# RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=GCN %s # GCN-LABEL: name: valu_write_vgpr_mfma_read # GCN: V_MOV_B32 # GCN: V_MOV_B32 # GCN-NEXT: S_NOP 1 # GCN-NEXT: V_MFMA name: valu_write_vgpr_mfma_read body: | bb.0: $vgpr0 = V_MOV_B32_e32 1, implicit $exec $vgpr1 = V_MOV_B32_e32 1, implicit $exec $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: valu_write_vgpr_accvgpr_write_read # GCN: V_MOV_B32 # GCN-NEXT: S_NOP 1 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: valu_write_vgpr_accvgpr_write_read body: | bb.0: $vgpr0 = V_MOV_B32_e32 1, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec ... --- # GCN-LABEL: name: mfma_write_agpr_mfma_read_same_agpr # GCN: V_MFMA # GCN-NEXT: V_MFMA name: mfma_write_agpr_mfma_read_same_agpr body: | bb.0: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: mfma_write_agpr_mfma_read_overlap # GCN: V_MFMA # GCN-NEXT: S_NOP 1 # GCN-NEXT: V_MFMA name: mfma_write_agpr_mfma_read_overlap body: | bb.0: $agpr1_agpr2_agpr3_agpr4 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: mfma_write_agpr_mfma_read_partial # GCN: V_MFMA # GCN-NEXT: S_NOP 1 # GCN-NEXT: V_MFMA name: mfma_write_agpr_mfma_read_partial body: | bb.0: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: mfma_write_agpr_mfma_srca_read_overlap # GCN: V_MFMA # GCN-NEXT: S_NOP 3 # GCN-NEXT: V_MFMA name: mfma_write_agpr_mfma_srca_read_overlap body: | bb.0: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $agpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: mfma_write_agpr_mfma_srcb_read_overlap # GCN: V_MFMA # GCN-NEXT: S_NOP 3 # GCN-NEXT: V_MFMA name: mfma_write_agpr_mfma_srcb_read_overlap body: | bb.0: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $agpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: mfma_4x4_write_agpr_accvgpr_read # GCN: V_MFMA_F32_4X4X1F32 # GCN-NEXT: S_NOP 3 # GCN-NEXT: V_ACCVGPR_READ_B32_e64 name: mfma_4x4_write_agpr_accvgpr_read body: | bb.0: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec ... --- # GCN-LABEL: name: mfma_16x16_write_agpr_accvgpr_read # GCN: V_MFMA_F32_16X16X1F32 # GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 1 # GCN-NEXT: V_ACCVGPR_READ_B32_e64 name: mfma_16x16_write_agpr_accvgpr_read body: | bb.0: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec ... --- # GCN-LABEL: name: mfma_32x32_write_agpr_accvgpr_read # GCN: V_MFMA_F32_32X32X2F32 # GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 1 # GCN-NEXT: V_ACCVGPR_READ_B32_e64 name: mfma_32x32_write_agpr_accvgpr_read body: | bb.0: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec ... --- # GCN-LABEL: name: mfma_4x4_write_agpr_accvgpr_write # GCN: V_MFMA_F32_4X4X1F32 # GCN-NEXT: S_NOP 0 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: mfma_4x4_write_agpr_accvgpr_write body: | bb.0: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec ... --- # GCN-LABEL: name: mfma_16x16_write_agpr_accvgpr_write # GCN: V_MFMA_F32_16X16X1F32 # GCN-NEXT: S_NOP 6 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: mfma_16x16_write_agpr_accvgpr_write body: | bb.0: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_16X16X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec ... --- # GCN-LABEL: name: mfma_32x32_write_agpr_accvgpr_write # GCN: V_MFMA_F32_32X32X2F32 # GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 6 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: mfma_32x32_write_agpr_accvgpr_write body: | bb.0: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X2F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec ... --- # GCN-LABEL: name: mfma_4x4_read_srcc_accvgpr_write # GCN: V_MFMA_F32_4X4X1F32 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: mfma_4x4_read_srcc_accvgpr_write body: | bb.0: $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr2, implicit $exec ... --- # GCN-LABEL: name: mfma_16x16_read_srcc_accvgpr_write # GCN: V_MFMA_F32_16X16X1F32 # GCN-NEXT: S_NOP 4 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: mfma_16x16_read_srcc_accvgpr_write body: | bb.0: $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 = V_MFMA_F32_16X16X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr2, implicit $exec ... --- # GCN-LABEL: name: mfma_32x32_read_srcc_accvgpr_write # GCN: V_MFMA_F32_32X32X2F32 # GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 4 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: mfma_32x32_read_srcc_accvgpr_write body: | bb.0: $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 = V_MFMA_F32_32X32X2F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr2, implicit $exec ... --- # GCN-LABEL: name: accvgpr_read_write_vgpr_valu_read # GCN: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: V_ADD_F32 name: accvgpr_read_write_vgpr_valu_read body: | bb.0: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec $vgpr1 = V_ADD_F32_e32 0, killed $vgpr0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: accvgpr_read_write_vgpr_mfma_read # GCN: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 1 # GCN-NEXT: V_MFMA name: accvgpr_read_write_vgpr_mfma_read body: | bb.0: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr4, implicit $exec $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr0, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: accvgpr_read_write_vgpr_accvgpr_write_read # GCN: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 1 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: