diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/llvm.round.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/llvm.round.ll | 54 |
1 files changed, 28 insertions, 26 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.round.ll b/llvm/test/CodeGen/AMDGPU/llvm.round.ll index 42671f9dd674..b241b9b800d2 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.round.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.round.ll @@ -830,21 +830,22 @@ define amdgpu_kernel void @round_f16(ptr addrspace(1) %out, i32 %x.arg) #0 { ; ; GFX11-TRUE16-LABEL: round_f16: ; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_clause 0x1 ; GFX11-TRUE16-NEXT: s_load_b32 s2, s[4:5], 0x2c -; GFX11-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.l, s2 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, s2 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) ; GFX11-TRUE16-NEXT: v_sub_f16_e32 v0.h, s2, v0.l -; GFX11-TRUE16-NEXT: v_cmp_ge_f16_e64 s0, |v0.h|, 0.5 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.l, 0, 0x3c00, s0 -; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24 -; GFX11-TRUE16-NEXT: v_bfi_b32 v1, 0x7fff, v1, s2 ; GFX11-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX11-TRUE16-NEXT: v_cmp_ge_f16_e64 s3, |v0.h|, 0.5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v2.l, 0, 0x3c00, s3 +; GFX11-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX11-TRUE16-NEXT: v_bfi_b32 v1, 0x7fff, v2, v1 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l -; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-TRUE16-NEXT: s_endpgm ; @@ -991,32 +992,33 @@ define amdgpu_kernel void @round_v2f16(ptr addrspace(1) %out, i32 %in.arg) #0 { ; ; GFX11-TRUE16-LABEL: round_v2f16: ; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_clause 0x1 ; GFX11-TRUE16-NEXT: s_load_b32 s2, s[4:5], 0x2c +; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: s_lshr_b32 s3, s2, 16 -; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.h, s2 -; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.l, s3 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_sub_f16_e32 v1.h, s2, v0.h -; GFX11-TRUE16-NEXT: v_sub_f16_e32 v1.l, s3, v0.l -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cmp_ge_f16_e64 s1, |v1.h|, 0.5 -; GFX11-TRUE16-NEXT: v_cmp_ge_f16_e64 s0, |v1.l|, 0.5 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cndmask_b16 v2.l, 0, 0x3c00, s1 -; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.l, 0, 0x3c00, s0 -; GFX11-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x24 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_bfi_b32 v2, 0x7fff, v2, s2 -; GFX11-TRUE16-NEXT: v_bfi_b32 v1, 0x7fff, v1, s3 +; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.l, s2 +; GFX11-TRUE16-NEXT: v_trunc_f16_e32 v0.h, s3 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.l, s3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_sub_f16_e32 v1.l, s2, v0.l +; GFX11-TRUE16-NEXT: v_sub_f16_e32 v1.h, s3, v0.h ; GFX11-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cmp_ge_f16_e64 s6, |v1.l|, 0.5 +; GFX11-TRUE16-NEXT: v_cmp_ge_f16_e64 s7, |v1.h|, 0.5 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, s2 ; GFX11-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v2.l, 0, 0x3c00, s6 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v3.l, 0, 0x3c00, s7 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_bfi_b32 v1, 0x7fff, v2, v1 +; GFX11-TRUE16-NEXT: v_bfi_b32 v2, 0x7fff, v3, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.h, v0.h, v2.l ; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l +; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.h, v0.h, v2.l ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h ; GFX11-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-TRUE16-NEXT: s_endpgm ; |