A lightweight WebAssembly runtime that is fast, secure, and standards-compliant
x64: Fix some required features of instructions (#13971)
Alex Crichtona5d923e5f853
diff --git a/cranelift/assembler-x64/meta/src/instructions/cvt.rs b/cranelift/assembler-x64/meta/src/instructions/cvt.rsindex 9f03f4c984..c6f630a74d 100644--- a/cranelift/assembler-x64/meta/src/instructions/cvt.rs+++ b/cranelift/assembler-x64/meta/src/instructions/cvt.rs@@ -65,6 +65,6 @@ pub fn list() -> Vec<Inst> {// * cvttpd2pi// * cvttps2pi- inst("vcvtudq2ps", fmt("A", [w(xmm1), r(xmm_m128)]), evex(L128, Full)._f2()._0f().w0().op(0x7A).r(), _64b | avx512vl | avx512f),+ inst("vcvtudq2ps", fmt("A", [w(xmm1), r(xmm_m128)]), evex(L128, Full)._f2()._0f().w0().op(0x7A).r(), (_64b | compat) & avx512vl & avx512f),]}diff --git a/cranelift/assembler-x64/meta/src/instructions/mov.rs b/cranelift/assembler-x64/meta/src/instructions/mov.rsindex 79746be00b..862be2b257 100644--- a/cranelift/assembler-x64/meta/src/instructions/mov.rs+++ b/cranelift/assembler-x64/meta/src/instructions/mov.rs@@ -77,46 +77,46 @@ pub fn list() -> Vec<Inst> {// destination must be modeled in the ModRM:r/m byte, not the ModRM:reg// byte. Currently our encoding based on format doesn't account for this// special case, so it's just dropped here.- inst("movss", fmt("A_M", [w(xmm1), r(m32)]), rex([0xF3, 0x0F, 0x10]).r(), compat | _64b | sse).alt(avx, "vmovss_d"),- inst("movss", fmt("A_R", [rw(xmm1), r(xmm2)]), rex([0xF3, 0x0F, 0x10]).r(), compat | _64b | sse).alt(avx, "vmovss_b"),- inst("movss", fmt("C_M", [w(m32), r(xmm1)]), rex([0xF3, 0x0F, 0x11]).r(), compat | _64b | sse).alt(avx, "vmovss_c_m"),- inst("movsd", fmt("A_M", [w(xmm1), r(m64)]), rex([0xF2, 0x0F, 0x10]).r(), compat | _64b | sse2).alt(avx, "vmovsd_d"),- inst("movsd", fmt("A_R", [rw(xmm1), r(xmm2)]), rex([0xF2, 0x0F, 0x10]).r(), compat | _64b | sse2).alt(avx, "vmovsd_b"),- inst("movsd", fmt("C_M", [w(m64), r(xmm1)]), rex([0xF2, 0x0F, 0x11]).r(), compat | _64b | sse2).alt(avx, "vmovsd_c_m"),- inst("vmovss", fmt("D", [w(xmm1), r(m32)]), vex(LIG)._f3()._0f().op(0x10).r(), compat | _64b | avx),- inst("vmovss", fmt("B", [w(xmm1), r(xmm2), r(xmm3)]), vex(LIG)._f3()._0f().op(0x10).r(), compat | _64b | avx),- inst("vmovss", fmt("C_M", [w(m32), r(xmm1)]), vex(LIG)._f3()._0f().op(0x11).r(), compat | _64b | avx),- inst("vmovsd", fmt("D", [w(xmm1), r(m64)]), vex(LIG)._f2()._0f().op(0x10).r(), compat | _64b | avx),- inst("vmovsd", fmt("B", [w(xmm1), r(xmm2), r(xmm3)]), vex(LIG)._f2()._0f().op(0x10).r(), compat | _64b | avx),- inst("vmovsd", fmt("C_M", [w(m64), r(xmm1)]), vex(LIG)._f2()._0f().op(0x11).r(), compat | _64b | avx),+ inst("movss", fmt("A_M", [w(xmm1), r(m32)]), rex([0xF3, 0x0F, 0x10]).r(), (_64b | compat) & sse).alt(avx, "vmovss_d"),+ inst("movss", fmt("A_R", [rw(xmm1), r(xmm2)]), rex([0xF3, 0x0F, 0x10]).r(), (_64b | compat) & sse).alt(avx, "vmovss_b"),+ inst("movss", fmt("C_M", [w(m32), r(xmm1)]), rex([0xF3, 0x0F, 0x11]).r(), (_64b | compat) & sse).alt(avx, "vmovss_c_m"),+ inst("movsd", fmt("A_M", [w(xmm1), r(m64)]), rex([0xF2, 0x0F, 0x10]).r(), (_64b | compat) & sse2).alt(avx, "vmovsd_d"),+ inst("movsd", fmt("A_R", [rw(xmm1), r(xmm2)]), rex([0xF2, 0x0F, 0x10]).r(), (_64b | compat) & sse2).alt(avx, "vmovsd_b"),+ inst("movsd", fmt("C_M", [w(m64), r(xmm1)]), rex([0xF2, 0x0F, 0x11]).r(), (_64b | compat) & sse2).alt(avx, "vmovsd_c_m"),+ inst("vmovss", fmt("D", [w(xmm1), r(m32)]), vex(LIG)._f3()._0f().op(0x10).r(), (_64b | compat) & avx),+ inst("vmovss", fmt("B", [w(xmm1), r(xmm2), r(xmm3)]), vex(LIG)._f3()._0f().op(0x10).r(), (_64b | compat) & avx),+ inst("vmovss", fmt("C_M", [w(m32), r(xmm1)]), vex(LIG)._f3()._0f().op(0x11).r(), (_64b | compat) & avx),+ inst("vmovsd", fmt("D", [w(xmm1), r(m64)]), vex(LIG)._f2()._0f().op(0x10).r(), (_64b | compat) & avx),+ inst("vmovsd", fmt("B", [w(xmm1), r(xmm2), r(xmm3)]), vex(LIG)._f2()._0f().op(0x10).r(), (_64b | compat) & avx),+ inst("vmovsd", fmt("C_M", [w(m64), r(xmm1)]), vex(LIG)._f2()._0f().op(0x11).r(), (_64b | compat) & avx),// Move aligned 128-bit values to and from XMM locations.- inst("movapd", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x28]).r(), compat | _64b | sse2).alt(avx, "vmovapd_a"),- inst("movapd", fmt("B", [w(align(xmm_m128)), r(xmm1)]), rex([0x66, 0x0F, 0x29]).r(), compat | _64b | sse2).alt(avx, "vmovapd_b"),- inst("movaps", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x0F, 0x28]).r(), compat | _64b | sse).alt(avx, "vmovaps_a"),- inst("movaps", fmt("B", [w(align(xmm_m128)), r(xmm1)]), rex([0x0F, 0x29]).r(), compat | _64b | sse).alt(avx, "vmovaps_b"),- inst("movdqa", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x6F]).r(), compat | _64b | sse2).alt(avx, "vmovdqa_a"),- inst("movdqa", fmt("B", [w(align(xmm_m128)), r(xmm1)]), rex([0x66, 0x0F, 0x7F]).r(), compat | _64b | sse2).alt(avx, "vmovdqa_b"),- inst("vmovapd", fmt("A", [w(xmm1), r(align(xmm_m128))]), vex(L128)._66()._0f().op(0x28).r(), compat | _64b | avx),- inst("vmovapd", fmt("B", [w(align(xmm_m128)), r(xmm1)]), vex(L128)._66()._0f().op(0x29).r(), compat | _64b | avx),- inst("vmovaps", fmt("A", [w(xmm1), r(align(xmm_m128))]), vex(L128)._0f().op(0x28).r(), compat | _64b | avx),- inst("vmovaps", fmt("B", [w(align(xmm_m128)), r(xmm1)]), vex(L128)._0f().op(0x29).r(), compat | _64b | avx),- inst("vmovdqa", fmt("A", [w(xmm1), r(align(xmm_m128))]), vex(L128)._66()._0f().op(0x6F).r(), compat | _64b | avx),- inst("vmovdqa", fmt("B", [w(align(xmm_m128)), r(xmm1)]), vex(L128)._66()._0f().op(0x7F).r(), compat | _64b | avx),+ inst("movapd", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x28]).r(), (_64b | compat) & sse2).alt(avx, "vmovapd_a"),+ inst("movapd", fmt("B", [w(align(xmm_m128)), r(xmm1)]), rex([0x66, 0x0F, 0x29]).r(), (_64b | compat) & sse2).alt(avx, "vmovapd_b"),+ inst("movaps", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x0F, 0x28]).r(), (_64b | compat) & sse).alt(avx, "vmovaps_a"),+ inst("movaps", fmt("B", [w(align(xmm_m128)), r(xmm1)]), rex([0x0F, 0x29]).r(), (_64b | compat) & sse).alt(avx, "vmovaps_b"),+ inst("movdqa", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x6F]).r(), (_64b | compat) & sse2).alt(avx, "vmovdqa_a"),+ inst("movdqa", fmt("B", [w(align(xmm_m128)), r(xmm1)]), rex([0x66, 0x0F, 0x7F]).r(), (_64b | compat) & sse2).alt(avx, "vmovdqa_b"),+ inst("vmovapd", fmt("A", [w(xmm1), r(align(xmm_m128))]), vex(L128)._66()._0f().op(0x28).r(), (_64b | compat) & avx),+ inst("vmovapd", fmt("B", [w(align(xmm_m128)), r(xmm1)]), vex(L128)._66()._0f().op(0x29).r(), (_64b | compat) & avx),+ inst("vmovaps", fmt("A", [w(xmm1), r(align(xmm_m128))]), vex(L128)._0f().op(0x28).r(), (_64b | compat) & avx),+ inst("vmovaps", fmt("B", [w(align(xmm_m128)), r(xmm1)]), vex(L128)._0f().op(0x29).r(), (_64b | compat) & avx),+ inst("vmovdqa", fmt("A", [w(xmm1), r(align(xmm_m128))]), vex(L128)._66()._0f().op(0x6F).r(), (_64b | compat) & avx),+ inst("vmovdqa", fmt("B", [w(align(xmm_m128)), r(xmm1)]), vex(L128)._66()._0f().op(0x7F).r(), (_64b | compat) & avx),// Move unaligned 128-bit values to and from XMM locations.