fix sdxl mlperf time bug

[huijuanzh]

What does this PR do?

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Bug found when test with num_batches = throughput_warmup_steps.

test cmd:
python text_to_image_generation.py --model_name_or_path /host/mnt/ctrl/disk1/personal/cg/models/stable-diffusion-xl-base-1.0 --prompts "Sailing ship painting by Van Gogh" --num_imag es_per_prompt 10 --batch_size 4 --image_save_dir /tmp/stable_diffusion_xl_images --scheduler euler_discrete --use_habana --use_hpu_graphs --num_inference_steps 30 --height 1024 --width 1024 --gaudi_config Habana/stable-diffusion --bf16 --optimize

Speed metrics: {'generation_runtime': 247.3958, 'generation_samples_per_second': 0.049, 'generation_steps_per_second': 1.31}

python text_to_image_generation.py --model_name_or_path /host/mnt/ctrl/disk1/personal/cg/models/stable-diffusion-xl-base-1.0 --prompts "Sailing ship painting by Van Gogh" --num_imag es_per_prompt 16 --batch_size 4 --image_save_dir /tmp/stable_diffusion_xl_images --scheduler euler_discrete --use_habana --use_hpu_graphs --num_inference_steps 30 --height 1024 --width 1024 --gaudi_config Habana/stable-diffusion --bf16 --optimize

Speed metrics: {'generation_runtime': 261.2659, 'generation_samples_per_second': 0.258, 'generation_steps_per_second': 7.751}

G2D:

<style> </style>

resolution	steps	bs	Throughput (image/s)
1024*1024	30	1	0.249
1024*1024	30	2	0.257
1024*1024	30	4	0.047

[huijuanzh]

G2D update patch test:

<style> </style>

resolution	steps	num_images_per_prompt	bs	Throughput (image/s)
1024*1024	30	4	4	0.24
1024*1024	30	8	4	0.241
1024*1024	30	10	4	0.244
1024*1024	30	12	4	0.245
1024*1024	30	16	4	0.258
1024*1024	30	20	4	0.258
1024*1024	30	30	4	0.259

[regisss]

There could still be more than 1 batch no?

[huijuanzh]

we just calculate the latest batch time in use_warmup_inference_steps case(add and j == num_batches - 1 in line 841), so the num_samples is batch_size