--- vgpr_good_asm 2015-10-11 23:42:07.606265715 +0200 +++ vgpr_bad_asm 2015-10-11 23:40:39.661750776 +0200 @@ -1,282 +1,280 @@  .text  .section .AMDGPU.config  .long 45096 -.long 645 +.long 648  .long 45100  .long 0  .long 165580  .long 2  .hsa_code_object_version 1,0  .hsa_code_object_isa 7,0,0,"AMD","AMDGPU"  .text  .globl main  .align 256  main: ; @main  ; BB#0: ; %main_body  s_wqm_b64 exec, exec  s_load_dwordx4 s[16:19], s[2:3], 0x0  s_load_dwordx4 s[12:15], s[2:3], 0x4  s_mov_b32 m0, s10  v_interp_p1_f32 v12, v0, 0, 0, [m0]  v_interp_p2_f32 v12, [v12], v1, 0, 0, [m0]  v_interp_p1_f32 v14, v0, 1, 0, [m0]  v_interp_p2_f32 v14, [v14], v1, 1, 0, [m0]  v_interp_p1_f32 v15, v0, 2, 0, [m0]  v_interp_p2_f32 v15, [v15], v1, 2, 0, [m0]  s_load_dwordx4 s[28:31], s[2:3], 0x8  s_waitcnt lgkmcnt(0) - s_buffer_load_dword s3, s[16:19], 0x0 - s_buffer_load_dword s2, s[16:19], 0x4 - s_buffer_load_dword s1, s[16:19], 0x8 + s_buffer_load_dword s10, s[16:19], 0x0 + s_buffer_load_dword s9, s[16:19], 0x4 + s_buffer_load_dword s8, s[16:19], 0x8  s_buffer_load_dword s18, s[12:15], 0xb8  s_buffer_load_dword s19, s[12:15], 0xb9  s_buffer_load_dword s20, s[12:15], 0xba - s_buffer_load_dword s8, s[12:15], 0xbb + s_buffer_load_dword s1, s[12:15], 0xbb  s_buffer_load_dword s16, s[12:15], 0xbc  s_buffer_load_dword s11, s[12:15], 0xbd  s_buffer_load_dword s17, s[12:15], 0xbe  s_buffer_load_dword s25, s[12:15], 0xc8  s_buffer_load_dword s26, s[12:15], 0xc9  s_buffer_load_dword s27, s[12:15], 0xca  s_buffer_load_dword s0, s[12:15], 0xd7  s_buffer_load_dword s22, s[12:15], 0x110  s_buffer_load_dword s21, s[12:15], 0x111  s_waitcnt lgkmcnt(0) - v_mov_b32_e32 v6, s8 + v_mov_b32_e32 v6, s1  s_buffer_load_dword s23, s[12:15], 0x112  s_buffer_load_dword s24, s[12:15], 0x114 - s_buffer_load_dword s8, s[28:31], 0x0 - s_buffer_load_dword s9, s[28:31], 0x1 - s_buffer_load_dword s10, s[28:31], 0x2 + s_buffer_load_dword s3, s[28:31], 0x0 + s_buffer_load_dword s2, s[28:31], 0x1 + s_buffer_load_dword s1, s[28:31], 0x2  v_interp_p1_f32 v18, v0, 0, 1, [m0]  v_interp_p2_f32 v18, [v18], v1, 0, 1, [m0]  v_interp_p1_f32 v19, v0, 1, 1, [m0]  v_interp_p2_f32 v19, [v19], v1, 1, 1, [m0]  v_interp_p1_f32 v2, v0, 0, 2, [m0]  v_interp_p2_f32 v2, [v2], v1, 0, 2, [m0]  v_interp_p1_f32 v3, v0, 1, 2, [m0]  v_interp_p2_f32 v3, [v3], v1, 1, 2, [m0]  v_interp_p1_f32 v4, v0, 2, 2, [m0]  v_interp_p2_f32 v4, [v4], v1, 2, 2, [m0]  v_interp_p1_f32 v5, v0, 3, 2, [m0]  v_interp_p2_f32 v5, [v5], v1, 3, 2, [m0]  v_interp_p1_f32 v11, v0, 0, 3, [m0]  v_interp_p2_f32 v11, [v11], v1, 0, 3, [m0]  