- inst("movupd", fmt("A", [w(xmm1), r(xmm_m128)]), rex([0x66, 0x0F, 0x10]).r(), compat | _64b | sse2).alt(avx, "vmovupd_a"),- inst("movupd", fmt("B", [w(xmm_m128), r(xmm1)]), rex([0x66, 0x0F, 0x11]).r(), compat | _64b | sse2).alt(avx, "vmovupd_b"),- inst("movups", fmt("A", [w(xmm1), r(xmm_m128)]), rex([0x0F, 0x10]).r(), compat | _64b | sse).alt(avx, "vmovups_a"),- inst("movups", fmt("B", [w(xmm_m128), r(xmm1)]), rex([0x0F, 0x11]).r(), compat | _64b | sse).alt(avx, "vmovups_b"),- inst("movdqu", fmt("A", [w(xmm1), r(xmm_m128)]), rex([0xF3, 0x0F, 0x6F]).r(), compat | _64b | sse2).alt(avx, "vmovdqu_a"),- inst("movdqu", fmt("B", [w(xmm_m128), r(xmm1)]), rex([0xF3, 0x0F, 0x7F]).r(), compat | _64b | sse2).alt(avx, "vmovdqu_b"),- inst("vmovupd", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._66()._0f().op(0x10).r(), compat | _64b | avx),- inst("vmovupd", fmt("B", [w(xmm_m128), r(xmm1)]), vex(L128)._66()._0f().op(0x11).r(), compat | _64b | avx),- inst("vmovups", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._0f().op(0x10).r(), compat | _64b | avx),- inst("vmovups", fmt("B", [w(xmm_m128), r(xmm1)]), vex(L128)._0f().op(0x11).r(), compat | _64b | avx),- inst("vmovdqu", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._f3()._0f().op(0x6F).r(), compat | _64b | avx),- inst("vmovdqu", fmt("B", [w(xmm_m128), r(xmm1)]), vex(L128)._f3()._0f().op(0x7F).r(), compat | _64b | avx),+ inst("movupd", fmt("A", [w(xmm1), r(xmm_m128)]), rex([0x66, 0x0F, 0x10]).r(), (_64b | compat) & sse2).alt(avx, "vmovupd_a"),+ inst("movupd", fmt("B", [w(xmm_m128), r(xmm1)]), rex([0x66, 0x0F, 0x11]).r(), (_64b | compat) & sse2).alt(avx, "vmovupd_b"),+ inst("movups", fmt("A", [w(xmm1), r(xmm_m128)]), rex([0x0F, 0x10]).r(), (_64b | compat) & sse).alt(avx, "vmovups_a"),+ inst("movups", fmt("B", [w(xmm_m128), r(xmm1)]), rex([0x0F, 0x11]).r(), (_64b | compat) & sse).alt(avx, "vmovups_b"),+ inst("movdqu", fmt("A", [w(xmm1), r(xmm_m128)]), rex([0xF3, 0x0F, 0x6F]).r(), (_64b | compat) & sse2).alt(avx, "vmovdqu_a"),+ inst("movdqu", fmt("B", [w(xmm_m128), r(xmm1)]), rex([0xF3, 0x0F, 0x7F]).r(), (_64b | compat) & sse2).alt(avx, "vmovdqu_b"),+ inst("vmovupd", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._66()._0f().op(0x10).r(), (_64b | compat) & avx),+ inst("vmovupd", fmt("B", [w(xmm_m128), r(xmm1)]), vex(L128)._66()._0f().op(0x11).r(), (_64b | compat) & avx),+ inst("vmovups", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._0f().op(0x10).r(), (_64b | compat) & avx),+ inst("vmovups", fmt("B", [w(xmm_m128), r(xmm1)]), vex(L128)._0f().op(0x11).r(), (_64b | compat) & avx),+ inst("vmovdqu", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._f3()._0f().op(0x6F).r(), (_64b | compat) & avx),+ inst("vmovdqu", fmt("B", [w(xmm_m128), r(xmm1)]), vex(L128)._f3()._0f().op(0x7F).r(), (_64b | compat) & avx),// Move and extend packed integers to and from XMM locations with sign extension.inst("pmovsxbw", fmt("A", [w(xmm1), r(xmm_m64)]), rex([0x66, 0x0F, 0x38, 0x20]).r(), (_64b | compat) & sse41).alt(avx, "vpmovsxbw_a"),