aboutsummaryrefslogtreecommitdiff
path: root/llvm/test/CodeGen/AMDGPU/calling-conventions.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/calling-conventions.ll')
-rw-r--r--llvm/test/CodeGen/AMDGPU/calling-conventions.ll202
1 files changed, 101 insertions, 101 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/calling-conventions.ll b/llvm/test/CodeGen/AMDGPU/calling-conventions.ll
index ddd3b152..363a248 100644
--- a/llvm/test/CodeGen/AMDGPU/calling-conventions.ll
+++ b/llvm/test/CodeGen/AMDGPU/calling-conventions.ll
@@ -2700,142 +2700,142 @@ define amdgpu_kernel void @amd_kernel_v32i8(<32 x i8> %arg0) {
;
; GFX1250-LABEL: amd_kernel_v32i8:
; GFX1250: ; %bb.0: ; %entry
-; GFX1250-NEXT: s_load_b256 s[0:7], s[4:5], 0x24
+; GFX1250-NEXT: s_load_b256 s[8:15], s[4:5], 0x24
; GFX1250-NEXT: v_mov_b64_e32 v[8:9], 16
; GFX1250-NEXT: v_mov_b64_e32 v[10:11], 0
; GFX1250-NEXT: s_wait_kmcnt 0x0
-; GFX1250-NEXT: s_lshr_b32 s16, s0, 16
-; GFX1250-NEXT: s_lshr_b32 s17, s0, 24
-; GFX1250-NEXT: s_lshr_b32 s20, s2, 16
-; GFX1250-NEXT: s_lshr_b32 s21, s2, 24
-; GFX1250-NEXT: s_lshr_b32 s14, s7, 16
-; GFX1250-NEXT: s_lshr_b32 s15, s7, 24
-; GFX1250-NEXT: s_bfe_u32 s27, s7, 0x80008
+; GFX1250-NEXT: s_lshr_b32 s16, s8, 16
+; GFX1250-NEXT: s_lshr_b32 s17, s8, 24
+; GFX1250-NEXT: s_lshr_b32 s6, s15, 16
+; GFX1250-NEXT: s_lshr_b32 s7, s15, 24
+; GFX1250-NEXT: s_bfe_u32 s27, s15, 0x80008
; GFX1250-NEXT: s_add_co_i32 s17, s17, s17
; GFX1250-NEXT: s_add_co_i32 s16, s16, s16
-; GFX1250-NEXT: s_lshr_b32 s18, s1, 16
-; GFX1250-NEXT: s_lshr_b32 s19, s1, 24
-; GFX1250-NEXT: s_lshr_b32 s22, s3, 16
-; GFX1250-NEXT: s_lshr_b32 s23, s3, 24
-; GFX1250-NEXT: s_bfe_u32 s29, s1, 0x80008
-; GFX1250-NEXT: s_bfe_u32 s30, s3, 0x80008
-; GFX1250-NEXT: s_add_co_i32 s21, s21, s21
-; GFX1250-NEXT: s_add_co_i32 s20, s20, s20
; GFX1250-NEXT: s_lshl_b32 s17, s17, 8
; GFX1250-NEXT: s_and_b32 s16, s16, 0xff
-; GFX1250-NEXT: s_add_co_i32 s7, s7, s7
-; GFX1250-NEXT: s_add_co_i32 s27, s27, s27
; GFX1250-NEXT: s_add_co_i32 s15, s15, s15
-; GFX1250-NEXT: s_add_co_i32 s14, s14, s14
-; GFX1250-NEXT: s_add_co_i32 s3, s3, s3
+; GFX1250-NEXT: s_add_co_i32 s27, s27, s27