v_subrev_f32_e32 v7, s25, v11  v_interp_p1_f32 v13, v0, 1, 3, [m0]  v_interp_p2_f32 v13, [v13], v1, 1, 3, [m0]  v_subrev_f32_e32 v8, s26, v13  v_interp_p1_f32 v10, v0, 2, 3, [m0]  v_interp_p2_f32 v10, [v10], v1, 2, 3, [m0]  v_subrev_f32_e32 v9, s27, v10 - s_load_dwordx4 s[40:43], s[4:5], 0xc - s_load_dwordx8 s[32:39], s[6:7], 0x18 + s_load_dwordx4 s[36:39], s[4:5], 0xc + s_load_dwordx8 s[28:35], s[6:7], 0x18  s_waitcnt lgkmcnt(0) - image_sample v[20:22], 7, 0, 0, 0, 0, 0, 0, 0, v[18:19], s[32:39], s[40:43] + image_sample v[20:22], 7, 0, 0, 0, 0, 0, 0, 0, v[18:19], s[28:35], s[36:39]  s_waitcnt vmcnt(0)  v_mul_f32_e32 v12, v20, v12  v_mul_f32_e32 v14, v21, v14  v_mul_f32_e32 v15, v22, v15 - v_mad_f32 v12, -s8, v12, v12 - v_mad_f32 v14, -s9, v14, v14 - v_mad_f32 v16, -s10, v15, v15 + v_mad_f32 v12, -s3, v12, v12 + v_mad_f32 v14, -s2, v14, v14 + v_mad_f32 v15, -s1, v15, v15  v_add_f32_e64 v12, 0, v12 clamp - s_buffer_load_dword s47, s[12:15], 0x115 - s_buffer_load_dword s46, s[12:15], 0x116 - s_buffer_load_dword s30, s[12:15], 0x118 - s_buffer_load_dword s28, s[12:15], 0x119 - s_buffer_load_dword s29, s[12:15], 0x11a - s_buffer_load_dword s45, s[12:15], 0x11c - s_buffer_load_dword s44, s[12:15], 0x11d - s_buffer_load_dword s31, s[12:15], 0x11e - s_buffer_load_dword s27, s[12:15], 0x198 - s_buffer_load_dword s26, s[12:15], 0x199 - s_buffer_load_dword s25, s[12:15], 0x19a - v_add_f32_e64 v15, 0, v14 clamp - v_add_f32_e64 v16, 0, v16 clamp - v_mul_f32_e32 v14, v7, v7 - v_mac_f32_e32 v14, v8, v8 - v_mac_f32_e32 v14, v9, v9 - v_sqrt_f32_e32 v17, v14 + s_buffer_load_dword s45, s[12:15], 0x115 + s_buffer_load_dword s43, s[12:15], 0x116 + s_buffer_load_dword s41, s[12:15], 0x118 + s_buffer_load_dword s42, s[12:15], 0x119 + s_buffer_load_dword s40, s[12:15], 0x11a + s_buffer_load_dword s27, s[12:15], 0x11c + s_buffer_load_dword s26, s[12:15], 0x11d + s_buffer_load_dword s25, s[12:15], 0x11e + s_buffer_load_dword s47, s[12:15], 0x198 + s_buffer_load_dword s46, s[12:15], 0x199 + s_buffer_load_dword s44, s[12:15], 0x19a + v_add_f32_e64 v14, 0, v14 clamp + v_add_f32_e64 v15, 0, v15 clamp + v_mul_f32_e32 v16, v7, v7 + v_mac_f32_e32 v16, v8, v8 + v_mac_f32_e32 v16, v9, v9 + v_sqrt_f32_e32 v17, v16  v_mov_b32_e32 v20, 0x358637bd - v_mov_b32_e32 v14, 0 + v_mov_b32_e32 v16, 0  v_cmp_gt_f32_e32 vcc, v20, v17  s_waitcnt lgkmcnt(0)  s_and_saveexec_b64 s[48:49], vcc  s_xor_b64 s[48:49], exec, s[48:49]  s_cbranch_execz BB0_2  ; BB#1: ; %IF37  s_buffer_load_dword s50, s[12:15], 0x7  s_buffer_load_dword s51, s[12:15], 0x8  s_buffer_load_dword s64, s[12:15], 