31     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
36       xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
39         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
42 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
49     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
54       xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
57         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
60 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
67     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
69     xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
71     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
74 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
79     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
81     xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
83     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
86 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
91     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
96       xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
99         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
102 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
109     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
114       xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
117         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
120 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
127     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
129     xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
131     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
134 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
139     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
141     xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
143     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
146 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
151     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
156       xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
159         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
162 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
169     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
174       xnn_math_f16_f32_cvt__sse2_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
177         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
180 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
189     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
194       xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
197         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
200 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
207     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
212       xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
215         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
218 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
225     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
227     xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
229     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
232 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
237     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
239     xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
241     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
244 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
249     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
254       xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
257         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
260 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
267     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
272       xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
275         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
278 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
285     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
287     xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
289     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
292 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
297     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
299     xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
301     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
304 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
309     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
314       xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
317         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
320 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
327     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
332       xnn_math_f16_f32_cvt__sse2_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
335         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
338 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
349     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
354       xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
357         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
360 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
369     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
374       xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
377         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
380 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
389     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
391     xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
393     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
396 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
403     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
405     xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
407     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
410 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
417     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
422       xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
425         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
428 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
437     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
442       xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
445         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
448 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
457     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
459     xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
461     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
464 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
471     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
473     xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
475     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
478 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
485     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
490       xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
493         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
496 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
505     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
510       xnn_math_f16_f32_cvt__sse41_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
513         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
516 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
527     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
532       xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
535         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
538 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
547     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
552       xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
555         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
558 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
567     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
569     xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
571     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
574 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
581     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
583     xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
585     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
588 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
595     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
600       xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
603         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
606 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
615     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
620       xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
623         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
626 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
635     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
637     xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
639     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
642 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
649     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
651     xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
653     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
656 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
663     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
668       xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
671         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
674 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
683     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
688       xnn_math_f16_f32_cvt__sse41_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
691         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
694 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
705     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
710       xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
713         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
716 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
725     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
730       xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
733         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
736 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
745     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
747     xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
749     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
752 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
759     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
761     xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
763     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
766 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
773     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
778       xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
781         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
784 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
793     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
798       xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
801         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
804 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
813     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
815     xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
817     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
820 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
827     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
829     xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
831     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
834 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
841     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
846       xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
849         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
852 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
861     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
866       xnn_math_f16_f32_cvt__f16c(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
869         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
872 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
883     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
888       xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
891         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
894 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
903     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
908       xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
911         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
914 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
923     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
925     xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
927     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
930 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
937     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
939     xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
941     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
944 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
951     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
956       xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
959         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
962 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
971     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
976       xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
979         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
982 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
991     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
993     xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
995     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
998 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1005     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1007     xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1009     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1012 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1019     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1024       xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1027         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1030 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1039     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1044       xnn_math_f16_f32_cvt__neon_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1047         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1050 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1061     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1066       xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1069         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1072 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1081     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1086       xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1089         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1092 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1101     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1103     xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1105     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1108 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1115     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1117     xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1119     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1122 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1129     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1134       xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1137         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1140 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1149     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1154       xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1157         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1160 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1169     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1171     xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1173     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1176 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1183     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1185     xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1187     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1190 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1197     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1202       xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1205         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1208 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1217     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1222       xnn_math_f16_f32_cvt__neon_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1225         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1228 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1239     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1244       xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1247         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1250 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1259     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1264       xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1267         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1270 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1279     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1281     xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1283     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1286 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1293     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1295     xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1297     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1300 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1307     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1312       xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1315         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1318 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1327     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1332       xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1335         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1338 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1347     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1349     xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1351     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1354 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1361     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1363     xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1365     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1368 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1375     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1380       xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1383         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1386 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1395     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1400       xnn_math_f16_f32_cvt__neonfp16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1403         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1406 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1415     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1420 …   xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1423         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1426 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1433     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1438 …   xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1441         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1444 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1451     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1453     xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1455     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1458 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1463     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1465     xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1467     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1470 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1475     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1480 …   xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1483         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1486 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1493     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1498 …   xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1501         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1504 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1511     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1513     xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1515     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1518 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1523     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1525     xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1527     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1530 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1535     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1540 …   xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1543         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1546 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1553     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1558 …   xnn_math_f16_f32_cvt__wasmsimd_int16(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1561         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1564 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1573     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1578 …   xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1581         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1584 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1591     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1596 …   xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1599         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1602 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1609     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1611     xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1613     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1616 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1621     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1623     xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1625     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1628 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1633     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1638 …   xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1641         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1644 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1651     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1656 …   xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1659         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1662 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1669     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1671     xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1673     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1676 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1681     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1683     xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1685     ASSERT_EQ(reference_output, float_as_uint32(outputs[0]))  in TEST()
1688 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[0]);  in TEST()
1693     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1698 …   xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1701         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1704 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()
1711     std::vector<float, AlignedAllocator<float, 64>> outputs(kBlockSize);  in TEST()  local
1716 …   xnn_math_f16_f32_cvt__wasmsimd_int32(kBlockSize * sizeof(float), inputs.data(), outputs.data());  in TEST()
1719         ASSERT_EQ(reference_output, float_as_uint32(outputs[i]))  in TEST()
1722 …", optimized = 0x" << std::hex << std::setw(8) << std::setfill('0') << float_as_uint32(outputs[i]);  in TEST()