Can't run terravox on linux on AMD processor

[noway]

Didn't tested Intel processor, so can't say if this is general issue for linux or not.

On my processor, terravox gives following error:

[1] 18295 illegal hardware instruction (core dumped) ./Terravox

Here is cat /proc/cpuinfo

Also, if i'm trying to compile it with gcc, i'm getting following error:

g++ -c -pipe -msse4.1 -mssse3 -msse3 -msse2 -msse -Wno-missing-braces -O2 -march=x86-64 -mtune=generic -O2 -pipe -fstack-protector-strong --param=ssp-buffer-size=4 -std=c++0x -Wall -W -D_REENTRANT -fPIE -DQT_NO_DEBUG -DQT_WIDGETS_LIB -DQT_GUI_LIB -DQT_CONCURRENT_LIB -DQT_CORE_LIB -I/usr/lib/qt/mkspecs/linux-g++ -I. -isystem /usr/include/qt -isystem /usr/include/qt/QtWidgets -isystem /usr/include/qt/QtGui -isystem /usr/include/qt/QtConcurrent -isystem /usr/include/qt/QtCore -I. -I. -o terrainview_p.o terrainview_p.cpp
terrainview_p.cpp: In lambda function:
terrainview_p.cpp:485:23: error: cannot convert ‘__m128i {aka __vector(2) long long int}’ to ‘__vector(4) float’ in assignment
                 coeff = _mm_cvttps_epi32(coeff);                // i32 * 4
                       ^
terrainview_p.cpp:486:53: error: cannot convert ‘__vector(4) float’ to ‘__m128i {aka __vector(2) long long int}’ for argument ‘1’ to ‘__m128i _mm_packs_epi32(__m128i, __m128i)’
                 coeff = _mm_packs_epi32(coeff, coeff);          // i16 * 4 + pad
                                                     ^
terrainview_p.cpp:487:56: error: cannot convert ‘__vector(4) float’ to ‘__m128i {aka __vector(2) long long int}’ for argument ‘1’ to ‘__m128i _mm_unpacklo_epi16(__m128i, __m128i)’
                 coeff = _mm_unpacklo_epi16(coeff, coeff);       // i16 [c1, c1, c2, c2, c3, c3, c4, c4]
                                                        ^
terrainview_p.cpp:492:94: error: cannot convert ‘__vector(4) float’ to ‘__m128i {aka __vector(2) long long int}’ for argument ‘1’ to ‘__m128i _mm_unpacklo_epi16(__m128i, __m128i)’
                 colLo = _mm_srai_epi16(_mm_mullo_epi16(colLo, _mm_unpacklo_epi16(coeff, coeff)), 6);
                                                                                              ^
terrainview_p.cpp:494:94: error: cannot convert ‘__vector(4) float’ to ‘__m128i {aka __vector(2) long long int}’ for argument ‘1’ to ‘__m128i _mm_unpackhi_epi16(__m128i, __m128i)’
                 colHi = _mm_srai_epi16(_mm_mullo_epi16(colHi, _mm_unpackhi_epi16(coeff, coeff)), 6);
                                                                                              ^
terrainview_p.cpp: At global scope:
terrainview_p.cpp:601:49: warning: unused parameter ‘alt’ [-Wunused-parameter]
         renderLandform1([](quint32 color, float alt) { return color; });
                                                 ^
Makefile:824: recipe for target 'terrainview_p.o' failed
make: *** [terrainview_p.o] Error 1

Which is probably somehow related to that.

Is there any hope to get it working?

[yvt]

Currently, Terravox requires SSE 4.1 because of _mm_floor_ps, _mm_round_ps, and _mm_extract_epi32.

[noway]

Oups, wrong link to gist file with cpuinfo. Fixed it in post and also here it is.
Unfortunately there is only sse4a, and looks like it's not sufficient for SSE 4.1.

[noway]

As workaround I found Intel SDE. This thing allow to emulate SSE instructions on processor without them.

Performance is not that good, but at least I was able to run it.

I just used following command:

path-to-kit/sde64 -- ./Terravox

Closing.