if-statement/ternary dataflow bug

[sarutob1]

I've encountered this bug with if-statements/contitional-expressions that are moving data larger than 32 bits.
e.g.:

        if (wordQ.size() > 0) begin
            myint = 100'(wordQ.pop_front());
        end else begin
            myint = 100'(99);
        end

The emitted C++ code can come out as this:

    VL_EXTENDS_WI(100,32, __Vtemp_h8963a0fe__1, vlSelf->t__DOT__wordQ.pop_front());
    t__DOT__myint[0U] = (VL_LTS_III(32, 0U, vlSelf->t__DOT__wordQ.size())
                          ? __Vtemp_h8963a0fe__1[0U]
                          : 0x63U);
    t__DOT__myint[1U] = (VL_LTS_III(32, 0U, vlSelf->t__DOT__wordQ.size())
                          ? __Vtemp_h8963a0fe__1[1U]
                          : 0U);
    t__DOT__myint[2U] = (VL_LTS_III(32, 0U, vlSelf->t__DOT__wordQ.size())
                          ? __Vtemp_h8963a0fe__1[2U]
                          : 0U);
    t__DOT__myint[3U] = (VL_LTS_III(32, 0U, vlSelf->t__DOT__wordQ.size())
                          ? (0xfU & __Vtemp_h8963a0fe__1[3U])
                          : 0U);

OR

VL_COND_WIWW(100, __Vtemp_11, VL_LTS_III(32, 0U, vlSelf->wordQ.size()), vlSelf->wordQ.pop_front(), __Vtemp_9);

Both of which use the value of pop_front without accounting for side effects.

Source code/test is included in the pull request, but I'm not familiar with with the C++ emission part of verilator. Which files/modules should I look at to try and fix this?

[wsnyder]

There's two bugs here. First, the "if", is because V3Premit inserts a temporary above the if, instead of on the if's branch. It should be straight forward to fix that, can you consider taking a look and making a pull to fix it (then likely comment out the conditional ?: in your test so we can merge the fix for the 'if' only).

Separately the conditional likewise requires a temporary, but unlike the "if" that can't be just put before the statement. Instead we need to see if there's side effects, and if so, put the new temporary into a AstExprStmt that returns the value of the temporary. We want to do that only when side effects as the AstExprStmt will disable other later optimizations.