accvgpr_read_write_vgpr_accvgpr_write_read body: | bb.0: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr1, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec ... --- # GCN-LABEL: name: accvgpr_write_agpr_mfma_read_srcc # GCN: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: S_NOP 0 # GCN-NEXT: V_MFMA name: accvgpr_write_agpr_mfma_read_srcc body: | bb.0: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr2, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: accvgpr_write_agpr_mfma_read_srca # GCN: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: S_NOP 2 # GCN-NEXT: V_MFMA name: accvgpr_write_agpr_mfma_read_srca body: | bb.0: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 killed $agpr8, killed $vgpr1, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: accvgpr_write_agpr_mfma_read_srcb # GCN: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: S_NOP 2 # GCN-NEXT: V_MFMA name: accvgpr_write_agpr_mfma_read_srcb body: | bb.0: $agpr8 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $agpr8, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: accvgpr_write_agpr_accvgpr_read # GCN: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: S_NOP 2 # GCN-NEXT: V_ACCVGPR_READ_B32_e64 name: accvgpr_write_agpr_accvgpr_read body: | bb.0: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec ... --- # GCN-LABEL: name: vcmpx_write_exec_mfma # GCN: V_CMPX_EQ_I32_e32 # GCN-NEXT: S_NOP 3 # GCN-NEXT: V_MFMA name: vcmpx_write_exec_mfma body: | bb.0: implicit $exec, implicit $vcc = V_CMPX_EQ_I32_e32 $vgpr0, $vgpr1, implicit $exec $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 killed $agpr8, killed $vgpr1, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec ... --- # GCN-LABEL: name: vcmpx_write_exec_accvgpr_write # GCN: V_CMPX_EQ_I32_e32 # GCN-NEXT: S_NOP 3 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 name: vcmpx_write_exec_accvgpr_write body: | bb.0: implicit $exec, implicit $vcc = V_CMPX_EQ_I32_e32 $vgpr0, $vgpr1, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr2, implicit $exec ... --- # GCN-LABEL: name: accvgpr_read_write_vgpr_load # GCN: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 1 # GCN-NEXT: FLAT_LOAD_DWORD name: accvgpr_read_write_vgpr_load body: | bb.0: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec $vgpr3 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ... --- # GCN-LABEL: name: accvgpr_read_write_vgpr_ds_permute # GCN: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 1 # GCN-NEXT: DS_PERMUTE_B32 name: accvgpr_read_write_vgpr_ds_permute body: | bb.0: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec $vgpr1 = DS_PERMUTE_B32 $vgpr0, $vgpr1, 0, implicit $exec ... --- # GCN-LABEL: name: accvgpr_read_write_vgpr_flat_load # GCN: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 1 # GCN-NEXT: FLAT_LOAD_DWORD name: accvgpr_read_write_vgpr_flat_load body: | bb.0: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec $vgpr4 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ... --- # GCN-LABEL: name: accvgpr_read_write_vgpr_buffer_store # GCN: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 1 # GCN-NEXT: BUFFER_STORE_DWORD_OFFSET name: accvgpr_read_write_vgpr_buffer_store body: | bb.0: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec BUFFER_STORE_DWORD_OFFSET $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, 0, 0, implicit $exec ... --- # GCN-LABEL: name: accvgpr_read_write_vgpr_store # GCN: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 1 # GCN-NEXT: DS_WRITE_B32 name: accvgpr_read_write_vgpr_store body: | bb.0: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec DS_WRITE_B32 $vgpr0, $vgpr1, 0, 0, implicit $m0, implicit $exec ... --- # GCN-LABEL: name: valu_write_vgpr_accvgpr_read_load_no_dependency # GCN: V_MOV_B32 # GCN-NEXT: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: FLAT_LOAD_DWORD name: valu_write_vgpr_accvgpr_read_load_no_dependency body: | bb.0: $vgpr0 = V_MOV_B32_e32 1, implicit $exec $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec $vgpr4 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ... --- # GCN-LABEL: name: valu_write_vgpr_accvgpr_read_load_1_and_3_depend # GCN: V_MOV_B32 # GCN-NEXT: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 0 # GCN-NEXT: FLAT_LOAD_DWORD name: valu_write_vgpr_accvgpr_read_load_1_and_3_depend body: | bb.0: $vgpr0 = V_MOV_B32_e32 1, implicit $exec $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec $vgpr4 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ... --- # GCN-LABEL: name: valu_write_vgpr_accvgpr_write_load_1_and_3_depend # GCN: V_MOV_B32 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: S_NOP 0 # GCN-NEXT: FLAT_LOAD_DWORD name: valu_write_vgpr_accvgpr_write_load_1_and_3_depend body: | bb.0: $vgpr0 = V_MOV_B32_e32 1, implicit $exec $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr2, implicit $exec $vgpr4 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ... --- # GCN-LABEL: name: valu_write_vgpr_accvgpr_read_load_2_and_3_depend # GCN: V_MOV_B32 # GCN-NEXT: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: S_NOP 1 # GCN-NEXT: FLAT_LOAD_DWORD name: valu_write_vgpr_accvgpr_read_load_2_and_3_depend body: | bb.0: $vgpr0 = V_MOV_B32_e32 1, implicit $exec $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr0, implicit $exec $vgpr4 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ... ---