0x9  s_buffer_load_dword s65, s[12:15], 0xb  s_buffer_load_dword s66, s[12:15], 0xc - v_interp_p1_f32 v14, v0, 3, 3, [m0] - v_interp_p2_f32 v14, [v14], v1, 3, 3, [m0] - image_sample v18, 8, 0, 0, 0, 0, 0, 0, 0, v[18:19], s[32:39], s[40:43] + v_interp_p1_f32 v16, v0, 3, 3, [m0] + v_interp_p2_f32 v16, [v16], v1, 3, 3, [m0] + image_sample v18, 8, 0, 0, 0, 0, 0, 0, 0, v[18:19], s[28:35], s[36:39]  v_interp_p1_f32 v0, v0, 3, 0, [m0]  v_interp_p2_f32 v0, [v0], v1, 3, 0, [m0]  s_waitcnt lgkmcnt(0)  v_mul_f32_e32 v1, s50, v13  s_waitcnt vmcnt(0)  v_mul_f32_e32 v0, v0, v18  s_buffer_load_dword s50, s[12:15], 0xd  s_buffer_load_dword s67, s[12:15], 0xf  s_buffer_load_dword s68, s[12:15], 0x9e  s_buffer_load_dword s69, s[12:15], 0x9f  s_buffer_load_dword s70, s[12:15], 0xa0  s_buffer_load_dword s71, s[12:15], 0x0  s_buffer_load_dword s72, s[12:15], 0x1  s_buffer_load_dword s73, s[12:15], 0x3  s_buffer_load_dword s74, s[12:15], 0x4  s_buffer_load_dword s75, s[12:15], 0x5  s_buffer_load_dword s76, s[12:15], 0xa1  s_buffer_load_dword s77, s[12:15], 0xa2  s_buffer_load_dword s78, s[12:15], 0xa3  s_load_dwordx4 s[52:55], s[4:5], 0x8  s_load_dwordx8 s[56:63], s[6:7], 0x10  s_waitcnt lgkmcnt(0)  v_mac_f32_e32 v1, s73, v11  v_mac_f32_e32 v1, s65, v10 - v_mac_f32_e32 v1, s67, v14 + v_mac_f32_e32 v1, s67, v16  v_mul_f32_e32 v18, s75, v13  v_mac_f32_e32 v18, s72, v11  v_mul_f32_e32 v13, s74, v13  v_mac_f32_e32 v13, s71, v11  v_mac_f32_e32 v18, s64, v10  v_rcp_f32_e32 v11, v1  v_mac_f32_e32 v13, s51, v10 - v_mac_f32_e32 v18, s50, v14 - v_mac_f32_e32 v13, s66, v14 + v_mac_f32_e32 v18, s50, v16 + v_mac_f32_e32 v13, s66, v16  v_mul_f32_e32 v10, v11, v13  s_buffer_load_dword s50, s[12:15], 0xd0  v_mov_b32_e32 v19, s78  v_mac_f32_e32 v19, s70, v10  v_mul_f32_e32 v10, v11, v18  v_mov_b32_e32 v20, s77  v_mac_f32_e32 v20, s76, v10  v_mov_b32_e32 v21, 0  image_sample_l v10, 1, 0, 0, 0, 0, 0, 0, 0, v[19:22], s[56:63], s[52:55]  v_mov_b32_e32 v11, 0x49742400  v_mul_f32_e32 v0, v11, v0  v_mov_b32_e32 v13, 0x2b8cbccc  v_mad_f32 v13, -v17, v17, v13  v_sqrt_f32_e32 v13, v13 - v_mov_b32_e32 v14, 0xc9742400 + v_mov_b32_e32 v16, 0xc9742400  v_mov_b32_e32 v18, s69  s_waitcnt vmcnt(0)  v_mad_f32 v10, s68, v10, -v18  v_rcp_f32_e32 v10, v10 - v_mul_f32_e32 v14, v17, v14 - v_mac_f32_e32 v0, v0, v14 - v_add_f32_e32 v14, v13, v1 - v_min_f32_e32 v10, v14, v10 + v_mul_f32_e32 v16, v17, v16 + v_mac_f32_e32 v0, v0, v16 + v_add_f32_e32 v16, v13, v1 + v_min_f32_e32 v10, v16, v10  v_subrev_f32_e32 v13, v13, v1  v_subrev_f32_e32 v10, v13, v10  v_mul_f32_e32 v0, v10, v0  v_exp_f32_e64 