEL-NEXT:    s_clause 0x1
837; GFX12-GISEL-NEXT:    s_load_b64 s[6:7], s[4:5], 0x24
838; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
839; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
840; GFX12-GISEL-NEXT:    s_mov_b32 s5, 1.0
841; GFX12-GISEL-NEXT:    s_mov_b32 s4, 0
842; GFX12-GISEL-NEXT:    s_mov_b32 s8, 0x42004600
843; GFX12-GISEL-NEXT:    s_mov_b32 s9, 0x44004700
844; GFX12-GISEL-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
845; GFX12-GISEL-NEXT:    s_mov_b32 s10, 0x45004800
846; GFX12-GISEL-NEXT:    v_mov_b32_e32 v6, 0xb36211c6
847; GFX12-GISEL-NEXT:    v_bfrev_b32_e32 v7, 4.0
848; GFX12-GISEL-NEXT:    v_mov_b32_e32 v3, s8
849; GFX12-GISEL-NEXT:    v_dual_mov_b32 v5, s10 :: v_dual_mov_b32 v4, s9
850; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
851; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s6 :: v_dual_mov_b32 v1, s7
852; GFX12-GISEL-NEXT:    s_mov_b32 s6, 2.0
853; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
854; GFX12-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, v0, v2
855; GFX12-GISEL-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
856; GFX12-GISEL-NEXT:    flat_load_b32 v8, v[0:1]
857; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5
858; GFX12-GISEL-NEXT:    s_wait_alu 0xfffe
859; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, s6
860; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
861; GFX12-GISEL-NEXT:    image_bvh64_intersect_ray v[0:3], [v[6:7], v8, v[0:2], v[3:5]], s[0:3] a16
862; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
863; GFX12-GISEL-NEXT:    flat_store_b128 v[0:1], v[0:3]
864; GFX12-GISEL-NEXT:    s_endpgm
865main_body:
866  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
867  %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
868  %ray_extent = load float, ptr %gep_ray, align 4
869  %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
870  %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1
871  %ray_origin = insertelement <3 x float> %ray_origin1, float 2.0, i32 2
872  %ray_dir0 = insertelement <3 x half> undef, half 3.0, i32 0
873  %ray_dir1 = insertelement <3 x half> %ray_dir0, half 4.0, i32 1
874  %ray_dir = insertelement <3 x half> %ray_dir1, half 5.0, i32 2
875  %ray_inv_dir0 = insertelement <3 x half> undef, half 6.0, i32 0
876  %ray_inv_dir1 = insertelement <3 x half> %ray_inv_dir0, half 7.0, i32 1
877  %ray_inv_dir = insertelement <3 x half> %ray_inv_dir1, half 8.0, i32 2
878  %v = call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i64.v4f16(i64 1111111111110, float %ray_extent, <3 x float> %ray_origin, <3 x half> %ray_dir, <3 x half> %ray_inv_dir, <4 x i32> %tdescr)
879  store <4 x i32> %v, ptr undef
880  ret void
881}
882
883declare i32 @llvm.amdgcn.workitem.id.x()
884