Archives Discussions

[Raistmer]

			for (int j = 1; j < ndivs ; j++)
			{
				//perdiv *=2;
				tmp0 = di & 1;
				di >>= 1;
				tmp0 += di;
				dis_thresh = t_funct(di, num_adds, j,PulseMax,t_funct_cache_fp) * avg;
				{
					tmp_max = sumtop2(tmp_pot, tmp_pot, di, fft_len4, tmp0);
				}

#if 1

	if(ul_PoT == 1 && y == 3 && z == 7 && p == 4454){
		printf("tmp_max=(%f,%f,%f,%f) dis_thresh=(%f,%f,%f,%f) j=%d\n",
			tmp_max.x,tmp_max.y,tmp_max.z,tmp_max.w,
			dis_thresh.x,dis_thresh.y,dis_thresh.z,dis_thresh.w,
			j);
	}

#endif

				if (tmp_max.x>dis_thresh.x || tmp_max.y>dis_thresh.y ||
					tmp_max.z>dis_thresh.z || tmp_max.w>dis_thresh.w) {
					// unscale for reporting
					tmp_max /= (float4)num_adds_2;
					cur_thresh = (dis_thresh / (float4)num_adds_2 - avg) * rcfg_dis_thresh + avg;

float4 _snr = (tmp_max-avg)/(cur_thresh-avg);

#if 1

	if(ul_PoT == 1 && y == 3 && z == 7 && p == 4454){
		printf("_snr=(%f,%f,%f,%f) best_pulse_score=(%f)\n",
			_snr.x,_snr.y,_snr.z,_snr.w,best_pulse_score);
	}

#endif

					if (_snr.x  > best_pulse_score ||
						_snr.y  > best_pulse_score ||
						_snr.z  > best_pulse_score ||
						_snr.w  > best_pulse_score) {
						result_flag[0].x = 1;printf("location 3 hit: ul_PoT=%d,y=%d,z=%d,p=%d\n",ul_PoT,y,z,p);return;
					} When it was %d instead %f for float values it worked OK. tmp_max=(1016206848,1016042150,1014254674,1015103447) dis_thresh=(1017574742,1017521425,1017343526,1017306023) j=1 tmp_max=(1019540776,1019234484,1019323582,1019736860) dis_thresh=(1022113170,1022037738,1021786049,1021732991) j=2 tmp_max=(1025885083,1025687250,1025025298,1024181522) dis_thresh=(1026938408,1026880340,1026686587,1026645743) j=3 tmp_max=(1032427341,1033697660,1031554466,1031172886) dis_thresh=(1033265612,1033217589,1033057352,1033023573) j=4 _snr=(1052921234,1058128081,1050160947,1049360638) best_pulse_score=(1063430413) tmp_max=(1039569588,1041180477,1038290570,1038685748) dis_thresh=(1040415552,1040373564,1040233466,1040203932) j=5 _snr=(1052941357,1060001794,1049353865,1051071391) best_pulse_score=(1063430413) tmp_max=(1046772721,1049205382,1045513371,1045367322) dis_thresh=(1047492886,1047416412,1047161246,1047107455) j=6 _snr=(1052854581,1063899244,1046138112,1044991040) best_pulse_score=(1063430413) location 3 hit: ul_PoT=1,y=3,z=7,p=4454 Now I changed %d to %f where needed but output broken to signle line with ridiculous j value. What is wrong ????

[Raistmer]

btw, forum formatting just CRAZY. Why it used table for copy/paste from plain text editor???