v0, -v0  v_sub_f32_e32 v0, 1.0, v0  v_add_f32_e64 v0, 0, v0 clamp  v_mov_b32_e32 v10, 0xb58637bd  v_add_f32_e32 v1, v1, v10  s_waitcnt lgkmcnt(0)  v_subrev_f32_e32 v1, s50, v1  v_mul_f32_e32 v1, v11, v1  v_add_f32_e64 v1, 0, v1 clamp - v_mul_f32_e32 v14, v1, v0 + v_mul_f32_e32 v16, v1, v0  BB0_2: ; %ENDIF36  s_or_b64 exec, exec, s[48:49]  v_mad_f32 v0, v6, v12, s18 - v_mad_f32 v1, v15, v6, s19 - v_mad_f32 v6, v16, v6, s20 + v_mad_f32 v1, v14, v6, s19 + v_mad_f32 v6, v15, v6, s20  v_subrev_f32_e32 v10, s22, v7  v_subrev_f32_e32 v7, s24, v7  v_subrev_f32_e32 v11, s21, v8 - v_subrev_f32_e32 v8, s47, v8 + v_subrev_f32_e32 v8, s45, v8  v_subrev_f32_e32 v12, s23, v9 - v_subrev_f32_e32 v9, s46, v9 + v_subrev_f32_e32 v9, s43, v9  v_add_f32_e32 v13, s16, v0 - v_add_f32_e32 v15, s11, v1 - v_add_f32_e32 v16, s17, v6 + v_add_f32_e32 v14, s11, v1 + v_add_f32_e32 v15, s17, v6  v_mov_b32_e32 v17, 0x3f62dfce - v_mad_f32 v18, v10, s30, -0.5 - s_load_dwordx4 s[12:15], s[4:5], 0x0 - s_load_dwordx4 s[16:19], s[4:5], 0x4 - s_load_dwordx8 s[32:39], s[6:7], 0x0 - s_load_dwordx8 s[48:55], s[6:7], 0x8 - v_mul_f32_e32 v19, s45, v7 - v_mul_f32_e32 v20, s44, v8 - v_mul_f32_e32 v21, s31, v9 - v_mul_f32_e32 v7, s30, v10 - v_mul_f32_e32 v8, s28, v11 - v_mul_f32_e32 v9, s29, v12 - v_mov_b32_e32 v22, 0 - v_mov_b32_e32 v10, v22 + v_mul_f32_e32 v18, s10, v17 + v_mul_f32_e32 v18, s47, v18 + v_mul_f32_e32 v18, v0, v18 + v_mul_f32_e32 v19, s9, v17 + v_mul_f32_e32 v19, s46, v19 + v_mul_f32_e32 v19, v1, v19 + v_mul_f32_e32 v20, s8, v17 + v_mul_f32_e32 v20, s44, v20 + v_mul_f32_e32 v20, v6, v20 + v_mad_f32 v21, v10, s41, -0.5 + v_sub_f32_e64 v21, 0.5, |v21| + s_load_dwordx4 s[20:23], s[4:5], 0x0 + s_load_dwordx4 s[8:11], s[4:5], 0x4 + s_load_dwordx8 s[28:35], s[6:7], 0x0 + s_load_dwordx8 s[12:19], s[6:7], 0x8 + v_mad_f32 v22, v11, s42, -0.5 + v_sub_f32_e64 v22, 0.5, |v22| + v_mad_f32 v23, v12, s40, -0.5 + v_sub_f32_e64 v23, 0.5, |v23| + v_mov_b32_e32 v24, 0x40c00000 + v_mul_f32_e32 v25, s41, v10 + v_mul_f32_e32 v26, s42, v11 + v_mul_f32_e32 v27, s40, v12 + v_mul_f32_e32 v29, s27, v7 + v_mul_f32_e32 v30, s26, v8 + v_mul_f32_e32 v31, s25, v9 + v_mul_f32_e32 v7, v24, v21 + v_mul_f32_e32 v8, v24, v22 + v_mov_b32_e32 v32, 0 + v_mul_f32_e32 v9, v24, v23 + v_mov_b32_e32 v28, v32  s_waitcnt lgkmcnt(0) - image_sample_l v[19:21], 7, 0, 0, 0, 0, 0, 0, 0, v[19:22], s[32:39], s[12:15] - image_sample_l v[7:9], 7, 0, 0, 0, 0, 0, 0, 0, v[7:10], s[48:55], s[16:19] - v_mad_f32 v10, v11, s28, -0.5 - v_mad_f32 v11, v12, s29, -0.5 - v_sub_f32_e64 v12, 0.5, |v18| - v_sub_f32_e64 v10, 