diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/nor-divergent-lanemask.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/nor-divergent-lanemask.ll | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/nor-divergent-lanemask.ll b/llvm/test/CodeGen/AMDGPU/nor-divergent-lanemask.ll index 4205393..a9c9db6 100644 --- a/llvm/test/CodeGen/AMDGPU/nor-divergent-lanemask.ll +++ b/llvm/test/CodeGen/AMDGPU/nor-divergent-lanemask.ll @@ -13,6 +13,7 @@ define amdgpu_ps i64 @test_nor(i64 inreg %a, i64 inreg %b) { ; SDAG-W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; SDAG-W64-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] ; SDAG-W64-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v0 +; SDAG-W64-NEXT: s_waitcnt_depctr depctr_va_sdst(0) ; SDAG-W64-NEXT: ; return to shader part epilog ; ; GISEL-W64-LABEL: test_nor: @@ -52,13 +53,13 @@ define amdgpu_ps i64 @test_or_two_uses(i64 inreg %a, i64 inreg %b) { ; SDAG-W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2) ; SDAG-W64-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] ; SDAG-W64-NEXT: s_xor_b64 s[0:1], s[0:1], -1 -; SDAG-W64-NEXT: s_waitcnt_depctr 0xfffe +; SDAG-W64-NEXT: s_waitcnt_depctr depctr_sa_sdst(0) ; SDAG-W64-NEXT: v_cndmask_b32_e64 v1, 0, 1, s[0:1] ; SDAG-W64-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 ; SDAG-W64-NEXT: s_delay_alu instid0(VALU_DEP_2) ; SDAG-W64-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v1 +; SDAG-W64-NEXT: s_waitcnt_depctr depctr_va_sdst(0) ; SDAG-W64-NEXT: s_and_b64 s[0:1], s[0:1], vcc -; SDAG-W64-NEXT: s_waitcnt_depctr 0xfffe ; SDAG-W64-NEXT: ; return to shader part epilog ; ; GISEL-W64-LABEL: test_or_two_uses: @@ -104,4 +105,4 @@ define amdgpu_ps i64 @test_or_two_uses(i64 inreg %a, i64 inreg %b) { %r1 = call i64 @llvm.amdgcn.ballot.i64(i1 %or) %r = and i64 %r0, %r1 ret i64 %r -}
\ No newline at end of file +} |
