Bug 23076 – SIMD assert fail with -inline -O converting float to short

Created attachment 1851 main source DMD32 D Compiler v2.100.0-beta.1-dirty Please consider the following program: ------------------ main.d -------------------------- float mul3(float a, float b, float t) { return t * b * a; } class A { ushort depthLow = 0; ushort depthHigh = 30000; void fun(short* pixels) { short depthA = (cast(short)(mul3(depthHigh, depthLow, 0))); short depthB = (cast(short)(mul3(depthLow, depthHigh, 0))); pixels[depthA] = depthB; } } ----------------------------------------------------- You may now crash the compiler with the following build command: $ dmd -m64 -inline -O -w main.d This is a regression, no ICE with dmd.2.099.1

Reduced it to: struct S { int depthLow = 0; int depthHigh = 30000; void fun(int* pixels) { float b = depthLow; int depthA = cast(int)(b); int depthB = cast(short)(cast(float)depthHigh * cast(float)depthLow); pixels[depthA] = depthB; } } compile with -O

@WalterBright created dlang/dmd pull request #14174 "fix Issue 23076 - SIMD assert fail with -inline -O converting float t…" fixing this issue: - fix Issue 23076 - SIMD assert fail with -inline -O converting float to short https://github.com/dlang/dmd/pull/14174

dlang/dmd pull request #14174 "fix Issue 23076 - SIMD assert fail with -inline -O converting float t…" was merged into master: - 99e10f87d12cac36db27d2b86d6d2ef116f3f59c by Walter Bright: fix Issue 23076 - SIMD assert fail with -inline -O converting float to short https://github.com/dlang/dmd/pull/14174