0.5, |v10| - v_sub_f32_e64 v11, 0.5, |v11| - v_mov_b32_e32 v18, 0x40c00000 - v_mul_f32_e32 v12, v18, v12 - v_mul_f32_e32 v10, v18, v10 - v_mul_f32_e32 v11, v18, v11 - v_add_f32_e64 v12, 0, v12 clamp - v_add_f32_e64 v10, 0, v10 clamp - v_mul_f32_e32 v10, v10, v12 - v_add_f32_e64 v11, 0, v11 clamp - v_mul_f32_e32 v10, v11, v10 - s_waitcnt vmcnt(1) - v_mad_f32 v11, -v10, v19, v19 - v_mad_f32 v12, -v10, v20, v20 - v_mad_f32 v18, -v10, v21, v21 + image_sample_l v[10:12], 7, 0, 0, 0, 0, 0, 0, 0, v[29:32], s[28:35], s[20:23] + v_add_f32_e64 v7, 0, v7 clamp + v_add_f32_e64 v8, 0, v8 clamp + v_mul_f32_e32 v7, v8, v7 + v_add_f32_e64 v8, 0, v9 clamp + v_mul_f32_e32 v7, v8, v7 + s_waitcnt vmcnt(0) + v_mad_f32 v8, -v7, v10, v10 + v_mad_f32 v9, -v7, v11, v11 + v_mad_f32 v10, -v7, v12, v12 + image_sample_l v[21:23], 7, 0, 0, 0, 0, 0, 0, 0, v[25:28], s[12:19], s[8:11]  s_waitcnt vmcnt(0) - v_mac_f32_e32 v11, v7, v10 - v_mac_f32_e32 v12, v8, v10 - v_mac_f32_e32 v18, v9, v10 - v_mul_f32_e32 v7, s3, v17 - v_mul_f32_e32 v7, s27, v7 - v_mul_f32_e32 v7, v0, v7 - v_mul_f32_e32 v8, v17, v11 - v_mac_f32_e32 v7, v8, v0 - v_mul_f32_e32 v0, s2, v17 - v_mul_f32_e32 v0, s26, v0 - v_mul_f32_e32 v0, v1, v0 - v_mul_f32_e32 v8, v17, v12 - v_mac_f32_e32 v0, v8, v1 - v_mul_f32_e32 v1, s1, v17 - v_mul_f32_e32 v1, s25, v1 - v_mul_f32_e32 v1, v6, v1 - v_mul_f32_e32 v8, v17, v18 - v_mac_f32_e32 v1, v8, v6 - v_mad_f32 v6, -s0, v7, v7 - v_mac_f32_e32 v6, s0, v13 - v_mad_f32 v0, -s0, v0, v0 + v_mac_f32_e32 v8, v7, v21 + v_mac_f32_e32 v9, v7, v22 + v_mac_f32_e32 v10, v7, v23 + v_mul_f32_e32 v7, v17, v8 + v_mac_f32_e32 v18, v7, v0 + v_mul_f32_e32 v0, v17, v9 + v_mac_f32_e32 v19, v0, v1 + v_mul_f32_e32 v0, v17, v10 + v_mac_f32_e32 v20, v0, v6 + v_sub_f32_e64 v0, 1.0, s0 + v_mad_f32 v1, v18, v0, s3 + v_mac_f32_e32 v1, s0, v13 + v_mad_f32 v6, v19, v0, s2 + v_mac_f32_e32 v6, s0, v14 + v_mad_f32 v0, v20, v0, s1  v_mac_f32_e32 v0, s0, v15 - v_mad_f32 v1, -s0, v1, v1 - v_mac_f32_e32 v1, s0, v16 - v_add_f32_e32 v6, s8, v6 - v_add_f32_e32 v0, s9, v0 - v_add_f32_e32 v1, s10, v1 - v_mac_f32_e32 v2, v5, v6 - v_mac_f32_e32 v3, v5, v0 - v_mac_f32_e32 v4, v5, v1 - v_add_f32_e64 v0, 0, v14 clamp + v_mac_f32_e32 v2, v5, v1 + v_mac_f32_e32 v3, v5, v6 + v_mac_f32_e32 v4, v5, v0 + v_add_f32_e64 v0, 0, v16 clamp  exp 15, 0, 0, 1, 1, v2, v3, v4, v0  s_endpgm    .section .AMDGPU.csdata  ; Kernel info: -; codeLenInByte = 892 +; codeLenInByte = 888  ; NumSgprs: 81 -; NumVgprs: 23 +; NumVgprs: 33  ; FloatMode: 192  ; IeeeMode: 0  ; ScratchSize: 0    .section ".note.GNU-stack"  .text  EndOfTextLabel: