This was never constructed and only provided one static helper
function.

remove unnecessary COPY for SDAG for mad-mix pattern

New intrinsics / instructions :
int_amdgcn_ds_bvh_stack_push4_pop1_rtn / ds_bvh_stack_push4_pop1_rtn_b32
int_amdgcn_ds_bvh_stack_push8_pop1_rtn / ds_bvh_stack_push8_pop1_rtn_b32
int_amdgcn_ds_bvh_stack_push8_pop2_rtn / ds_bvh_stack_push8_pop2_rtn_b64

Co-authored-by: Mateja Marjanovic <mateja.marjanovic@amd.com>

This is an extension of #131357. Hopefully this would be the last one.

For VALU shuffles, this saves an instruction in some case.

These instructions return a 64-bit result and a 1-bit carry, unlike
smul_lohi and umul_lohi which return a pair of 32-bit results.

This does not appear to make any difference in practice because the DAG
types are not used for anything before these nodes are converted to
MachineInstrs.

APInt will fail when given a negative offset. SelectScratchSVAddr
utilizes this function and can be given a negative offset as well, so
this change modifies it to use APSInt instead.

ImmOffset is signed and we will hit an assert with negative ImmOffset
when getTargetConstant is used.

Fixes: SWDEV-506453