+; GFX1250-NEXT: s_add_co_i32 s7, s7, s7
+; GFX1250-NEXT: s_add_co_i32 s6, s6, s6
+; GFX1250-NEXT: s_or_b32 s16, s16, s17
+; GFX1250-NEXT: s_and_b32 s15, s15, 0xff
+; GFX1250-NEXT: s_lshl_b32 s17, s27, 8
+; GFX1250-NEXT: s_lshl_b32 s7, s7, 8
+; GFX1250-NEXT: s_and_b32 s6, s6, 0xff
+; GFX1250-NEXT: s_or_b32 s15, s15, s17
+; GFX1250-NEXT: s_or_b32 s6, s6, s7
+; GFX1250-NEXT: s_bfe_u32 s26, s14, 0x80008
+; GFX1250-NEXT: s_and_b32 s7, s15, 0xffff
+; GFX1250-NEXT: s_lshl_b32 s6, s6, 16
+; GFX1250-NEXT: s_lshr_b32 s20, s10, 16
+; GFX1250-NEXT: s_lshr_b32 s21, s10, 24
+; GFX1250-NEXT: s_lshr_b32 s4, s14, 16
+; GFX1250-NEXT: s_lshr_b32 s5, s14, 24
+; GFX1250-NEXT: s_or_b32 s6, s7, s6
+; GFX1250-NEXT: s_add_co_i32 s7, s14, s14
+; GFX1250-NEXT: s_add_co_i32 s26, s26, s26
+; GFX1250-NEXT: s_lshr_b32 s18, s9, 16
+; GFX1250-NEXT: s_lshr_b32 s19, s9, 24
+; GFX1250-NEXT: s_lshr_b32 s22, s11, 16
+; GFX1250-NEXT: s_lshr_b32 s23, s11, 24
+; GFX1250-NEXT: s_bfe_u32 s29, s9, 0x80008
+; GFX1250-NEXT: s_bfe_u32 s30, s11, 0x80008
+; GFX1250-NEXT: s_add_co_i32 s21, s21, s21
+; GFX1250-NEXT: s_add_co_i32 s20, s20, s20
+; GFX1250-NEXT: s_lshr_b32 s2, s13, 16
+; GFX1250-NEXT: s_lshr_b32 s3, s13, 24
+; GFX1250-NEXT: s_and_b32 s7, s7, 0xff
+; GFX1250-NEXT: s_lshl_b32 s14, s26, 8
+; GFX1250-NEXT: s_add_co_i32 s5, s5, s5
+; GFX1250-NEXT: s_add_co_i32 s4, s4, s4
+; GFX1250-NEXT: s_add_co_i32 s11, s11, s11
; GFX1250-NEXT: s_add_co_i32 s30, s30, s30
; GFX1250-NEXT: s_add_co_i32 s23, s23, s23
; GFX1250-NEXT: s_add_co_i32 s22, s22, s22
; GFX1250-NEXT: s_lshl_b32 s21, s21, 8
; GFX1250-NEXT: s_and_b32 s20, s20, 0xff
-; GFX1250-NEXT: s_add_co_i32 s1, s1, s1
+; GFX1250-NEXT: s_add_co_i32 s9, s9, s9
; GFX1250-NEXT: s_add_co_i32 s29, s29, s29
; GFX1250-NEXT: s_add_co_i32 s19, s19, s19
; GFX1250-NEXT: s_add_co_i32 s18, s18, s18
-; GFX1250-NEXT: s_lshr_b32 s10, s5, 16
-; GFX1250-NEXT: s_lshr_b32 s11, s5, 24
-; GFX1250-NEXT: s_lshr_b32 s12, s6, 16
-; GFX1250-NEXT: s_lshr_b32 s13, s6, 24
-; GFX1250-NEXT: s_or_b32 s16, s16, s17
-; GFX1250-NEXT: s_and_b32 s7, s7, 0xff
-; GFX1250-NEXT: s_lshl_b32 s17, s27, 8
-; GFX1250-NEXT: s_lshl_b32 s15, s15, 8
