resolve SLEEF before QBLAS so disable_fma actually applies

Meson caches the first subproject() call by name. When the QBLAS
dependency was resolved first, qblas's internal subproject('sleef')
ran with no options, configuring SLEEF with FMA enabled. Our later
subproject('sleef', default_options: ['disable_fma=true']) was then
silently a no-op - meson returned the already-configured instance.

Net effect on the old-CPU CI: the SLEEF in the wheel still had the
PURECFMA scalar code path enabled, and Intel SDE trapped on the first
vfnmadd132sd from Sleef_log10q1_u10purecfma when emulating Sandy
Bridge.

Move the whole SLEEF resolution block above the qblas dependency line.
SLEEF gets configured with the right options on its first (and only)
load; QBLAS's later subproject('sleef') call returns the same
instance, so there's now exactly one SLEEF in the build with the
options we wanted.

Author: SwayamInSync

meson.build

@@ -15,25 +15,11 @@ if is_windows
   add_project_arguments('-DWIN32', '-D_WINDOWS', language : ['c', 'cpp'])
 endif
 
-# QBLAS does not build under MSVC (GCC-only flags, POSIX-only APIs, GCC
-# built-ins for CPUID); force-disable it on Windows. Users on other
-# platforms can opt out via -Ddisable_quadblas=true to fall back to the
-# naive matmul kernel.
-disable_quadblas = is_windows or get_option('disable_quadblas')
-if disable_quadblas
-  if is_windows
-    message('QBLAS disabled (Windows / MSVC) - using naive matmul kernel')
-  else
-    message('QBLAS disabled by user option - using naive matmul kernel')
-  endif
-  add_project_arguments('-DDISABLE_QUADBLAS', language: ['c', 'cpp'])
-  qblas_dep = declare_dependency()
-else
-  qblas_dep = dependency('qblas', fallback: ['qblas', 'qblas_dep'])
-endif
-
-# Try to find SLEEF system-wide first, fall back to subproject if not found
-# Required SLEEF version (must match sleef.wrap revision)
+# Resolve SLEEF BEFORE the QBLAS subproject. QBLAS calls subproject('sleef')
+# without options when it's loaded as a subproject, and meson caches the
+# first subproject() call - so if QBLAS were loaded first, our disable_fma
+# option would be silently dropped (the SLEEF instance with FMA enabled
+# would be the one shipped, breaking Sandy Bridge SDE tests).
 required_sleef_version = '3.9.0'
 # Don't use fallback here - we need to call subproject() explicitly later with disable_fma option
 sleef_dep = dependency('sleef', version: '>=' + required_sleef_version, required: false)
@@ -45,7 +31,7 @@ if sleef_dep.found() and sleef_dep.version().startswith(required_sleef_version)
   # SLEEF found system-wide - verify quad-precision support
   cpp = meson.get_compiler('cpp')
   sleefquad_lib = cpp.find_library('sleefquad', required: false)
-  
+
   if sleefquad_lib.found()
     sleefquad_test_code = '''
     #include <sleefquad.h>
@@ -63,7 +49,7 @@ if sleef_dep.found() and sleef_dep.version().startswith(required_sleef_version)
       dependencies: [sleef_dep, sleefquad_lib],
       name: 'SLEEF quad-precision support'
     )
-    
+
     if quad_works
       sleefquad_dep = declare_dependency(
         dependencies: [sleef_dep, sleefquad_lib]
@@ -95,6 +81,23 @@ else
   message('Proceeding with vendored SLEEF subproject instead')
 endif
 
+# QBLAS does not build under MSVC (GCC-only flags, POSIX-only APIs, GCC
+# built-ins for CPUID); force-disable it on Windows. Users on other
+# platforms can opt out via -Ddisable_quadblas=true to fall back to the
+# naive matmul kernel.
+disable_quadblas = is_windows or get_option('disable_quadblas')
+if disable_quadblas
+  if is_windows
+    message('QBLAS disabled (Windows / MSVC) - using naive matmul kernel')
+  else
+    message('QBLAS disabled by user option - using naive matmul kernel')
+  endif
+  add_project_arguments('-DDISABLE_QUADBLAS', language: ['c', 'cpp'])
+  qblas_dep = declare_dependency()
+else
+  qblas_dep = dependency('qblas', fallback: ['qblas', 'qblas_dep'])
+endif
+
 incdir_numpy = run_command(py,
   ['-c', 'import numpy; print(numpy.get_include())'],
   check : true