Quantizelinear nearbyint fix

[CharlieL7]

Fixes #2661.

For floating point types do not use the nearbyint operation.
Updates tests to catch this error and fixes gold values.

I tested and it fixes the issue seen on https://github.com/ROCm/rocMLIR-internal/issues/1718

[migraphx-bot]

Test	Batch	Rate new 5a3918	Rate old bfc14e	Diff	Compare
torchvision-resnet50	64	3,233.96	3,233.96	0.00%	✅
torchvision-resnet50_fp16	64	6,874.33	6,874.33	0.00%	✅
torchvision-densenet121	32	2,438.02	2,436.94	0.04%	✅
torchvision-densenet121_fp16	32	4,175.33	4,181.18	-0.14%	✅
torchvision-inceptionv3	32	1,613.72	1,612.90	0.05%	✅
torchvision-inceptionv3_fp16	32	2,675.59	2,677.82	-0.08%	✅
cadene-inceptionv4	16	750.12	750.47	-0.05%	✅
cadene-resnext64x4	16	808.90	809.72	-0.10%	✅
slim-mobilenet	64	6,660.48	6,659.15	0.02%	✅
slim-nasnetalarge	64	196.81	196.77	0.02%	✅
slim-resnet50v2	64	3,429.83	3,430.42	-0.02%	✅
bert-mrpc-onnx	8	1,133.14	1,133.14	0.00%	✅
bert-mrpc-tf	1	502.51	500.00	0.50%	✅
pytorch-examples-wlang-gru	1	477.31	470.97	1.35%	✅
pytorch-examples-wlang-lstm	1	440.19	437.32	0.66%	✅
torchvision-resnet50_1	1	800.00	793.65	0.80%	✅
cadene-dpn92_1	1	427.94	430.36	-0.56%	✅
cadene-resnext101_1	1	389.11	389.11	0.00%	✅
onnx-taau-downsample	1	370.94	371.31	-0.10%	✅
dlrm-criteoterabyte	1	31.83	31.79	0.11%	✅
dlrm-criteoterabyte_fp16	1	51.04	51.06	-0.03%	✅
agentmodel	1	8,412.69	8,658.38	-2.84%	✅
unet_fp16	2	58.34	58.36	-0.03%	✅
resnet50v1_fp16	1	1,063.83	1,075.27	-1.06%	✅
resnet50v1_int8	1	806.45	800.00	0.81%	✅
bert_base_cased_fp16	64	1,158.58	1,159.42	-0.07%	✅
bert_large_uncased_fp16	32	359.83	359.96	-0.03%	✅
bert_large_fp16	1	215.98	215.98	0.00%	✅
distilgpt2_fp16	16	2,185.79	2,188.78	-0.14%	✅
yolov5s	1	521.16	531.08	-1.87%	✅
tinyllama	1	49.07	49.12	-0.10%	✅
vicuna-fastchat	1	43.59	43.55	0.10%	✅
whisper-tiny-encoder	1	436.68	436.68	0.00%	✅
whisper-tiny-decoder	1	450.45	448.43	0.45%	✅

This build is OK for merge ✅

[migraphx-bot]

     ✅ bert-mrpc-onnx: PASSED: MIGraphX meets tolerance

     ✅ bert-mrpc-tf: PASSED: MIGraphX meets tolerance

     ✅ pytorch-examples-wlang-gru: PASSED: MIGraphX meets tolerance

     ✅ pytorch-examples-wlang-lstm: PASSED: MIGraphX meets tolerance

     ✅ torchvision-resnet50_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-dpn92_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-resnext101_1: PASSED: MIGraphX meets tolerance

     ✅ dlrm-criteoterabyte: PASSED: MIGraphX meets tolerance

     ✅ agentmodel: PASSED: MIGraphX meets tolerance

     ✅ unet: PASSED: MIGraphX meets tolerance

     ✅ resnet50v1: PASSED: MIGraphX meets tolerance

     ✅ bert_base_cased_fp16: PASSED: MIGraphX meets tolerance

🔴bert_large_uncased_fp16: FAILED: MIGraphX is not within tolerance - check verbose output

     ✅ bert_large: PASSED: MIGraphX meets tolerance

     ✅ yolov5s: PASSED: MIGraphX meets tolerance

     ✅ tinyllama: PASSED: MIGraphX meets tolerance

     ✅ vicuna-fastchat: PASSED: MIGraphX meets tolerance

     ✅ whisper-tiny-encoder: PASSED: MIGraphX meets tolerance

     ✅ whisper-tiny-decoder: PASSED: MIGraphX meets tolerance

     ✅ distilgpt2_fp16: PASSED: MIGraphX meets tolerance

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.35%. Comparing base (bfc14e5) to head (5a39189).
Report is 2 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #3819   +/-   ##
========================================
  Coverage    92.35%   92.35%           
========================================
  Files          519      519           
  Lines        22307    22311    +4     
========================================
+ Hits         20601    20606    +5     
+ Misses        1706     1705    -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ahsan-ca]

Besides a typo, looks good to me. Thanks for doing this.

[ahsan-ca]

Typo

[lakhinderwalia]

Perhaps you should saturate quantized + zero_pts. Else that sum will overflow in the next line.

[CharlieL7]

I'll make a followup PR with this