-; GFX1250-NEXT: s_and_b32 s14, s14, 0xff
-; GFX1250-NEXT: s_and_b32 s3, s3, 0xff
+; GFX1250-NEXT: s_bfe_u32 s25, s13, 0x80008
+; GFX1250-NEXT: s_lshl_b32 s5, s5, 8
+; GFX1250-NEXT: s_and_b32 s4, s4, 0xff
+; GFX1250-NEXT: s_or_b32 s7, s7, s14
+; GFX1250-NEXT: s_add_co_i32 s3, s3, s3
+; GFX1250-NEXT: s_add_co_i32 s2, s2, s2
+; GFX1250-NEXT: s_and_b32 s11, s11, 0xff
; GFX1250-NEXT: s_lshl_b32 s30, s30, 8
; GFX1250-NEXT: s_lshl_b32 s23, s23, 8
; GFX1250-NEXT: s_and_b32 s22, s22, 0xff
; GFX1250-NEXT: s_or_b32 s20, s20, s21
-; GFX1250-NEXT: s_and_b32 s1, s1, 0xff
+; GFX1250-NEXT: s_and_b32 s9, s9, 0xff
; GFX1250-NEXT: s_lshl_b32 s21, s29, 8
; GFX1250-NEXT: s_lshl_b32 s19, s19, 8
; GFX1250-NEXT: s_and_b32 s18, s18, 0xff
-; GFX1250-NEXT: s_lshr_b32 s8, s4, 16
-; GFX1250-NEXT: s_lshr_b32 s9, s4, 24
-; GFX1250-NEXT: s_bfe_u32 s24, s4, 0x80008
-; GFX1250-NEXT: s_bfe_u32 s25, s5, 0x80008
-; GFX1250-NEXT: s_bfe_u32 s26, s6, 0x80008
-; GFX1250-NEXT: s_or_b32 s7, s7, s17
-; GFX1250-NEXT: s_or_b32 s14, s14, s15
-; GFX1250-NEXT: s_add_co_i32 s13, s13, s13
-; GFX1250-NEXT: s_add_co_i32 s12, s12, s12
-; GFX1250-NEXT: s_add_co_i32 s11, s11, s11
-; GFX1250-NEXT: s_add_co_i32 s10, s10, s10
-; GFX1250-NEXT: s_bfe_u32 s28, s0, 0x80008
-; GFX1250-NEXT: s_or_b32 s3, s3, s30
+; GFX1250-NEXT: s_lshr_b32 s0, s12, 16
+; GFX1250-NEXT: s_lshr_b32 s1, s12, 24
+; GFX1250-NEXT: s_bfe_u32 s24, s12, 0x80008
+; GFX1250-NEXT: s_or_b32 s4, s4, s5
+; GFX1250-NEXT: s_and_b32 s5, s7, 0xffff
+; GFX1250-NEXT: s_add_co_i32 s7, s13, s13
+; GFX1250-NEXT: s_add_co_i32 s25, s25, s25
+; GFX1250-NEXT: s_lshl_b32 s3, s3, 8
+; GFX1250-NEXT: s_and_b32 s2, s2, 0xff
+; GFX1250-NEXT: s_bfe_u32 s28, s8, 0x80008
+; GFX1250-NEXT: s_or_b32 s11, s11, s30
; GFX1250-NEXT: s_or_b32 s22, s22, s23
-; GFX1250-NEXT: s_bfe_u32 s23, s2, 0x80008
-; GFX1250-NEXT: s_or_b32 s1, s1, s21
+; GFX1250-NEXT: s_bfe_u32 s23, s10, 0x80008
+; GFX1250-NEXT: s_or_b32 s9, s9, s21
; GFX1250-NEXT: s_or_b32 s18, s18, s19
-; GFX1250-NEXT: s_and_b32 s7, s7, 0xffff
-; GFX1250-NEXT: s_lshl_b32 s14, s14, 16
-; GFX1250-NEXT: s_add_co_i32 s6, s6, s6
-; GFX1250-NEXT: s_add_co_i32 s26, s26, s26