[Raistmer]

Just trying to make code more easy to read:

                              //          int j = 1;

                              //float4 tmp_max2=(float4)0.f;

                              for (int j = 1; j < ndivs ; j++)

                              {

                                        //perdiv *=2;

                                        tmp0 = di & 1;

                                        di >>= 1;

                                        tmp0 += di;

                                        dis_thresh = t_funct(di, num_adds, j,PulseMax,t_funct_cache_fp) * avg;

                                        {

                                                  tmp_max = sumtop2(tmp_pot, tmp_pot, di, fft_len4, tmp0);

                                        }

#if 1

          if(ul_PoT == 1 && y == 3 && z == 7 && p == 4454){

                    printf("tmp_max=(%d,%d,%d,%d) dis_thresh=(%d,%d,%d,%d) j=%d\n",

                              tmp_max.x,tmp_max.y,tmp_max.z,tmp_max.w,dis_thresh.x,dis_thresh.y,dis_thresh.z,dis_thresh.w,j);

          }

#endif

                                        if (tmp_max.x>dis_thresh.x || tmp_max.y>dis_thresh.y ||

                                                  tmp_max.z>dis_thresh.z || tmp_max.w>dis_thresh.w) {

                                                  // unscale for reporting

                                                  tmp_max /= (float4)num_adds_2;

                                                  cur_thresh = (dis_thresh / (float4)num_adds_2 - avg) * rcfg_dis_thresh + avg;

                                                  float4 _snr = (tmp_max-avg)/(cur_thresh-avg);

#if 1

          if(ul_PoT == 1 && y == 3 && z == 7 && p == 4454){

                    printf("_snr=(%d,%d,%d,%d) best_pulse_score=(%d)\n",

                              _snr.x,_snr.y,_snr.z,_snr.w,best_pulse_score);

          }

#endif

                                                  if (_snr.x  > best_pulse_score ||

                                                            _snr.y  > best_pulse_score ||

                                                            _snr.z  > best_pulse_score ||

                                                            _snr.w  > best_pulse_score) {

                                                            result_flag[0].x = 1;printf("location 3 hit: ul_PoT=%d,y=%d,z=%d,p=%d\n",ul_PoT,y,z,p);return;

                                                  }

when %d changed (only for floating point values of course) to %.4g or %f printf fails to print anything useful.

[himanshu_gautam]

Hi Raistmer,

Thanks for reporting the issue. Will it be possible for you attach a small testcase, which I can directly run and check if the issue is still present in our internal drivers.

One small suggestion, use %vn format strings when printing vectors. They are quite handy.

[Raistmer]

Well, you could try attached pack.

But when I tried to reproduce issue on HD6950-based PC I got compilation failure:

INFO: can't open binary kernel file: .\\MultiBeam_Kernels_r1812.clHD5_Cayman.bin_V7, continue with recompile...

Error : Building Program (source, clBuildProgram):main kernels: not OK code -11

C:\Users\R\AppData\Local\Temp\OCL7BF4.tmp.cl(2200): error: function "printf"

          declared implicitly

            if(ul_PoT == 0 && y == 0 && z == 0) printf("Entering PulseFind\n");

                                                ^

C:\Users\R\AppData\Local\Temp\OCL7BF4.tmp.cl(2262): error: function "printf"

          declared implicitly

                                          printf("label 1\n");

Catalyst 12.1

And it was compiled w/o issues on my dev netbook with some another Cat version.

[Another version is:

Name:						Loveland
Vendor:					Advanced Micro Devices, Inc.
Driver version:				CAL 1.4.1646 (VM)
Version:					OpenCL 1.2 AMD-APP (938.2)
Extensions:					cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_sharing cl_ext_atomic_counters_32 cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_popcnt cl_khr_d3d10_sharing

]

This return us to previously reported issue: should printf extension be manually enabled or not. Looks like for this particular host with this partcular Catalyst version it should be...

EDIT: same error on Catalyst 12.8 on Trinity APU.

Name:						Devastator
Vendor:					Advanced Micro Devices, Inc.
Driver version:				CAL 1.4.1741 (VM)
Version:					OpenCL 1.1 AMD-APP (851.4)
Extensions:					cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_sharing cl_ext_atomic_counters_32 cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_popcnt cl_khr_d3d10_sharing

[pinform]

Hi,

Sorry for the delayed response.

Does the problem persist?  Can you try with the latest driver and report back?

--Prasad