diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.48bit.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.48bit.ll | 61 |
1 files changed, 33 insertions, 28 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.48bit.ll b/llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.48bit.ll index 68312b89142c..fd190b23dd8c 100644 --- a/llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.48bit.ll +++ b/llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.48bit.ll @@ -138,46 +138,51 @@ define <3 x half> @bitcast_v3bf16_to_v3f16(<3 x bfloat> %a, i32 %b) { ; GFX11-TRUE16-LABEL: bitcast_v3bf16_to_v3f16: ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_mov_b32 s0, exec_lo -; GFX11-TRUE16-NEXT: v_cmpx_ne_u32_e32 0, v2 -; GFX11-TRUE16-NEXT: s_xor_b32 s0, exec_lo, s0 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v2 +; GFX11-TRUE16-NEXT: ; implicit-def: $vgpr2_vgpr3 +; GFX11-TRUE16-NEXT: s_and_saveexec_b32 s0, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX11-TRUE16-NEXT: s_xor_b32 s0, exec_lo, s0 +; GFX11-TRUE16-NEXT: ; %bb.1: ; %cmp.false +; GFX11-TRUE16-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 +; GFX11-TRUE16-NEXT: ; implicit-def: $vgpr1 +; GFX11-TRUE16-NEXT: ; %bb.2: ; %Flow ; GFX11-TRUE16-NEXT: s_and_not1_saveexec_b32 s0, s0 -; GFX11-TRUE16-NEXT: s_cbranch_execz .LBB0_2 -; GFX11-TRUE16-NEXT: ; %bb.1: ; %cmp.true +; GFX11-TRUE16-NEXT: s_cbranch_execz .LBB0_4 +; GFX11-TRUE16-NEXT: ; %bb.3: ; %cmp.true ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) ; GFX11-TRUE16-NEXT: v_add_f32_e32 v1, 0x40c00000, v2 ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.h, v0.l -; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 ; GFX11-TRUE16-NEXT: v_add_f32_e32 v2, 0x40c00000, v2 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, 0x40c00000, v0 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v1, 16, 1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_bfe_u32 v4, v0, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v5, 0x400000, v2 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v7, 0x400000, v2 ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v7, 0x400000, v0 ; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v4, v4, v0, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v5, vcc_lo -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v3, 0x400000, v1 -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v6, v1, 0x7fff -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.l, v2.h -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v0, v4, v7, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_dual_cndmask_b32 v7, v3, v7 :: v_dual_and_b32 v0, 0xffff0000, v0 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, 0x40c00000, v0 +; GFX11-TRUE16-NEXT: v_bfe_u32 v4, v1, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v5, 0x400000, v1 ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0x7fc0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_bfi_b32 v0, 0xffff, v2, v0 -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc_lo -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, v3.h -; GFX11-TRUE16-NEXT: .LBB0_2: ; %end +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0x7fc0 +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v0, 16, 1 +; GFX11-TRUE16-NEXT: v_add3_u32 v4, v4, v1, 0x7fff +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v0, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v1, v4, v5, vcc_lo +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v2, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.l, v7.h +; GFX11-TRUE16-NEXT: .LBB0_4: ; %end ; GFX11-TRUE16-NEXT: s_or_b32 exec_lo, exec_lo, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_dual_mov_b32 v0, v2 :: v_dual_mov_b32 v1, v3 ; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-FAKE16-LABEL: bitcast_v3bf16_to_v3f16: |