-; GFX1250-NEXT: s_lshl_b32 s13, s13, 8
-; GFX1250-NEXT: s_and_b32 s12, s12, 0xff
-; GFX1250-NEXT: s_add_co_i32 s5, s5, s5
-; GFX1250-NEXT: s_add_co_i32 s25, s25, s25
-; GFX1250-NEXT: s_lshl_b32 s11, s11, 8
-; GFX1250-NEXT: s_and_b32 s10, s10, 0xff
-; GFX1250-NEXT: s_add_co_i32 s4, s4, s4
+; GFX1250-NEXT: s_lshl_b32 s4, s4, 16
+; GFX1250-NEXT: s_and_b32 s7, s7, 0xff
+; GFX1250-NEXT: s_lshl_b32 s13, s25, 8
+; GFX1250-NEXT: s_or_b32 s2, s2, s3
+; GFX1250-NEXT: s_add_co_i32 s3, s12, s12
; GFX1250-NEXT: s_add_co_i32 s24, s24, s24
-; GFX1250-NEXT: s_add_co_i32 s9, s9, s9
-; GFX1250-NEXT: s_add_co_i32 s8, s8, s8
-; GFX1250-NEXT: s_and_b32 s3, s3, 0xffff
+; GFX1250-NEXT: s_add_co_i32 s1, s1, s1
+; GFX1250-NEXT: s_add_co_i32 s0, s0, s0
+; GFX1250-NEXT: s_and_b32 s11, s11, 0xffff
; GFX1250-NEXT: s_lshl_b32 s22, s22, 16
-; GFX1250-NEXT: s_add_co_i32 s2, s2, s2
+; GFX1250-NEXT: s_add_co_i32 s10, s10, s10
; GFX1250-NEXT: s_add_co_i32 s23, s23, s23
-; GFX1250-NEXT: s_and_b32 s1, s1, 0xffff
+; GFX1250-NEXT: s_and_b32 s9, s9, 0xffff
; GFX1250-NEXT: s_lshl_b32 s18, s18, 16
-; GFX1250-NEXT: s_add_co_i32 s0, s0, s0
+; GFX1250-NEXT: s_add_co_i32 s8, s8, s8
; GFX1250-NEXT: s_add_co_i32 s28, s28, s28
-; GFX1250-NEXT: s_or_b32 s7, s7, s14
-; GFX1250-NEXT: s_and_b32 s6, s6, 0xff
-; GFX1250-NEXT: s_lshl_b32 s14, s26, 8
-; GFX1250-NEXT: s_or_b32 s12, s12, s13
-; GFX1250-NEXT: s_and_b32 s5, s5, 0xff
-; GFX1250-NEXT: s_lshl_b32 s13, s25, 8
-; GFX1250-NEXT: s_or_b32 s10, s10, s11
-; GFX1250-NEXT: s_and_b32 s4, s4, 0xff
-; GFX1250-NEXT: s_lshl_b32 s11, s24, 8
-; GFX1250-NEXT: s_lshl_b32 s9, s9, 8
-; GFX1250-NEXT: s_and_b32 s8, s8, 0xff
-; GFX1250-NEXT: s_or_b32 s3, s3, s22
-; GFX1250-NEXT: s_and_b32 s2, s2, 0xff
-; GFX1250-NEXT: s_lshl_b32 s22, s23, 8
-; GFX1250-NEXT: s_or_b32 s1, s1, s18
+; GFX1250-NEXT: s_or_b32 s4, s5, s4
+; GFX1250-NEXT: s_or_b32 s5, s7, s13
+; GFX1250-NEXT: s_and_b32 s3, s3, 0xff
+; GFX1250-NEXT: s_lshl_b32 s7, s24, 8
+; GFX1250-NEXT: s_lshl_b32 s1, s1, 8
; GFX1250-NEXT: s_and_b32 s0, s0, 0xff
+; GFX1250-NEXT: s_or_b32 s11, s11, s22
+; GFX1250-NEXT: s_and_b32 s10, s10, 0xff
+; GFX1250-NEXT: s_lshl_b32 s22, s23, 8
+; GFX1250-NEXT: s_or_b32 s9, s9, s18
+; GFX1250-NEXT: s_and_b32 s8, s8, 0xff
; GFX1250-NEXT: s_lshl_b32 s18, s28, 8
-; GFX1250-NEXT: s_or_b32 s6, s6, s14
-; GFX1250-NEXT: s_or_b32 s5, s5, s13
-; GFX1250-NEXT: s_or_b32 s4, s4, s11
-; GFX1250-NEXT: s_or_b32 s8, s8, s9
-; GFX1250-NEXT: s_or_b32 s2, s2, s22
-; GFX1250-NEXT: s_or_b32 s0, s0, s18
-; GFX1250-NEXT: s_and_b32 s6, s6, 0xffff
-; GFX1250-NEXT: s_lshl_b32 s12, s12, 16
+; GFX1250-NEXT: s_or_b32 s3, s3, s7
+; GFX1250-NEXT: s_or_b32 s0, s0, s1
+; GFX1250-NEXT: s_or_b32 s10, s10, s22
+; GFX1250-NEXT: s_or_b32 s8, s8, s18
; GFX1250-NEXT: s_and_b32 s5, s5, 0xffff
-; GFX1250-NEXT: s_and_b32 s4, s4, 0xffff
-; GFX1250-NEXT: s_lshl_b32 s8, s8, 16
-; GFX1250-NEXT: s_lshl_b32 s9, s10, 16
-; GFX1250-NEXT: s_and_b32 s2, s2, 0xffff
+; GFX1250-NEXT: s_and_b32 s1, s3, 0xffff
+; GFX1250-NEXT: s_lshl_b32 s0, s0, 16
+; GFX1250-NEXT: s_lshl_b32 s2, s2, 16
+; GFX1250-NEXT: s_and_b32 s10, s10, 0xffff
; GFX1250-NEXT: s_lshl_b32 s20, s20, 16
-; GFX1250-NEXT: s_and_b32 s0, s0, 0xffff
+; GFX1250-NEXT: s_and_b32 s8, s8, 0xffff
; GFX1250-NEXT: s_lshl_b32 s16, s16, 16
-; GFX1250-NEXT: s_or_b32 s6, s6, s12
-; GFX1250-NEXT: s_or_b32 s4, s4, s8
-; GFX1250-NEXT: s_or_b32 s5, s5, s9
-; GFX1250-NEXT: s_or_b32 s2, s2, s20
-; GFX1250-NEXT: s_or_b32 s0, s0, s16
-; GFX1250-NEXT: v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5
-; GFX1250-NEXT: v_dual_mov_b32 v2, s6 :: v_dual_mov_b32 v3, s7
-; GFX1250-NEXT: v_dual_mov_b32 v4, s0 :: v_dual_mov_b32 v5, s1
-; GFX1250-NEXT: v_dual_mov_b32 v6, s2 :: v_dual_mov_b32 v7, s3
+; GFX1250-NEXT: s_or_b32 s0, s1, s0
+; GFX1250-NEXT: s_or_b32 s1, s5, s2
+; GFX1250-NEXT: s_or_b32 s10, s10, s20
+; GFX1250-NEXT: s_or_b32 s8, s8, s16
+; GFX1250-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX1250-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s6
+; GFX1250-NEXT: v_dual_mov_b32 v4, s8 :: v_dual_mov_b32 v5, s9
+; GFX1250-NEXT: v_dual_mov_b32 v6, s10 :: v_dual_mov_b32 v7, s11
; GFX1250-NEXT: s_clause 0x1
; GFX1250-NEXT: global_store_b128 v[8:9], v[0:3], off
; GFX1250-NEXT: global_store_b128 v[10:11], v[4:7], off