(mahout) branch main updated: Hoist loop-invariant norm computation in iqp_encode_batch_kernel_naive (#1198)

[hc...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

hcr pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/mahout.git


The following commit(s) were added to refs/heads/main by this push:
     new 54fcd13dc Hoist loop-invariant norm computation in iqp_encode_batch_kernel_naive (#1198)
54fcd13dc is described below

commit 54fcd13dcb8c3cffd69fe7df1ee64c41286b5e5b
Author: wdskuki <ds...@gmail.com>
AuthorDate: Tue Mar 17 11:52:44 2026 +0800

    Hoist loop-invariant norm computation in iqp_encode_batch_kernel_naive (#1198)
---
 qdp/qdp-kernels/src/iqp.cu | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/qdp/qdp-kernels/src/iqp.cu b/qdp/qdp-kernels/src/iqp.cu
index 2316c1252..a8a14c505 100644
--- a/qdp/qdp-kernels/src/iqp.cu
+++ b/qdp/qdp-kernels/src/iqp.cu
@@ -251,6 +251,8 @@ __global__ void iqp_encode_batch_kernel_naive(
     const size_t total_elements = num_samples * state_len;
     const size_t stride = gridDim.x * blockDim.x;
     const size_t state_mask = state_len - 1;
+    // Normalize by 1/2^n (state_len = 2^n) - hoisted outside the loop
+    const double norm = 1.0 / (double)state_len;
 
     for (size_t global_idx = blockIdx.x * blockDim.x + threadIdx.x;
          global_idx < total_elements;
@@ -261,7 +263,6 @@ __global__ void iqp_encode_batch_kernel_naive(
 
         cuDoubleComplex amp = compute_amplitude_naive(data, z, state_len, num_qubits, enable_zz);
 
-        double norm = 1.0 / (double)state_len;
         state_batch[global_idx] = make_cuDoubleComplex(cuCreal(amp) * norm, cuCimag(amp) * norm);
     }
 }