nir: add amul instruction

Used for address/offset calculation (ie. array derefs), where we can
potentially use less than 32b for the multiply of array idx by element
size.  For backends that support `imul24`, this gives a lowering pass
an easy way to find multiplies that potentially can be converted to
`imul24`.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

src/compiler/nir/nir_opcodes.py

@@ -1047,6 +1047,18 @@ dst.z = src2.x;
 dst.w = src3.x;
 """)
 
+# An integer multiply instruction for address calculation.  This is
+# similar to imul, except that the results are undefined in case of
+# overflow.  Overflow is defined according to the size of the variable
+# being dereferenced.
+#
+# This relaxed definition, compared to imul, allows an optimization
+# pass to propagate bounds (ie, from an load/store intrinsic) to the
+# sources, such that lower precision integer multiplies can be used.
+# This is useful on hw that has 24b or perhaps 16b integer multiply
+# instructions.
+binop("amul", tint, _2src_commutative + associative, "src0 * src1")
+
 # ir3-specific instruction that maps directly to mul-add shift high mix,
 # (IMADSH_MIX16 i.e. ah * bl << 16 + c). It is used for lowering integer
 # multiplication (imul) on Freedreno backend..