v3d: Fix copy-propagation of input unpacks.

I had a single function for "does this do float input unpacking" with two
major flaws: It was missing the most common thing to try to copy propagate
a f32 input nunpack to (the VFPACK to an FP16 render target) along with
several other ALU ops, and also would try to propagate an f32 unpack into
a VFMUL which only does f16 unpacks.

instructions in affected programs: 659232 -> 655895 (-0.51%)
uniforms in affected programs: 132613 -> 135336 (2.05%)

and a couple of programs increase their thread counts.

The uniforms hit appears to be a pattern in generated code of doing (-a >=
a) comparisons, which when a is abs(b) can result in the abs instruction
being copy propagated once but not fully DCEed.

2 files changed, 69 insertions, 0 deletions

diff --git a/src/broadcom/qpu/qpu_instr.c b/src/broadcom/qpu/qpu_instr.c
index add2d2a23c8..12a9c32c831 100644
--- a/src/broadcom/qpu/qpu_instr.c
+++ b/src/broadcom/qpu/qpu_instr.c
@@ -867,3 +867,70 @@ v3d_qpu_writes_flags(const struct v3d_qpu_instr *inst)
 
         return false;
 }
+
+bool
+v3d_qpu_unpacks_f32(const struct v3d_qpu_instr *inst)
+{
+        if (inst->type != V3D_QPU_INSTR_TYPE_ALU)
+                return false;
+
+        switch (inst->alu.add.op) {
+        case V3D_QPU_A_FADD:
+        case V3D_QPU_A_FADDNF:
+        case V3D_QPU_A_FSUB:
+        case V3D_QPU_A_FMIN:
+        case V3D_QPU_A_FMAX:
+        case V3D_QPU_A_FCMP:
+        case V3D_QPU_A_FROUND:
+        case V3D_QPU_A_FTRUNC:
+        case V3D_QPU_A_FFLOOR:
+        case V3D_QPU_A_FCEIL:
+        case V3D_QPU_A_FDX:
+        case V3D_QPU_A_FDY:
+        case V3D_QPU_A_FTOIN:
+        case V3D_QPU_A_FTOIZ:
+        case V3D_QPU_A_FTOUZ:
+        case V3D_QPU_A_FTOC:
+        case V3D_QPU_A_VFPACK:
+                return true;
+                break;
+        default:
+                break;
+        }
+
+        switch (inst->alu.mul.op) {
+        case V3D_QPU_M_FMOV:
+        case V3D_QPU_M_FMUL:
+                return true;
+                break;
+        default:
+                break;
+        }
+
+        return false;
+}
+bool
+v3d_qpu_unpacks_f16(const struct v3d_qpu_instr *inst)
+{
+        if (inst->type != V3D_QPU_INSTR_TYPE_ALU)
+                return false;
+
+        switch (inst->alu.add.op) {
+        case V3D_QPU_A_VFMIN:
+        case V3D_QPU_A_VFMAX:
+                return true;
+                break;
+        default:
+                break;
+        }
+
+        switch (inst->alu.mul.op) {
+        case V3D_QPU_M_VFMUL:
+                return true;
+                break;
+        default:
+                break;
+        }
+
+        return false;
+}
diff --git a/src/broadcom/qpu/qpu_instr.h b/src/broadcom/qpu/qpu_instr.h
index 1e2dcb78af6..a77430ff882 100644
--- a/src/broadcom/qpu/qpu_instr.h
+++ b/src/broadcom/qpu/qpu_instr.h
@@ -464,5 +464,7 @@ bool v3d_qpu_reads_flags(const struct v3d_qpu_instr *inst) ATTRIBUTE_CONST;
 bool v3d_qpu_writes_flags(const struct v3d_qpu_instr *inst) ATTRIBUTE_CONST;
 bool v3d_qpu_sig_writes_address(const struct v3d_device_info *devinfo,
                                 const struct v3d_qpu_sig *sig) ATTRIBUTE_CONST;
+bool v3d_qpu_unpacks_f32(const struct v3d_qpu_instr *inst) ATTRIBUTE_CONST;
+bool v3d_qpu_unpacks_f16(const struct v3d_qpu_instr *inst) ATTRIBUTE_CONST;
 
 #endif