nak: Use the hardware's max warps_per_sm value

This should improve our occupancy estimates.

Reviewed-by: Mary Guillemard <mary@mary.zone>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38913>

src/nouveau/compiler/nak/api.rs

@@ -433,7 +433,7 @@ fn nak_compile_shader_internal(
         Some(unsafe { &*fs_key })
     };
 
-    let sm = ShaderModelInfo::new(nak.sm);
+    let sm = ShaderModelInfo::new(nak.sm, nak.warps_per_sm);
     let mut s = nak_shader_from_nir(nak, nir, &sm);
 
     if DEBUG.print() {

src/nouveau/compiler/nak/hw_tests.rs

@@ -37,7 +37,8 @@ impl RunSingleton {
 
             let run = Runner::new(dev_id);
             let sm_nr = run.dev_info().sm;
-            let sm = ShaderModelInfo::new(sm_nr);
+            let sm =
+                ShaderModelInfo::new(sm_nr, run.dev_info().max_warps_per_mp);
             RunSingleton { sm, run }
         })
     }

src/nouveau/compiler/nak/ir.rs

@@ -9305,11 +9305,12 @@ pub trait ShaderModel {
 
 pub struct ShaderModelInfo {
     sm: u8,
+    warps_per_sm: u8,
 }
 
 impl ShaderModelInfo {
-    pub fn new(sm: u8) -> Self {
+    pub fn new(sm: u8, warps_per_sm: u8) -> Self {
-        ShaderModelInfo { sm }
+        ShaderModelInfo { sm, warps_per_sm }
     }
 }
 
@@ -9430,7 +9431,7 @@ pub fn gpr_limit_from_local_size(local_size: &[u16; 3]) -> u32 {
     min(out, 255)
 }
 
-pub fn max_warps_per_sm(gprs: u32) -> u32 {
+pub fn max_warps_per_sm(sm: &ShaderModelInfo, gprs: u32) -> u32 {
     fn prev_multiple_of(x: u32, y: u32) -> u32 {
         (x / y) * y
     }
@@ -9440,7 +9441,7 @@ pub fn max_warps_per_sm(gprs: u32) -> u32 {
     // GPRs are allocated in multiples of 8
     let gprs = gprs.next_multiple_of(8);
     let max_warps = prev_multiple_of((total_regs / 32) / gprs, 4);
-    min(max_warps, 48)
+    min(max_warps, sm.warps_per_sm.into())
 }
 
 pub struct Shader<'a> {
@@ -9600,6 +9601,7 @@ impl Shader<'_> {
         self.info.uses_fp64 = uses_fp64;
 
         self.info.max_warps_per_sm = max_warps_per_sm(
+            self.sm,
             self.info.num_gprs as u32 + self.sm.hw_reserved_gprs(),
         );
 

src/nouveau/compiler/nak/nvdisasm_tests.rs

@@ -87,7 +87,7 @@ fn disassemble_instrs(instrs: Vec<Instr>, sm: u8) -> Vec<String> {
         io: ShaderIoInfo::None,
     };
 
-    let sm = ShaderModelInfo::new(sm);
+    let sm = ShaderModelInfo::new(sm, 0);
     let s = Shader {
         sm: &sm,
         info: info,